Skip to main content
Sergei Grebnov
Member of Technical Staff at Spice AI
View all authors

Spice v2.4.2 (Nov 9, 2026)

ยท 15 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Spice v2.4.2 is now available! ๐Ÿ› ๏ธ

Spice v2.4.2 is a patch release that continues security hardening and fixes Snowflake key-pair authentication. Security fixes limit the nesting depth, the expanded size, and the memory use of SQL queries, and HTTP datasets refuse redirects to other internal hosts. The release also fixes several SQL, Spice Cayenne, and runtime issues.

What's New in v2.4.2โ€‹

SQL Query Limits Protect the Runtimeโ€‹

Spice now limits the nesting depth, the expanded size, and the memory use of SQL queries. A query over a limit fails with HTTP 400 on /v1/sql, or with InvalidArgument on Flight SQL. The runtime continues to serve other queries. Before this release, any client that could send SQL could stop the runtime with one query. A deeply nested query overflowed the stack, and the process stopped with fatal runtime error: stack overflow, aborting. Other queries made one function or one plan allocate memory without a limit.

The checks apply to HTTP /v1/sql, /v1/nsql, Flight SQL statements and prepared statements, Flight SQL updates, and transactions. The checks cover these shapes:

  • Long expressions. A chain of +, ||, AND, or OR terms deeper than 512 levels fails. This includes a /v1/search where condition.
  • Nested queries. Subqueries, or common table expressions (CTEs) that each read the CTE before them, fail past 64 levels.
  • CTE expansion. A query fails when its CTE references expand to more than 4096 table references. The planner makes a copy of the plan of a CTE for each reference to it.
  • Nested data types. A data type such as ARRAY<ARRAY<โ€ฆ>> fails past 32 levels. A type such as INT[][]โ€ฆ fails past 64 levels.
  • Arrow type strings. An Arrow type string, such as an arrow_cast argument or a column type that a dataset declares, fails past 64 levels.
  • Grouping sets. A GROUP BY that expands to more than 4096 grouping sets fails. CUBE of 13 expressions makes 8192 sets. 4096 is also the limit in PostgreSQL.
  • Search keywords. A /v1/search request with more than 512 keywords fails with HTTP 400.

These scalar functions now keep their output within runtime.query.memory_limit, or within a fixed cap:

  • lpad and rpad charge the whole output buffer to the memory limit. This includes an empty fill string and negative lengths. A query that does not fit fails with Resources exhausted.
  • array_repeat and list_repeat charge the real size of a list, struct, binary, boolean, or NULL element. Before this release, they charged 1 byte for each copy of such an element.
  • flatten_json_properties, json_tree, and flatten_json stop at 32 ร— max_bytes of output for one document. max_depth is limited to 128. A small JSON schema with repeated allOf or $ref entries no longer expands exponentially.
  • truncate cuts text by characters, not bytes. Non-ASCII text made the query fail with Query driver task ended unexpectedly.

Flight SQL metadata requests also have limits. GetSqlInfo info IDs and GetTables table types are limited to 1024 entries. GetTables and GetDbSchemas filter patterns are limited to 1024 bytes.

The limits are:

ShapeLimit
Expression depth, such as a chain of AND, OR, +, or ||512 levels
Query nesting: subqueries, or CTEs that read each other64 levels
Table references after CTE expansion4096
SQL data type nesting with ARRAY, STRUCT, MAP, and similar keywords32 levels
SQL data type nesting with []64 levels
Arrow type string nesting64 levels
GROUP BY grouping sets4096 (CUBE of up to 12 expressions)
/v1/search keywords512
Flight SQL GetSqlInfo info IDs and GetTables table types1024 entries
Flight SQL GetTables and GetDbSchemas filter patterns1024 bytes

CTEs that do not read each other are not limited by depth. Table references outside of CTEs do not count toward the expansion limit.

Credentials Stay out of API Responses and Logsโ€‹

GET /v1/datasets and GET /v1/catalogs no longer return credentials from the from value of a component. Before this release, any API key that could list components could read a password in a source URL, such as https://user:password@host/file.csv. This included a read-only key. The runtime now removes user:password@ from each URL. It replaces values of query keys such as password, secret, token, and sig with ***. With status=true, /v1/datasets also removes credentials from error_message. Sources with no credentials do not change.

The runtime also removes credentials from these errors and log lines:

  • ADBC connection errors. Setup errors no longer contain adbc_uri. The runtime replaces each configured secret in the driver message with ***. Before this release, the full URI went to the warning log and to /v1/datasets?status=true.
  • MCP transport errors. The error no longer contains the URL of the MCP server. A URL can contain a credential.
  • .env files. The warning for a malformed line gives only the line number, not the line content.

HTTP Datasets Refuse Redirects to Other Internal Hostsโ€‹

An http or https dataset in a file format, such as CSV or Parquet, no longer follows a redirect to an internal host. An internal host is a different host that is a loopback, private, or link-local address, or that resolves to one. The dataset fails with this error:

Refused to connect to <target>: the HTTPS data connector does not follow a redirect to a private, loopback, or link-local address

The host of the dataset itself is always allowed, also when it is internal. Redirects that stay on that host still work. SPICE_URL_TABLE_ALLOW_PRIVATE_HOSTS=true turns off the check.

Snowflake Key-Pair Authentication Fixโ€‹

Snowflake datasets with key-pair authentication now load correctly. Before this release, such a dataset stopped at Initializing with this error:

Could not automatically determine the process-level CryptoProvider from jsonwebtoken crate features.

Other Bug Fixesโ€‹

  • IS [NOT] DISTINCT FROM with AND or OR. These queries no longer return wrong boolean results or fail with an integer type error. For example, SELECT true IS DISTINCT FROM true OR true now returns true, not false. The comparison now binds before AND or OR.
  • COUNT(DISTINCT โ€ฆ) on list or struct columns. A query with two or more COUNT(DISTINCT โ€ฆ) aggregates no longer fails with Resources exhausted when one of them reads a list or struct column. Before this release, the query failed at about 50,000 rows and requested tens of GB. Each stored value held a reference to its whole input batch. This applied to federated and Cayenne queries, with or without GROUP BY.
  • Spice Cayenne file mode. At startup, a file-mode acceleration checks that the data files in its metastore exist and have the expected size. If they do not match, the acceleration reloads from the source or fails to load, and it does not serve incomplete data. Accelerations on S3 and in memory mode are not checked.
  • Queries after a burst of stale cache hits. With stale_while_revalidate_ttl set on the SQL results cache, new client queries no longer wait behind background revalidations. In a test, 600 stale hits made the next uncached query wait 321.8 seconds, and 6.3 seconds with this fix. At most one quarter of the query admission limit runs revalidations at the same time. A stale hit over that limit gets the cached result with no revalidation, and results_cache_stale_swr_count counts it. The runtime cancels a revalidation after 5 minutes and records the outcome timed_out.
  • truncate on numbers. truncate no longer returns wrong values for integers and decimals near the limits of their type. For example, truncate(100, arrow_cast(50, 'Int8')) returned 56, and the correct result is 0. A result that does not fit its type is now an error.
  • SQL scalar functions with no arguments. A from: sql scalar function with no arguments no longer fails on every call.
  • Certificate errors in the CLI. spice HTTP commands, such as status and refresh, report a TLS certificate failure as a TLS error. In v2.4.1, the release build reported it as The Spice runtime is unavailable.
  • AWS session tokens. The runtime no longer sends an empty AWS session token together with static keys. A secret that is not set gives an empty token. This applies to S3, Delta Lake, and AWS SDK connectors.

Dependency Updatesโ€‹

  • arrow-rs (Spice fork): 59.3.0 โ†’ 59.3.2. This update fixes RUSTSEC-2026-0442 in run-end-encoded array validation, and it adds the 64-level limit for Arrow type strings.
  • azure_security_keyvault_secrets: 0.10.0 โ†’ 1.0.1. This update fixes RUSTSEC-2026-0438 in the Azure Key Vault secret store. azure_identity and azure_core move to the 1.x line for that store.
  • sqlparser: 0.62.0, vendored with the precedence fix from apache/datafusion-sqlparser-rs#2436.
  • DataFusion (Spice fork): the pin adds the fixes for COUNT(DISTINCT โ€ฆ) memory and for nested SQL data types.

Breaking Changesโ€‹

Query limits. Queries and requests over the new limits fail with HTTP 400 or Flight InvalidArgument. Most of them stopped the runtime before this release, but some ran in v2.4.1. Examples are a GROUP BY CUBE of 13 expressions, a /v1/search request with more than 512 keywords, and a query whose CTEs expand to more than 4096 table references. For each limit, see SQL Query Limits Protect the Runtime.

HTTP dataset redirects. An http or https dataset in a file format fails when its server redirects to a different internal host. To allow it, set SPICE_URL_TABLE_ALLOW_PRIVATE_HOSTS=true.

JSON flattening. flatten_json, json_tree, and flatten_json_properties stop at 32 ร— max_bytes of output for one document (256 MiB by default). They return the rows up to that point, with no error. The limit is on output, so a small document with long, deeply nested keys can reach it. To get all rows, increase max_bytes.

Upgradingโ€‹

To upgrade to v2.4.2, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:2.4.2 image:

docker pull spiceai/spiceai:2.4.2

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 2.4.2

AWS Marketplace:

Spice is available in the AWS Marketplace.

Cookbook Updatesโ€‹

No new cookbook recipes.

The Spice Cookbook provides recipes to help you get started with Spice.

Contributorsโ€‹

What's Changedโ€‹

Changelogโ€‹

  • [release/2.4] fix(sql): honour IS [NOT] DISTINCT FROM precedence before AND/OR (backport #15935) by @lukekim in #15936
  • fix(snowflake): select the jsonwebtoken crypto provider at startup by @sgrebnov in #15988
  • fix: reject over-deep SQL expressions and search keyword lists by @phillipleblanc in #15348
  • Reject over-deep expressions before they overflow the query stack by @phillipleblanc in #15433
  • Reject over-deep query nesting before planning by @phillipleblanc in #15722
  • fix(sql): reject queries whose CTE references expand to too many table references by @phillipleblanc in #15766
  • fix: bound nested SQL data types before parsing by @phillipleblanc in #15764
  • fix(sql): limit the nesting depth of SQL data types by @phillipleblanc in #15705
  • fix(query): reject GROUP BY that expands past 4096 grouping sets before planning by @phillipleblanc in #15843
  • fix: limit Arrow type-string nesting depth on every parse path by @phillipleblanc in #15845
  • fix(datafusion): charge lpad/rpad output buffers to the query memory limit by @phillipleblanc in #15846
  • fix(runtime): reserve array_repeat output by element size by @phillipleblanc in #15885
  • fix(udfs): bound JSON function expansion and output; fix truncate overflow and non-ASCII text by @phillipleblanc in #15912
  • fix(flight): bound Flight SQL metadata request list and pattern sizes by @phillipleblanc in #15848
  • fix(http): remove URL credentials from from in /v1/datasets and /v1/catalogs by @lukekim in #15834
  • fix(adbc): keep the connection URI out of ADBC setup errors by @lukekim in #15835
  • fix(http): refuse dataset redirects to other internal hosts by @phillipleblanc in #15877
  • [2.4] fix(cayenne): don't serve a file-mode acceleration whose data files no longer match its metastore by @lukekim in #15943
  • [release/2.4]fix(datafusion): charge each nested COUNT(DISTINCT) value for its own bytes (2.4) by @lukekim in #15986
  • fix(cache): bound stale-while-revalidate background revalidations and give them a deadline by @phillipleblanc in #15820
  • fix(cache): stop a timed-out stale-while-revalidate query that still waits for admission by @phillipleblanc in #15880
  • fix(cli): use rustls for the HTTP client so certificate errors are recognized (2.4 backport of #15811) by @lukekim in #15984
  • fix(deps): bump spiceai/arrow-rs to upstream 59.3.2 (2.4) by @lukekim in #16041
  • fix(deps): bump azure_security_keyvault_secrets to 1.0.1 on release/2.4 (RUSTSEC-2026-0438) by @lukekim in #16043
  • [release/2.4] fix(deps): replace temporary datafusion branch with spiceai-55-release-2.4 by @sgrebnov in #16046
  • Update spicepod.schema.json by @github-actions in #15828
  • ci: add rustfs-credentials action for GitHub OIDC credentials to RustFS by @phillipleblanc in #15976
  • ci: move CI credentials to OIDC and environment secrets by @phillipleblanc in #16010
  • ci: get RustFS credentials from the GitHub OIDC token by @phillipleblanc in #16005
  • test: remove CI tests that need a live Dremio server by @phillipleblanc in #15973
  • ci: authenticate BigQuery tests with workload identity federation by @sgrebnov in #16009
  • ci: authenticate CI calls to Anthropic with workload identity federation by @lukekim in #15954
  • ci: pin MongoDB 8.0 integration tests to 8.0.4 by @krinart in #15958
  • ci: run the released-runtime spice chat E2E test against xAI by @Jeadie in #15974
  • ci(cuda): least-privilege tokens and drop the MinIO copy by @sgrebnov in #16020
  • ci: scope workflow jobs to GitHub environments by @Jeadie in #15966
  • ci: pass secrets to run scripts through env, and use the read-write Postgres for load runs by @Jeadie in #15975
  • [release/2.4] ci: move CI credentials to OIDC and environment secrets (backport) by @sgrebnov in #16035
  • ci: fail a tag build when the tag does not match version.txt by @sgrebnov in #16021
  • ci(docker): build release images on GitHub-hosted runners by @sgrebnov in #16022
  • ci(release): cancel an older build of the same tag by @peasee in #16032
  • [release/2.4] ci: hosted Docker release builds and release hardening by @sgrebnov in #16044
  • ci: Use OIDC-based AWS roles for short-lived AWS credentials by @Jeadie in #16039
  • [release/2.4] Revert workload identity for models and GitHub App keys in Key Vault/KMS by @Jeadie in #16052
  • v2.4.2: remaining security backports, version bump to 2.4.2 by @sgrebnov in #16055

Full Changelog: https://github.com/spiceai/spiceai/compare/v2.4.1...v2.4.2

Spice v2.4.0-rc.3 (Oct 22, 2026)

ยท 20 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Spice v2.4.0-rc.3 is now available! ๐Ÿ› ๏ธ

Spice v2.4.0-rc.3 is the third release candidate for v2.4.0, with mostly fixes since rc.2. By default, full and append refreshes now write on their own threads, not on the threads that serve queries. In a Cayenne test, P99 query latency at each refresh commit fell from 130โ€“690 ms to the steady P99 of 44โ€“50 ms. On Unix, the runtime also raises its soft open file limit at startup, up to the hard limit. Before, a few hundred client connections in a container could make Cayenne files fail to open.

Twelve other changes apply to Iceberg append refresh and table reads, Spice Cayenne, Bedrock decisions, OTLP logs, distributed queries, and TLS verification. Some apply only in specific setups. Two fix regressions from earlier v2.4.0 release candidates: Failed to open Vortex file errors after a Cayenne load, and failed MySQL CDC snapshots into Cayenne.

Testers of rc.2 should retest Iceberg append refresh, Cayenne refresh_mode: caching, and ai_* functions on Bedrock models. If you are new to v2.4.0, start with the rc.1 and rc.2 notes for its main features and breaking changes.

Highlights in this release candidate include:

  • Lower query latency during refreshes: By default, full and append refreshes write to the accelerator on dedicated refresh threads. In a Cayenne test, P99 latency at each refresh commit fell from 130โ€“690 ms to the steady P99 of 44โ€“50 ms.
  • Higher open file limit on Unix: The runtime raises its soft open file limit at startup, up to the hard limit. In a container with a soft limit of 1024 and a hard limit of 524288, the runtime raised its soft limit to 524288.

What's New in v2.4.0-rc.3โ€‹

Lower Query Latency During Refreshesโ€‹

Full and append refreshes now write to the accelerator on the dedicated refresh threads. Before this release, the write ran on the threads that also serve HTTP requests. For Cayenne, the write sorts, encodes, checks for duplicate keys, and publishes the data. A large write held those threads, and query latency rose until the write ended.

The change applies to these refresh writes, for all accelerators:

  • refresh_mode: full, and the initial load
  • refresh_mode: append
  • refresh_mode: changes, only when it rebuilds the table from a snapshot

Steady CDC apply, the caching and snapshot refresh modes, and retention deletes do not change. With runtime.params.dedicated_thread_pool: disabled, there are no refresh threads, and refresh writes run as before.

In a test, Cayenne ran a full refresh every 5 minutes on a table of 2.2 million rows and about 8 GB. A client sent 500 point-lookup joins per second over HTTP. Before the change, P99 latency in the 15-second window of each refresh commit was 130โ€“690 ms. After the change, it matched the steady P99 of 44โ€“50 ms. The test did not cover append refreshes or other accelerators. See #15262.

Higher Open File Limit on Unixโ€‹

On Unix systems such as Linux and macOS, the runtime now raises its soft open file limit (RLIMIT_NOFILE) to the hard limit at startup. Client connections, acceleration files, and pipes all count against this one limit. Containers often start the runtime with a soft limit of 1024 and a hard limit of 524288. Before this release, a few hundred client connections could make Cayenne files fail to open. Queries then failed with Unable to open Cayenne acceleration file, or the client saw Connection reset by peer.

  • Measured: On macOS with a soft limit of 256, v2.3.1 failed queries with 176 to 220 idle client connections open. A build with the fix raised the limit to 245760, the kern.maxfilesperproc value of that Mac. It answered queries with 600 connections open.
  • Containers: In a Docker container with --ulimit nofile=1024:524288, the runtime raised its soft limit from 1024 to 524288.
  • macOS: The runtime raises the limit to at most kern.maxfilesperproc.
  • Logs: The runtime logs the result at DEBUG, for example Raised the open file limit from 1024 to 524288. The runtime also logs a failed raise only at DEBUG. Start the runtime with -v or set SPICED_LOG to see it.
  • Windows: Nothing changes.

If the hard limit is also low, the runtime cannot raise the soft limit above it. For this case, the Cayenne error now names the effective limit and says that client connections count against it. It also keeps the Too many open files (os error 24) text from the operating system:

Failed to open Cayenne acceleration file 'โ€ฆ/01a15b2f5c697c01b2a8acdede2d637d_p000_00011.vortex' because the process reached its open file limit of 256, which client connections and all other open files also count against. Raise the limit where spiced is started (`ulimit -n`, or the container's `nofile` limit), or write fewer, larger acceleration files with `cayenne_target_file_size_mb`. See: https://spiceai.org/docs/components/data-accelerators/cayenne#params. Cause: โ€ฆ Too many open files (os error 24)

See #15339.

Iceberg Append Refresh Follows Table Snapshotsโ€‹

An Iceberg dataset with refresh_mode: append and a time_column now loads the rows that each new commit adds, whatever their time values. This applies to Cayenne datasets and to datasets without a primary key. It covers Iceberg tables read through the Iceberg or Glue data connectors. Glue tables in Parquet, CSV, or ORC format still read by time.

Before this release, an append refresh read only the rows with a time_column value later than the stored maximum minus refresh_append_overlap. A writer that commits out of time order can put rows outside that window. A sink with several shards, one of them behind, is an example. Those rows did not load until a full refresh.

The refresh now saves the Iceberg snapshot that it loaded last. The next refresh reads the data files that later snapshots added. An unchanged table scans no data. This change adds no setting:

datasets:
- from: iceberg:https://iceberg-catalog.example.com/v1/namespaces/sales/tables/orders
name: orders
time_column: updated_at
acceleration:
enabled: true
engine: cayenne
mode: file
primary_key: order_id
refresh_mode: append
  • Newest version by time: With a primary key, Cayenne keeps the newest version of each key by time_column. A late row with an older time does not replace a newer stored row.
  • Skipped commits: The refresh skips compactions and commits that add no data file. An append refresh never removes a row.
  • Fallback to reads by time: If the runtime cannot follow the snapshots, it reads by time for one refresh. This happens after an overwrite on a dataset without a primary key, or after an expired snapshot in the interval. A rollback, a branch switch, or a recreated table has the same effect.
  • After a fallback: A refresh that reads by time does not load rows older than refresh_append_overlap. A full refresh loads them. The runtime logs a warning when the reason changes. The new dataset_acceleration_refresh_fallbacks metric counts each fallback by reason.
  • Saved position: File-mode accelerations store the last loaded snapshot with the acceleration data. Other accelerations keep it in memory. A full refresh saves the current snapshot, so the next append refresh continues from it.
  • Computed columns: A dataset that computes columns over the Iceberg table, such as embedding columns, still reads by time.
  • Overlap: refresh_append_overlap now applies only to the first load and to refreshes that read by time. Datasets that follow snapshots no longer log the startup warning that asks for refresh_append_overlap.

With a primary key, a dataset on DuckDB, SQLite, or another accelerator that keeps the last row to arrive still reads by time. This matches rc.2. A primary key that includes the time_column is the exception: such a dataset follows snapshots. On those accelerators, a late row with an older time would replace the newer stored row. If such a dataset has no refresh_append_overlap, the runtime logs this warning when it loads the dataset:

Dataset 'acc' (`engine: duckdb`) has no `refresh_append_overlap`, so a row that arrives after a newer one is never loaded. Use `engine: cayenne` to load every new row, or set `refresh_append_overlap` to how late rows can arrive. See: https://spiceai.org/docs/features/data-acceleration/data-refresh

To load every late row on such a dataset, use engine: cayenne, or set refresh_append_overlap to how late rows can arrive. Append refresh on other data sources does not change. See append refresh, #15147, and #15346.

Spice Cayenne Bug Fixesโ€‹

  • Stale file lists (rare): In rare cases, Cayenne kept a stale file list for a dataset in mode: file. Queries then missed the rows of an appended file with no error. After retention removed a file, every query failed with Failed to open Vortex file โ€ฆ not found instead. Later appends did not repair the list, but a restart did.

    The fault needed a secondary-index build or some queries to list the files while retention or an append changed them. Tests reproduced it only with a delayed file listing. No spiced run has shown it so far. Cayenne now lists files under the same lock that its writers take. See #15304.

  • Vortex file errors after a load: Some keyed Cayenne datasets on network block storage no longer fail every query and refresh with Failed to open Vortex file โ€ฆ not found. This was a regression in earlier v2.4.0 release candidates. The fault needed all of these conditions:

    • mode: file, a primary_key, and key deletion mode (cayenne_deletion_mode: key, or auto when it uses key deletes)
    • a first load or a full refresh with more than one version of some keys
    • network block storage: storage_profile: ebs, or auto when it resolves to EBS or NAS

    Cayenne rewrote the files that held older versions and deleted the originals, but a cached file list still named the deleted files. The data on disk was correct, and a restart cleared the error. Tests reproduced the bug on rc.2. Cayenne now clears the cached list after it deletes the files. See #15303.

  • MySQL CDC snapshots (rc.1 regression): A MySQL dataset with refresh_mode: changes no longer fails its initial snapshot on large tables when a Cayenne file acceleration has maintained aggregates. The error was Bootstrap snapshot error: snapshot row read: โ€ฆ bytes remaining on stream, and the dataset stayed in the Error state.

    Since rc.1, Cayenne ran a checkpoint before each snapshot batch. The checkpoint could rebuild the maintained aggregates with a scan of the whole table. The MySQL reader then waited longer than net_write_timeout, and MySQL closed the connection. Cayenne now runs this checkpoint only when it can hold CDC rows in memory, as before rc.1. In a test with 40 million rows and net_write_timeout=2, the snapshot completed in 36โ€“44 seconds. See #15315.

Cayenne Caching Mode Fixesโ€‹

These fixes apply to Cayenne datasets with refresh_mode: caching.

  • Removed rows served from the cache: An empty origin response now replaces the cached entry of a GET request. Before this release, Cayenne refused that write when the origin started to answer [] for a cached request. Each refusal logged An empty cache replacement requires the complete logical request key. The query that received [] still returned no rows. Later queries could return the rows that the origin had removed:

    • With caching_stale_if_error: enabled, the next origin failure served the removed rows.
    • With caching_stale_while_revalidate_ttl, lookups served the removed rows for the rest of that window. Each lookup started another refresh that Cayenne refused.

    An empty response now replaces only the entry of its own request. Cayenne still refuses an empty response when the lookup filters do not name one request, and it logs An empty cache replacement requires filters that name one request. Entries for a POST request or another query string under the same path stay. DuckDB, SQLite, and Arrow caching datasets still have this fault, because their writer drops empty responses. To get this fix, use engine: cayenne. See #15365.

  • Dropped cache writes under load: Cayenne no longer drops most cache writes above about 100 cache misses per second. Before this release, Cayenne applied each replacement of a cached entry as a separate write. Under load, Cayenne fell behind. The runtime then dropped most writes with Cache write admission timed out. The cache held only part of what it served, so caching_stale_if_error had little data to fall back on. Cayenne now combines replacements and new entries into one write when their requests cannot overlap.

    In a test, a client sent 500 lookups per second over 2.5 million random keys for 300 seconds. Cayenne used mode: file, caching_ttl: 60s, and caching_stale_if_error: enabled. Two runs of each build gave these results:

    MeasureBeforeAfter
    Cache writes dropped121,957โ€“122,5080
    Keys stored15,915โ€“16,624143,356โ€“143,488
    Request P99361โ€“363 ms9 ms

    In a second test, the client cached 60,000 keys, and then the origin stopped. Before the change, 2,932 of 5,001 repeated lookups failed. After the change, 0 of 5,000 failed.

    Cayenne also deletes a replaced entry by its primary key when the request names every primary-key column and the new response has rows. Before, each delete read every stored row, so its cost grew with the cache. In a 60-minute test, the share of CPU for these deletes during origin recovery fell from 3.6% to 0.2%.

    In that phase, P99 latency rose from 303 ms to 433 ms, and throughput fell from 487 to 459 requests per second. The extra writes added compaction work. The test limited the runtime to 8 CPUs and 4 GiB. See #15253 and #15312.

Iceberg Bug Fixesโ€‹

  • Old or uncommitted versions of Hadoop tables: Spice now reads the committed version of a Hadoop-layout Iceberg table. Before this release, a stale version-hint.text kept SQL queries and the Iceberg REST API on an older snapshot. Without a hint, Spice could read the uncommitted metadata file of a writer as the current table. The fix applies to a table read from its root directory, a dataset pinned to version-hint.text, and an Iceberg catalog over a Hadoop warehouse.

    A Hadoop writer commits a version when it renames <uuid>.metadata.json to v<N>.metadata.json. It updates version-hint.text after that, on a best-effort basis. Spice now treats the hint as a starting point and checks v<N+1>, v<N+2>, and later versions, as Iceberg Java does. A metadata file that a bare UUID names no longer counts as a version. See #15142.

  • Tables whose last commit changed only the schema: Queries on an Iceberg table no longer fail when its last commit was ADD COLUMN or RENAME COLUMN with no write after it. Such queries failed with DataInvalid => Column value not found in table. A renamed column did not resolve, an added column was hidden, and a dropped column stayed visible. An Arrow acceleration of the table never loaded, so /v1/ready returned 503.

    A schema-only commit creates no snapshot, and the scan used the schema of the last snapshot. Scans now use the current schema of the table. A scan pinned with snapshot_id keeps the schema of its snapshot. v2.3.2, rc.1, and rc.2 had this bug. A column that someone adds while the runtime runs stays hidden until a restart. Distributed scans pinned to a snapshot still use the schema of that snapshot. See #15334.

AI and Observability Bug Fixesโ€‹

  • Decisions on Bedrock models: ai_if, ai_probability, ai_classify, ai_score, and /v1/decisions now return answers from Bedrock models. Before this release, each such request failed with Evaluation model '<name>' failed: the model refused to answer: . rc.1 and rc.2 had this bug.

    The Bedrock adapter set an empty refusal on every non-streaming reply, and the evaluator read it as a refusal. The adapter no longer sets it, so /v1/chat/completions replies from Bedrock models no longer carry "refusal": "". The evaluator now ignores an empty refusal beside an answer.

    This also covers an OpenAI-compatible model whose upstream is an rc.1 or rc.2 runtime that serves Bedrock. A refusal with text still fails the request. ai_decide uses the same evaluator, but the tests for this fix did not cover it. See #15270.

  • OTLP log trace IDs for pinned queries: A query that a client pins to a trace ID with spice-trace-id or traceparent now exports all of its OTLP log records under that ID. Before this release, records that the runtime logged while the query plan ran carried another trace ID. That ID matched neither the spice-trace-id response header nor the runtime.task_history row. A backend filter on the trace ID missed those records.

    Chat, NSQL, Responses, and decisions requests with a pinned trace ID use the same code path. The tests for this fix covered SQL queries only. This bug did not affect queries without a pinned trace ID. OTLP log export first shipped in rc.2. See observability and #15363.

Distributed Queries Lend Task Slots During HTTP Waitsโ€‹

In a distributed query, a task that waits on a slow HTTP response no longer holds its executor task slot for the whole wait. The task lends the slot to other queries until the response arrives. It then takes a slot back before it continues. Before this release, a few HTTP requests to a slow origin could hold every slot, and other queries waited behind them.

This change needs no configuration. Other data sources and non-distributed queries do not change. In a test, slow HTTP requests filled every executor slot, and a fast query still finished while those requests waited. See #15123.

TLS Certificate Verificationโ€‹

The runtime now uses rustls-webpki 0.103.16 and version 1.0.10 of the webpki-roots and webpki-root-certs root certificate bundles. The bundles follow the current root store of Mozilla. rustls-webpki 0.103.16 rejects a certificate chain that passes through a CA key that upstream no longer trusts. This also applies to cross-signed copies of that key.

The check applies wherever rustls verifies a chain: the bundled roots, rustls-native-certs, and rustls-platform-verifier on Linux. A connection to a server with such a chain now fails certificate verification. Connections through OpenSSL or native-tls use the trust store of the operating system and do not change. See #15374.

Dependency Updatesโ€‹

DependencyUpdate
rustls-webpkiv0.103.15 โ†’ v0.103.16
webpki-roots, webpki-root-certsv1.0.7 โ†’ v1.0.10
Apache Iceberg (spiceai/iceberg-rust, v0.11.0)Fork revision a6b07b8b, which adds the schema fix from apache/iceberg-rust#2904
Apache DataFusion (spiceai/datafusion, based on 55.2.0-rc1)Fork revision 63df5106, which adds the I/O wait hook from spiceai/datafusion#258
Apache Ballista (spiceai/datafusion-ballista, v55)Fork revision 432924cd, which adds task-slot lending from spiceai/datafusion-ballista#76

The upstream versions of DataFusion, Arrow, Iceberg, and Ballista do not change.

Upgradingโ€‹

An upgrade from rc.2 needs no configuration change. TLS connections that rustls verifies now fail if the server's certificate chain passes through a CA that Mozilla no longer trusts. See TLS Certificate Verification.

To upgrade to v2.4.0-rc.3 once the release artifacts are available, use one of the following methods:

CLI:

spice upgrade v2.4.0-rc.3

Homebrew:

brew upgrade spiceai/spiceai/spice

Homebrew follows the version published in the tap. Use the explicit CLI version or Docker tag to select this release candidate.

Docker:

Pull the spiceai/spiceai:2.4.0-rc.3 image:

docker pull spiceai/spiceai:2.4.0-rc.3

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 2.4.0-rc.3

AWS Marketplace:

Spice is available in the AWS Marketplace. Marketplace availability follows its published versions.

Cookbook Updatesโ€‹

The Spice Cookbook provides recipes to help you get started with Spice.

Contributorsโ€‹

What's Changedโ€‹

Changelogโ€‹

  • fix(iceberg): find the committed Hadoop metadata version by @phillipleblanc in #15142
  • feat(iceberg): append refresh follows Iceberg snapshots so late-arriving rows are not missed by @sgrebnov in #15147
  • Drive accelerator refresh writes on the refresh runtime by @sgrebnov in #15262
  • fix(models): answer ai_* decisions from Bedrock models instead of failing them as refusals by @krinart in #15270
  • fix(cache): coalesce Cayenne cache replacements so caching mode keeps up with sustained misses by @krinart in #15253
  • feat(cluster): lend a distributed task's slot while it waits on an HTTP source by @peasee in #15123
  • fix(cayenne): evict a folded snapshot's cached listing once its superseded files are removed by @bjchambers in #15303
  • fix(cache): delete replaced Cayenne cache entries by primary key, not by scanning the table by @krinart in #15312
  • fix(cayenne): list snapshot directories under the listing fence by @bjchambers in #15304
  • fix(cayenne): checkpoint before a durable CDC write only when a source callback is installed by @phillipleblanc in #15315
  • fix(iceberg): read tables whose last commit is schema-only (ADD/RENAME COLUMN) by @lukekim in #15334
  • fix(runtime): skip the refresh_append_overlap warning for Iceberg append datasets by @sgrebnov in #15340
  • fix(spiced): raise the soft open file limit to the hard limit at startup by @bjchambers in #15339
  • fix(iceberg): keyed DuckDB/SQLite append refreshes read by time so late rows don't overwrite newer versions by @sgrebnov in #15346
  • fix(acceleration): restore the strict high-water mark in append dedupe by @sgrebnov in #15358
  • fix(deps): bump rustls-webpki to 0.103.16 and webpki-roots to 1.0.10 by @phillipleblanc in #15374
  • fix(observability): export a pinned query's execution log records under its trace id by @krinart in #15363
  • fix(caching): let an empty origin response replace a cached HTTP entry by @krinart in #15365

Full Changelog: https://github.com/spiceai/spiceai/compare/v2.4.0-rc.2...v2.4.0-rc.3

Spice v2.4.0-rc.1 (Oct 8, 2026)

ยท 64 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Spice v2.4.0-rc.1 is now available! ๐Ÿ”ฅ

Spice v2.4.0-rc.1 is the first release candidate for v2.4.0. It adds performance improvements, S3 event-driven ingestion, SQL results-cache warmup, and adaptive HTTP rate controls. The release also upgrades to DataFusion v55, Ballista v55, Arrow v59, Vortex v0.86, Iceberg v0.11, and Turso v0.81.

Highlights in v2.4.0-rc.1 include:

What's New in v2.4.0-rc.1โ€‹

Performance & Query Engineโ€‹

This release upgrades Apache DataFusion to the v55.2.0 dependency line and Apache Arrow to v59.3.0. It also upgrades Vortex to v0.86.1, Apache Iceberg to the v0.11.0 fork, and Apache Ballista to v55.

Apache DataFusion v55โ€‹

The DataFusion v55 release adds the following improvements:

  • Sort pushdown and TopK pruning: Parquet scans reevaluate each unread row group as the threshold for ORDER BY ... LIMIT tightens. They skip groups that cannot contribute to the result. TopK pruning also supports multiple sort columns.
  • Join planning: The optimizer converts eligible inner joins to semi joins and removes redundant sides of outer joins. It also orders filter predicates by estimated cost.
  • Aggregation and expressions: Multi-column GROUP BY uses column-oriented storage for all supported key types, such as fixed-size binary UUIDs. More string functions preserve dictionary encoding, and IN lists use specialized paths for small integer types.
  • Parquet reads: Scans skip nested fields that the declared schema does not contain. They also skip page-index reads when a file has no page index.
  • Spill handling: Sorts bound the number of streams in a merge. If memory is insufficient, they spill the largest stream again in smaller batches.
  • SQL diagnostics and functions: EXPLAIN accepts PostgreSQL-style options and FORMAT pgjson. New array functions cover element-wise addition, subtraction, scaling, sums, and averages.

Spice carries these changes through its query plans and preserves statistics across plan wrappers. Cayenne keeps Vortex scans below 10 MiB unsplit to avoid repeated footer reads. See #14612.

Vortex v0.79.0 to v0.86.1โ€‹

Spice v2.3.2 used the Vortex v0.79.0 fork. This upgrade covers the full upstream range from v0.79.0 through v0.86.1, not only the v0.86 changes:

  • Types and arithmetic: Vortex adds native Map arrays, Arrow map conversion, and operations and compression for maps. It also adds union arrays and decimal addition, subtraction, multiplication, and division.
  • Row selection: Piecewise-sequence indices represent contiguous selections without an expanded index for every row. Specialized paths handle chunked arrays, fixed-size lists, variable-length lists, and binary values.
  • Scans and expressions: Layout scans gain a physical plan and expression optimization. Filters can pass through scalar functions with multiple arguments. Filters on wide lists restrict child elements to the selected range, and constant masks can resolve from metadata.
  • Compression: Binary arrays support FSST compression with variable-length offsets. OnPair becomes a stable encoding for reads and gains a storage-backed dictionary. Vortex can convert run-end arrays of lists and decimals.
  • File metadata: Files can store custom metadata. Readers cache decoded type descriptions, and file-format editions define the supported encodings and types.
  • Memory and execution: Builders append nested values in batches, preserve list views, and propagate buffer allocators. Expression rewrites retain unchanged nodes, and row functions support batch execution.
  • Correctness and validation: NULL handling changes cover dictionary predicates, BETWEEN bounds, and empty arrays. Readers add validation for footer offsets, compression metadata, and array indices. Other corrections cover nested scalar hashes, decimal operations, and variable-length binary selections above 4 GiB.

See the complete Vortex v0.79.0 to v0.86.1 changelog for every upstream change. Changes to standalone Vortex bindings and GPU execution do not imply new Spice features.

Apache Iceberg v0.11.0โ€‹

Spice updates its Iceberg reader and catalog integrations from v0.10.1 to the v0.11.0 fork for DataFusion 55 and Arrow 59. The upstream v0.11 changes extend reads and catalog compatibility:

  • Iceberg v3 reads: The reader applies deletion vectors from Puffin files and carries row identifiers and sequence numbers through scans.
  • Metadata-only scans: Metadata-only projections do not require data-column reads. Manifest reads reuse partition types, and positional-delete processing buffers runs instead of allocating a key per row.
  • REST catalogs: Clients negotiate server-advertised endpoints and use session-scoped OAuth2 authentication.

The upgrade also aligns Iceberg storage with OpenDAL v0.58. See #14771.

Apache Ballista v55โ€‹

Spice.ai Enterprise feature. See the Enterprise documentation.

The Ballista v55 upgrade adds virtual-core resource accounting for distributed tasks. A protocol-version handshake detects incompatible schedulers and executors. Cancellation identifies tasks by their task IDs, and task-state records use an append-only model.

Schedulers and executors must run the same version. Upgrade all cluster components together.

S3 Event-Driven Ingestionโ€‹

S3 listing datasets can use refresh_mode: changes with S3 event notifications delivered through SQS:

datasets:
- from: s3://my-bucket/events/
name: events
params:
file_format: parquet
s3_region: us-east-1
s3_auth: iam_role
s3_changes_queue_url: ${ secrets:events_queue_url }
acceleration:
enabled: true
engine: cayenne
mode: file
refresh_mode: changes

New object notifications append the object's rows. A periodic listing backfill covers missed or expired notifications. Each dataset needs its own queue and permissions to read and delete SQS messages, alongside its S3 read/list permissions.

This is object ingestion: removal notifications are ignored by default. Set s3_on_object_removed: rebuild to rebuild the entire prefix when an object is removed. An overwrite of an already applied object key is not ingested again; use new object keys for incoming data. See #14121.

SQL Results-Cache Warmup and Shared Fetchesโ€‹

The SQL results cache can persist query plan shapes and replay them after a dataset's first full or append refresh:

runtime:
caching:
sql_results:
enabled: true
warmup: on_first_refresh

For example, run these queries against an accelerated orders dataset with warmup enabled:

SELECT id, status FROM orders WHERE id = 1;
SELECT id, status FROM orders WHERE id = 2;

Spice records one query shape because only the equality-filter value differs. After a restart and the dataset's first full or append refresh, warmup reruns that shape with distinct id values from the refreshed dataset, filling the cache before the dataset becomes ready. Keep the local .spice/data directory across restarts, or configure runtime.state.location, to retain recorded query shapes.

Warmup replays up to ten distinct recorded query shapes and tries up to 1,024 distinct filter-value combinations per shape, stopping when the cache is full. A dataset stays not ready until warmup completes. Later refreshes do not repeat warmup. The feature requires the default plan-based cache key; cache_key_type: sql is incompatible. See #14178.

Concurrent cache-miss fetches for the same request now share a source fetch. SQL results caching also includes changes for tables updated during query execution and for stale results served while revalidation runs. HTTP dataset caching supports RFC 5861 stale-if-error handling. See #14142, #14710, #14708, and #14134.

SQL, search, and embedding caches now use Spice's sharded cache backend. Existing engine: moka and engine: pingora values are accepted for configuration compatibility but no longer select a backend.

Adaptive HTTP Rate Controlsโ€‹

HTTP rate controls now adapt admission to upstream failures while staying within configured request limits. rate_control_acquire_timeout bounds how long a request waits for capacity and defaults to the connector's client timeout. rate_control_failure_threshold and rate_control_window control the response to upstream failures.

In Spice.ai Enterprise, instances that share a runtime.state.location coordinate per-second and per-minute limits through shared state; OSS instances keep these limits in memory independently. Concurrency limits remain local to each instance. Components sharing an upstream origin must use matching rate-control settings. See HTTP rate-control documentation and #14143.

ORC Files and Object Metadata Queriesโ€‹

Listing connectors support file_format: orc. Object-store listing also uses predicates on metadata columns, including _last_modified, to narrow eligible objects. Queries that select only partition or metadata columns can use those values without reading file contents. See #14075, #14265, #14303, and #14116.

Hugging Face Datasetsโ€‹

The new Hugging Face data connector queries and accelerates datasets from the Hugging Face Hub. It supports Parquet, CSV, TSV, JSON, and ORC files. Public datasets need no credentials:

datasets:
- from: hf://datasets/stanfordnlp/imdb/plain_text/
name: imdb
acceleration:
enabled: true

The location format is hf://datasets/<owner>/<dataset>[@<revision>][/<path>]. A path can select a file, a folder, or a glob. A revision can name a branch, a tag, or a commit. Use @~parquet to select the Hub's automatic Parquet conversion.

Set hf_token for private or gated datasets. Set hf_endpoint for a Hub mirror or proxy. Each scan reads one commit. Refreshes follow the selected branch, but the dataset keeps its registered schema until reload. See #14877.

Automatic Primary-Key Handlingโ€‹

Cayenne keeps one row per primary_key without requiring an on_conflict policy. When a dataset sets time_column, the row with the newest time wins; without it, the last arrival wins. This applies to full and append refreshes as well as writes. Existing explicit conflict policies remain accepted during the deprecation period.

datasets:
- from: s3://my-bucket/orders/
name: orders
time_column: updated_at
params:
file_format: parquet
acceleration:
enabled: true
engine: cayenne
mode: file
primary_key: id

For a read-write dataset whose writes should stay in its acceleration, set acceleration.write_mode: acceleration. Its source need not support writes. This mode cannot be combined with a dataset that refreshes by changes.

Cayenne secondary indexes also support dynamic join filters, and their write handling covers inserts, updates, deletes, and refreshes. See #14726, #14282, and #14593.

PostgreSQL and MySQL replication, and MongoDB change streams, rejected Cayenne datasets that omitted on_conflict. Their validation still required an explicit upsert policy. These sources now accept Cayenne datasets with primary_key alone. See #14880.

For file-mode Cayenne datasets, append refreshes failed with a configuration that combined primary_key, time_column, and retention_sql. The refresh selected a version-resolution path that did not support retention. The refresh now resolves each key's newest version before Cayenne applies retention. See #14878.

Cayenne maintained aggregates now share one compact index of per-key contributions across views. Rebuilds capture concurrent writes and apply them after the scan. The aggregate budget uses 10% of a bounded query pool without the former 512 MiB cap. An unbounded pool retains the 512 MiB budget. See #14762.

Read Datasets from Published Snapshotsโ€‹

Spice.ai Enterprise feature. See the Enterprise documentation.

A dataset can now read published acceleration snapshots directly, without configuring the original source connector:

datasets:
- from: s3://my-bucket/spice/snapshots/orders/
name: orders
params:
file_format: snapshot
s3_region: us-east-1

Spice reads the snapshot metadata to select the engine, restores the published data, and checks for newer snapshots. The dataset is read-only. Snapshot-mode readers can also use S3 notifications delivered through SQS, with periodic checks retained for missed notifications. Each reader process needs its own queue.

This release includes changes to snapshot retries, slow-connection bootstrap, publication metadata, and coordination between snapshot archiving and Cayenne maintenance. Cayenne datasets with a datalake tier cannot create acceleration snapshots. See snapshot documentation, #14529, and #14335.

Connector and Protocol Updatesโ€‹

  • Connector status: ADBC, Databricks Spark Connect and SQL Warehouse, FlightSQL, Glue, HTTP/HTTPS, Iceberg, Localpod, and MongoDB are now Stable data connectors.
  • MCP: support for specification version 2026-07-28, alongside the earlier protocol era. See #14043.
  • GitHub: nested GraphQL pagination and rate-limit pacing updates; the default concurrency limit is now four. See #14179 and #14431.
  • GitHub nested pages: Scans could return incomplete reviews or comments because pagination accepted a short page as complete. The connector now rejects incomplete connections and repeated cursors. It retries a failed nested page without another fetch of the outer page. Datasets with the same token also share the REST quota. See #14862.
  • Iceberg REST: Clients could read an empty dataset because the catalog synthesized metadata without snapshots. The catalog now returns the source metadata for Iceberg datasets that Spice reads unchanged. Other datasets return 400 BadRequestException. Clients need their own storage credentials. See #14588 and Breaking Changes.

Other Fixesโ€‹

The release includes fixes in the following areas; the linked PRs provide details of the changes:

  • Cayenne queries and writes: NULL-aware NOT IN, maintained aggregates, dynamic filters, partition-filter forwarding, memory-mode DML and retention, and primary-key handling across CDC checkpoints. See #14429, #14761, #14370, #14047, and #14344.
  • Startup and reloads: retry datasets with unavailable sources, serve existing accelerations during source outages, and invalidate cached plans and results on catalog replacement or dataset unload. See #14623, #14624, #13914, and #14365.
  • Federation: local evaluation of casts and functions whose source semantics differ, plus filter pushdown changes for DynamoDB, Cosmos DB, and MongoDB. See #14484, #14601, and #14419.
  • HTTP and GraphQL: response-status handling during refresh, retry-budget handling, non-JSON gateway responses, and URL redaction in HTTP errors. See #13538, #14313, #14781, and #14490.
  • Search: deletion of obsolete Elasticsearch chunks, non-finite embedding handling, and source-scan coordination during full-text refresh. See #13960, #13902, and #14663.
  • Models and tools: tool-call-only assistant turns, required tool choices, streaming tool-use completion, model-load diagnostics, and propagation of the API-key principal into MCP tool calls. See #14232, #14460, #14548, and #14828.
  • CDC shutdown and reconnects: source-position recording before accelerations close, and MySQL shared-stream reconnect handling. See #14702 and #14751.
  • SQL weekdays: date_part('dow') aligns with EXTRACT(dow), with Sunday represented as zero. See #14796.
  • Cayenne schema statistics: Decimal bounds could retain an old scale after schema evolution because maintenance published statistics from the previous schema. Cayenne now rejects statistics from an obsolete schema and keeps row counts conservative. See #14856.
  • Vector search: vector_search planning failed after the DataFusion 55 upgrade because a second optimization pass tried to reorder a join with a dynamic filter. The planner now preserves that join's input order. See #14857.

Default accelerator: Datasets and views that enable acceleration without engine now use Cayenne. Explicit engine settings keep their behavior. Storage still defaults to memory. Set mode: file for persistent acceleration. See Breaking Changes for migration guidance.

Dependency Updatesโ€‹

Dependency / ComponentVersion
DataFusionv55.2.0
Apache Arrowv59.3.0
Vortexv0.86.1
Apache Icebergv0.11.0
Apache Ballistav55.0.0
Tursov0.8.1
ADBCv0.24
Rust toolchainv1.98.1

Contributorsโ€‹

Breaking Changesโ€‹

Cayenne is the default accelerator on supported platforms. A dataset or view that omits acceleration.engine switches from Arrow to Cayenne. To retain Arrow, set engine: arrow explicitly before upgrading. Windows keeps Arrow as its default. Persistent datasets should continue to name their engine and use mode: file.

on_conflict is deprecated and scheduled for removal in v3.0. Cayenne automatically keeps one row per primary key, choosing the newest time_column value when configured, or the last arrival otherwise. Existing explicit policies remain supported during the deprecation period. Review those policies before removing them, especially drop or policies that reject conflicting rows.

on_conflict no longer routes writes to the acceleration. For read-write datasets whose writes should stay in the acceleration, use:

acceleration:
enabled: true
engine: cayenne
write_mode: acceleration

This setting cannot be used with refresh_mode: changes, including a connector's default change-stream mode. The default write_through and write_back modes require a writable source.

Cache engine selection is retired. engine: moka and engine: pingora remain accepted but are ignored. Remove the field and use caching_policy to select eviction behavior.

GitHub connector default concurrency is four. Review explicit concurrency settings if your deployment relied on the previous default.

HTTP rate-control waits are bounded by default. Requests waiting for rate-control capacity now time out after the connector's client timeout. Set rate_control_acquire_timeout to a suitable duration, or 0 to retain the previous unbounded wait behavior.

Cayenne acceleration snapshots are unavailable for datalake-tier datasets. Review snapshot settings on datasets using cayenne_datalake_location; this release disables snapshotting that configuration.

Iceberg REST no longer synthesizes metadata for unsupported datasets. GET /v1/namespaces/{namespace}/tables/{table} returns 400 BadRequestException for accelerated datasets, views, and other datasets that Spice does not read unchanged from Iceberg. If a client used this endpoint for schema discovery, use SQL DESCRIBE or information_schema.columns instead. Query these datasets through /v1/sql or Arrow Flight SQL. For eligible Iceberg datasets, clients read the source metadata and need their own storage access. See the Get a table API.

Cookbook Updatesโ€‹

The Spice Cookbook provides recipes to help you get started with Spice.

Upgradingโ€‹

To upgrade to v2.4.0-rc.1 once the release artifacts are available, use one of the following methods:

CLI:

spice upgrade v2.4.0-rc.1

Docker:

Pull the spiceai/spiceai:2.4.0-rc.1 image:

docker pull spiceai/spiceai:2.4.0-rc.1

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 2.4.0-rc.1

AWS Marketplace:

Spice is available in the AWS Marketplace. Marketplace availability follows its published versions.

What's Changedโ€‹

Changelogโ€‹

  • fix(runtime): discard cached logical plans when a hot reload replaces a catalog (fixes #13910) by @claudespice in #13914
  • fix(acceleration): let a schema repair correct a checkpoint without resetting the freshness clock (fixes #13817) by @claudespice in #13894
  • fix(search): filter a chunked Elasticsearch delete on a field that can match the key (fixes #13714) by @claudespice in #13926
  • fix(search): classify a partially non-finite embedding as unindexable on every backend (fixes #13872) by @claudespice in #13902
  • fix: stabilize GitHub tests and bound GraphQL registration (fixes #13762) by @lukekim in #13939
  • fix(postgres): decode versioned JSONB binary replication values by @phillipleblanc in #13962
  • docs: require a reviewed Enhancement before any user-facing surface changes by @lukekim in #13970
  • ci: upgrade spiceio setup action to v0.9.0 by @lukekim in #13971
  • fix(postgres): preserve microseconds in timestamp writeback by @phillipleblanc in #13963
  • feat(hash-index): verify the bloom filter's block index with Verus by @lukekim in #13777
  • build(deps-dev): bump js-yaml by @dependabot in #13989
  • docs: release notes for v2.3.0 by @bjchambers in #13999
  • fix(ci): drop the dangling substrait-compliance submodule pointer by @bjchambers in #14002
  • fix(cayenne): release the keyset bytes an abandoned PK checkout accounted (fixes #13668) by @grokspice in #13925
  • fix(caching): keep a declared key from disabling eviction and stranding stale rows (fixes #13976) by @bjchambers in #13992
  • Add Substrait compliance harness (IBM TPC-H Mode A + FlightSQL Mode B stub) by @lukekim in #13879
  • ci: skip DynamoDB TPC-H benches in OSS testoperator dispatch by @phillipleblanc in #14016
  • docs: update security support and roadmap after v2.3.0 by @phillipleblanc in #14024
  • chore: post v2.3.0 release housekeeping by @bjchambers in #13969
  • test(adbc): guard BigQuery corpus offline and in release gate by @phillipleblanc in #14017
  • fix(duckdb): deny the regexp built-ins DuckDB cannot answer faithfully (fixes #13809) by @claudespice in #13871
  • fix: Update tpch benchmark snapshots for federated/adbc[bigquery].yaml by @app/github-actions in #13984
  • fix(search): drop the chunks a shortened row no longer produces from a chunked index (refs #13717) by @claudespice in #13960
  • Reduce Cayenne allocations during primary-key validation and filtering by @lukekim in #14009
  • test(forks): guard seven fork patches that had no repo-side test by @krinart in #13996
  • fix(deps): bump arrow-rs to correctly-rounded Decimalโ†’Float cast (closes #13978) by @Jeadie in #14012
  • perf(vortex): defer projection setup on filtered scans until the filter resolves by @bjchambers in #14035
  • endgame: include spiceai/skills versioned release by @lukekim in #14031
  • fix(caching): partition doomed entries at the survivor cutoff so eviction converges (closes #13994) by @Jeadie in #14021
  • fix(deps): bump arrow-rs fork pin for Decimal->Float rounding fix by @Jeadie in #14049
  • Fix subqueries with use_source acceleration by @phillipleblanc in #14022
  • fix(cayenne): make DELETE, UPDATE and INSERT work on a mode: memory acceleration (fixes #12008) by @bjchambers in #14047
  • fix: clarify OpenDAL S3 retry warnings by @lukekim in #14040
  • test(s3): run the parquet-overwrite fixtures on RustFS by @bjchambers in #14067
  • feat(cayenne): materialize multi-reference CTEs on the query path by @lukekim in #13918
  • perf(vortex): answer a constant IN list by probing a set, and falsify it by interval by @bjchambers in #14061
  • perf(vortex): skip a scan split whose zones cannot satisfy the filter by @peasee in #14064
  • fix(arrow): report an exact row count from the indexed point-lookup scan by @krinart in #13972
  • fix(cayenne): apply sort_columns with refresh_mode: full by @peasee in #14063
  • fix(smb): pad an empty CREATE buffer so Samba lists the share root (fixes #13293) by @grokspice in #14050
  • perf(cache): key the logical-plan cache on SQL text, not parameter values by @bjchambers in #14069
  • fix: harden HuggingFace E2E chat against slow Metal generation by @lukekim in #14072
  • fix(cayenne): move accelerator filesystem I/O off Tokio workers by @lukekim in #14073
  • test(chbench): enable CTE materialization and IVM on mysql/postgres adaptive HTAP by @lukekim in #14070
  • ci: run Substrait Mode A TPC-H on pull requests and the merge queue by @lukekim in #14071
  • feat(mcp): support MCP specification 2026-07-28 (dual-era) by @lukekim in #14043
  • fix(ci): call a linker that died of a signal an infrastructure failure, not a check failure (fixes #13614) by @grokspice in #14044
  • fix: Provide temporary directory in docker images by @Jeadie in #14089
  • fix: restore OSS installer, CLI and test workflow coverage by @phillipleblanc in #14025
  • Delete v2.2.0.md by @Jeadie in #14095
  • docs(release): add v2.3.1 release notes by @phillipleblanc in #14094
  • fix(test): allow DELETE in the CORS allow-methods assertion by @claudespice in #14098
  • fix(ci): stop install-protoc unzipping into a shared ~/.local by @lukekim in #14097
  • feat(connectors): add ORC listing format via in-repo FileFormat by @lukekim in #14075
  • fix(cayenne): run snapshot bootstrap check before opening the metastore by @Jeadie in #14093
  • feat(cache): verify the results-cache namespace prefix with Verus by @lukekim in #14074
  • feat(cloud-connect): add a GetDatasets command that answers the /v1/datasets document (refs #13369) by @grokspice in #14051
  • Suppress Cayenne startup logs when no Cayenne dataset is configured by @Jeadie in #14042
  • fix(cache): re-bind parameter values when revalidating a stale result (fixes #14099) by @bjchambers in #14100
  • fix(cluster): support distributed HTTP scans by @phillipleblanc in #14108
  • fix(bigquery): keep ILIKE evaluation local by @phillipleblanc in #14110
  • docs: update security support for v2.3.1 by @phillipleblanc in #14117
  • feat(cayenne): reuse ScanView until write, lag only for read-only CDC by @lukekim in #14055
  • fix(deps): remediate open Dependabot alerts by @phillipleblanc in #14111
  • fix(testoperator): validate results in every scale factor 1 TPC-H, TPC-DS and ClickBench benchmark by @lukekim in #14119
  • fix(cayenne): reject ambiguous metastore paths by @phillipleblanc in #14130
  • perf: serve results-cache hits where the request arrives and cut per-hit overhead by @lukekim in #14103
  • test(runtime): record query previews in the management export test by @lukekim in #14155
  • ci: upgrade spiceio setup action to v0.11.0 by @lukekim in #14152
  • feat(caching): Make caching_stale_if_error RFC-5861 compliant (with stale-if-error header) by @Jeadie in #14134
  • perf(runtime-table): defer cache-eviction key extraction to entries a delete actually names by @Jeadie in #14138
  • fix(runtime): report the acceleration.ready_state deprecation once per component (fixes #13749) by @claudespice in #14006
  • fix(runtime): write the inferred Arrow sort order under the prefixed key its validation accepts (fixes #14023) by @claudespice in #14032
  • fix(connectors): Fix JSON/Orca files using metadata columns by @Jeadie in #14115
  • feat(cayenne): build secondary indexes from indexes in file and memory mode by @phillipleblanc in #14149
  • fix(cayenne): round-trip decimal, binary, and time stats and drop them on scale change by @lukekim in #14139
  • feat(cayenne): cluster warm and datalake tiers, and write full refreshes as key-range files by @lukekim in #14124
  • fix(cayenne): compile the cold-tier pruning test and backtick a doc literal by @lukekim in #14175
  • fix(cayenne): make the crates own targets lint and compile by @phillipleblanc in #14200
  • test(forks): guard five more fork patches, and drop a row that is not fork state by @krinart in #14015
  • perf(cache): promote encoded SQL results to raw after the second decode by @lukekim in #14199
  • fix(turso): build a dictionary column directly so a dictionary over a list, map or boolean value reads back (fixes #13033) by @grokspice in #14181
  • fix(ci): probe the macOS toolchain before reaching for brew in the release builds by @grokspice in #14203
  • fix(ci): skip Metal kernel precompilation in the macOS release build by @grokspice in #14204
  • fix(github): paginate nested GraphQL connections and pace to GitHub's rate limits by @lukekim in #14179
  • fix(vortex): stop an IN list holding a NULL from panicking the scan by @krinart in #14163
  • bench(cayenne): use std::hint::black_box in the clustering bench by @lukekim in #14129
  • fix(runtime-table): stop rebuilding SessionContext on every cache fetch by @Jeadie in #14141
  • test(chbench): cluster order_line, oorder and customer on the adaptive HTAP arms by @lukekim in #14192
  • fix(runtime): count a first load as still loading in the Dataset load summary (fixes #13974) by @claudespice in #14020
  • fix(duckdb): push regexp_count down again at a rendering that counts as the kernel does (fixes #13870) by @claudespice in #14153
  • build: lint and test the sign-off under the same profile as the merge queue by @lukekim in #14180
  • ci: require the Verus proofs in the merge queue as one check by @lukekim in #14189
  • ci: run the longest macOS jobs on their own runner pool by @lukekim in #14229
  • fix(cayenne): build the DELETE sink inside the execution-time write lock (fixes #13828) by @claudespice in #14218
  • build(deps): bump the github-actions-dependencies group across 1 directory with 7 updates by @dependabot in #14231
  • fix(cluster): decide what a Flight message carries by its IPC header, not its body length (refs #13737) by @claudespice in #14212
  • ci: run Mode A TPC-H on merge queue and trunk/release push only by @lukekim in #14247
  • chore(deps): bump spiceai/duckdb-rs to 76655d2f by @lukekim in #14246
  • ci: stop exporting empty AWS and DuckLake endpoints to the schema test by @phillipleblanc in #14194
  • fix(runtime): reload a localpod dataset when the dataset it reads through is reloaded (fixes #3288) by @claudespice in #14208
  • build(deps): bump the aws-sdk group with 3 updates by @dependabot in #14255
  • fix(ci): resolve Homebrew prefix when brew is the spice flock wrapper by @lukekim in #14210
  • chore(deps): raise the datafusion-table-providers pin to include the NUMERIC result-column fix by @phillipleblanc in #14254
  • ci: align the DuckLake bootstrap with the embedded DuckDB, wait for Databricks startup, and stop dispatching legs that cannot pass by @phillipleblanc in #14250
  • fix(runtime): install the Spice function deny-list on the PostgreSQL catalog connector (refs #13664) by @claudespice in #14225
  • perf(cayenne): share inline-cache view entries by Arc instead of cloning them per scan by @krinart in #14191
  • build(deps): bump aws-actions/configure-aws-credentials by @dependabot in #14256
  • ci: stop dispatching the indexed turso TPC-H SF1 tests by @phillipleblanc in #14252
  • fix(catalog): keep the tables registered under an existing schema by @phillipleblanc in #14193
  • feat(cache): Spice sharded cache as the sole LruCache engine by @lukekim in #14206
  • ci: lint GitHub Actions definitions with actionlint, and fix the 73 findings it surfaced by @grokspice in #14223
  • perf(cache): tighten the Raw SQL results-cache serve path by @lukekim in #14205
  • ci: stop triggering the CUDA build on pull requests by @lukekim in #14267
  • ci: run CodeQL on pull requests and the merge queue by @lukekim in #14269
  • Release 2.3.2 release notes by @krinart in #14271
  • chore: make AGENTS.md the canonical agent instructions by @lukekim in #14237
  • ci: run CodeQL Analyze on spiceai-dev-runners by @lukekim in #14281
  • feat: TypeSafe Jev System One evaluation provider by @lukekim in #14215
  • fix(cayenne): tag file statistics bounds as the column's Arrow type (fixes #14280) by @phillipleblanc in #14283
  • ci: install spiceio when the runner has no gh (refs #14233) by @lukekim in #14288
  • test(forks): 11 repo guards by @krinart in #14261
  • docs: add Spice.ai in Action manuscript and companion labs by @lukekim in #13965
  • test(duckdb): add an integration test for the index CTE materialization by @sgrebnov in #13885
  • fix(cayenne): coalesce inline writes into one batch by @sgrebnov in #14279
  • release: Update SECURITY.md and endgame template after 2.3.2 by @peasee in #14293
  • fix(cayenne): count each Arrow allocation once in the inline-cache gauge by @krinart in #14272
  • feat(s3): SQS event-driven changes for refresh_mode: changes by @lukekim in #14121
  • feat(caching): single-flight coalesce concurrent cache-miss fetches by @Jeadie in #14142
  • fix(caching): make caching_stale_if_error detect transient HTTP failures on real schemas by @krinart in #14161
  • Use Cayenne secondary indexes for dynamic join filters by @phillipleblanc in #14282
  • fix(ci): compare DynamoDB sets without relying on element order by @bjchambers in #14328
  • docs(release): Remove QA analytics step from endgame by @peasee in #14329
  • ci: run CodeQL on the merge queue and trunk, not on pull requests by @lukekim in #14290
  • docs(cayenne): reposition the reference, add query serving, re-audit against trunk by @lukekim in #14338
  • docs(endgame): update versioned docs release steps by @ewgenius in #14277
  • test(forks): guard the ballista per-task file-scan restriction by @krinart in #14292
  • fix(cli): surface the full error chain for spice chat connection failures by @krinart in #14289
  • fix(ci): degrade the incomplete-sign-off handler when the runner has no gh (refs #14234) by @claudespice in #14304
  • fix(runtime-table): serialize a direct write against acceleration snapshot creation (fixes #13548) by @claudespice in #14310
  • fix(duckdb): screen regexp_like and regexp_replace as regexp_count is screened (fixes #14148) by @claudespice in #14321
  • fix(http): honor retry budget without an extra origin request by @phillipleblanc in #14313
  • Fix SchemaCastScanExec's schema conversion in fn partition_statistics by @Jeadie in #14258
  • fix(cluster): recognise a Flight keepalive by its empty envelope, not by what its header declares (fixes #13737) by @claudespice in #14327
  • perf(http): defer zero-TTL acceleration lookup until origin failure by @phillipleblanc in #14302
  • fix(cayenne): keep a key visible when it is re-inserted over a stale-insert tombstone by @sgrebnov in #14312
  • perf(cayenne): read only key and filter columns in filtered key deletes by @sgrebnov in #14374
  • fix(cayenne): let small protected-snapshot merges run during a long large-tier merge by @sgrebnov in #14296
  • perf(cayenne): serve primary-key lookups on a freshly loaded table in ~1 ms by @lukekim in #14314
  • fix(cayenne): skip min/max statistics for nested columns (fixes #14368) by @sgrebnov in #14392
  • test(caching): cover SchemaCastScanExec statistics projection by name, retype, and SQL filter by @Jeadie in #14146
  • fix(turso): keep a quantified comparison out of Turso SQL (fixes #14041) by @grokspice in #14393
  • fix(udfs): declare the local_embed dev-dependency the embed tests need (fixes #13092) by @claudespice in #14377
  • perf(cayenne): batch metastore manifest rewrites, upsert in place, and run every write on one writer connection by @lukekim in #14369
  • fix(runtime-table): ignore zero-row batches in stale fallback by @phillipleblanc in #14331
  • Prune object-store file listing by _last_modified predicates by @Jeadie in #14265
  • Reading only partition or metadata columns needlessly scans all file contents by @Jeadie in #14116
  • fix(duckdb): keep a concat over a binary operand out of the federated plan (fixes #13915) by @claudespice in #14333
  • fix(ci): keep the sign-off attribution inside GitHub's 140-character status cap (fixes #14076) by @grokspice in #14390
  • fix(ci): expose a present-but-unlinked cc tool on macOS runners instead of routing it through brew install (fixes #13479) by @grokspice in #14387
  • ci: re-measure integration.yml's job bounds after the archive consolidation (fixes #13429) by @grokspice in #14388
  • fix(ci): start DuckLake's local MinIO from an image that is still published by @grokspice in #14399
  • test(runtime-table): make the metric-scraping refresh tests pass under cargo test by @claudespice in #14381
  • fix(federation): keep a correlated subquery predicate above a join of two sources (refs #8220) by @claudespice in #14372
  • fix(caching): snapshot staleness before the origin fetch for stale_if_error by @Jeadie in #14263
  • perf(snapshots): skip unchanged snapshot metadata with a conditional GET by @sgrebnov in #14409
  • fix(search): prune the rest of a key group from an Elasticsearch chunked index (refs #13717) by @claudespice in #14320
  • fix(bench): derive MySQL's empty-field NULL handling from the column type (refs #13152) by @claudespice in #14345
  • fix(graphql): debit a LIMIT by the rows a page returned, not the declared page size (fixes #14308) by @claudespice in #14353
  • fix(ci): fit retention_oom's retry budget inside its workflow step, and guard the coupling (fixes #13512) by @grokspice in #14389
  • docs: say plainly what Spice is, refresh the README for v2.3, and promote connector statuses by @lukekim in #14410
  • fix(ci): pin, checksum and retry the oha download in the E2E graceful-shutdown jobs by @grokspice in #14418
  • fix(snapshots): resolve snapshot entries relative to the metadata location (#14425) by @sgrebnov in #14426
  • Lower the GitHub connector default concurrency limit to 4 by @lukekim in #14431
  • build(deps): bump nvidia/cuda in the docker-dependencies group by @dependabot in #14439
  • build(deps): bump the aws-sdk group with 3 updates by @dependabot in #14440
  • build(deps): bump the github-actions-dependencies group across 1 directory with 5 updates by @dependabot in #14441
  • test(cayenne): bound the refused-build guard by builds, not by the host's speed by @grokspice in #14424
  • fix(cache): don't report an invalidation cancelled by runtime shutdown as a failure by @grokspice in #14417
  • ci: run remote sign-off on the spiceai-macos pool by @lukekim in #14449
  • ci: run CodeQL Analyze on spiceai-macos by @lukekim in #14442
  • fix(runtime): keep the built-in date_part so both weekday spellings agree (fixes #13920) by @claudespice in #14154
  • fix(cayenne): include the in-memory CDC tier when an overwrite or a retention pass covers the whole table by @lukekim in #14428
  • fix: keep NOT IN null-aware through the join reorder and the Cayenne sort-merge rewrite by @lukekim in #14429
  • fix(cayenne): discard a compaction whose snapshot an overwrite replaced mid-pass by @lukekim in #14432
  • fix(cayenne): stop maintained views, Vortex IN lists and dynamic-filter sharing from returning wrong rows by @lukekim in #14427
  • fix(cayenne): hide spilled rows on CDC upsert fallback by @bjchambers in #14416
  • fix(ci): skip the integration, ADBC, chDB and E2E gate jobs on pull requests instead of passing them (fixes #13841) by @grokspice in #14454
  • fix(cayenne): judge a filtered key delete's captured sources by the index captured with them (refs #13913) by @claudespice in #14455
  • fix(ci): clear the macos-15 image's [email protected] symlink before installing MySQL by @lukekim in #14489
  • ci: produce CodeQL SARIF in the Analyze job on spiceai-macos by @lukekim in #14500
  • fix(cayenne): correctness fixes for the goal-driven adaptive controller, with a closed-loop simulation harness by @lukekim in #14443
  • fix(cdc): keep the newest source commit timestamp on a coalesced change batch by @lukekim in #14463
  • fix(cayenne): draw a sequence for a current-snapshot append so per-key OCC can order it (fixes #13685) by @claudespice in #14360
  • fix(http): name a configured model's load failure instead of reporting it not found (fixes #13303) by @claudespice in #14395
  • fix(duckdb): keep inferred source indexes off change-stream accelerations so upserts commit under concurrent reads (refs #13929) by @claudespice in #14396
  • fix(duckdb): keep a text cast over a binary operand out of the federated plan (fixes #14355) by @claudespice in #14448
  • fix(llms): honor tool_choice required and allowed_tools on mistral.rs-hosted models instead of panicking (fixes #14230) by @claudespice in #14460
  • test(runtime): run the load-error counter test in its own process (fixes #13085) by @claudespice in #14462
  • fix(runtime): stop a replaced dataset configuration's load from registering over the new one (fixes #1458) by @claudespice in #14367
  • fix(install): stop asking for sudo on a first install into a fresh HOME (fixes #14445) by @claudespice in #14495
  • perf(cayenne): serve primary-key point lookups in half the time by @lukekim in #14433
  • fix(deps): bump DataFusion for upstream fixes to wrong results from filter pushdown, simplification and planning by @lukekim in #14430
  • fix(runtime): size every internal DataFusion session from the CPU budget by @bjchambers in #14412
  • fix(runtime): serve nested, zoned and half-float columns from the Iceberg catalog API (fixes #4815) by @claudespice in #14480
  • fix(acceleration): keep cached results when a snapshot refresh finds no newer snapshot by @sgrebnov in #14497
  • fix(runtime): sleep cron tests to the next boundary, not one that already fired (fixes #13759) by @claudespice in #14474
  • fix(ci): derive every lint-rust guard make runs, whatever its separator or recipe layout (fixes #13783) by @claudespice in #14475
  • fix(cayenne): stop the small-file compaction of a position-mode PK table from deadlocking on its own write lock (fixes #14420) by @claudespice in #14481
  • fix(runtime-table): report refresh bytes for the rows each batch holds by @Jeadie in #14469
  • fix(spark,databricks): keep Spice-only functions out of SQL sent to Spark Connect and Databricks SQL Warehouse (refs #13664) by @claudespice in #14498
  • fix(vortex): size a cached footer by what it retains, not its serialized bytes (fixes #12917) by @claudespice in #14502
  • fix(cayenne,telemetry): Register accelerated sink dataset immediately from existing acceleration by @peasee in #13955
  • Update spicepod.yml by @Jeadie in #14533
  • fix(cayenne): keep a rewrite's count inexact when it retains a late protected snapshot (fixes #14383) by @claudespice in #14385
  • fix: Update tpch benchmark snapshots for accelerated/on_zero_results/file[parquet]-cayenne[file]-on_zero_results.yaml by @app/github-actions in #14408
  • build(rust): upgrade toolchain to 1.98.1 by @lukekim in #14560
  • fix(postgres): release a shared slot's hold on a table its publication cannot drop (fixes #13032) by @claudespice in #14527
  • fix(cayenne): size the build side before sort-merging a non-Cayenne outer join by @krinart in #14520
  • Update DF Upgrade template by @krinart in #14526
  • fix(ci): wait for the refresh to invalidate the results cache, not a fixed 3s by @grokspice in #14598
  • fix(deps): move the DataFusion pin past the four unparser fixes, and guard each of them (fixes #13022) by @grokspice in #14570
  • fix(dynamodb, cosmosdb, mongodb): push filters down only where the source evaluates them as SQL does by @lukekim in #14419
  • feat(snapshots): serve a dataset from published snapshots with from: s3://โ€ฆ and file_format: snapshot by @lukekim in #14529
  • perf(cayenne): keep the per-shard PK index across checkpoint flushes and back off futile bakes; fix q17 and two wrong-results races (fixes #14235) by @lukekim in #14555
  • test(chbench): lower the MySQL adaptive pods' CDC linger to 500 ms by @lukekim in #14615
  • fix(graphql): show the parse-failure bytes in JSON decode previews by @Jeadie in #14530
  • Defer source-first cache fallback planning by @phillipleblanc in #14556
  • perf(cayenne): order whole-table rewrites per scan partition, then merge by @bjchambers in #14488
  • fix(ci): tolerate the parquet-rename race's third DuckDB error in the E2E log scan by @grokspice in #14541
  • fix(graphql): retry an inferred credential refusal instead of failing the refresh by @Jeadie in #14539
  • fix(cayenne): serve a widened table's files from their persisted statistics (refs #13829) by @claudespice in #14316
  • fix(acceleration): checkpoint DuckDB's write-ahead log before a snapshot copies its file (fixes #13912) by @claudespice in #14325
  • fix(ci): let E2E cleanup run when the job failed before creating its working directory by @grokspice in #14553
  • Replace unmaintained backoff with a workspace crate by @phillipleblanc in #14621
  • test: read Expect stand-ins with the installed shell by @phillipleblanc in #14634
  • fix(models): apply a tool_choice that forces a call to one round of the tool-use loop (fixes #14459) by @claudespice in #14635
  • fix(runtime): don't end a tool-use stream on a stale tool_calls finish (fixes #13309) by @claudespice in #14548
  • fix(smb): name listed locations under the share so directory datasets load (fixes #14060) by @claudespice in #14549
  • fix(models): name a configured model's load failure in ai() and streaming nsql (fixes #14394) by @claudespice in #14632
  • test(chbench): remove the mysql-cayenne[file]-adaptive-split HTAP arm by @sgrebnov in #14620
  • fix(cayenne): record sharded CDC keys in the table-wide PK index by @lukekim in #14603
  • fix(sqlite): keep TRY_CAST and every cast SQLite answers differently out of the federated plan (fixes #14398) by @claudespice in #14496
  • ci: move remaining Ubuntu 22.04 references to 24.04 by @lukekim in #14677
  • fix(snapshots): retry a failed snapshot attempt with backoff by @sgrebnov in #14676
  • fix(snapshots): large snapshots no longer fail bootstrap on slow connection by @sgrebnov in #14571
  • fix(refresh): start a refresh's source scan only when the sink reads it, so the full-text index keeps its rows (fixes #14619) by @claudespice in #14663
  • test(cayenne): compare suite answers cell by cell against SQLite and chDB by @lukekim in #14473
  • Evaluate any chat model via /v1/evaluate by @lukekim in #14568
  • ci: quarantine the management API integration schedule until its dev OAuth client is reactivated (refs #12376) by @grokspice in #14684
  • ci: stop scheduled Testoperator Ballista benchmarks by @phillipleblanc in #14682
  • fix(cayenne): stop snapshotting datasets with a datalake tier, whose restored copies deleted each other's files by @lukekim in #14583
  • fix(cayenne): count each Arrow allocation once for the mem-tier limit and checkpoint write sizing by @krinart in #14675
  • fix(ci): classify a sign-off that reached its step budget as signalled, so an expiry publishes no verdict (fixes #13843) by @grokspice in #14685
  • fix(runtime): retry a dataset whose source is unreachable at startup (fixes #14609) by @bjchambers in #14623
  • fix(cayenne): archive only the snapshot directories a dataset snapshot references (fixes #14605) by @sgrebnov in #14627
  • ci: route trunk-push macOS builds to the standard pool by @phillipleblanc in #14645
  • ci: install strip for the retention OOM regression test by @phillipleblanc in #14701
  • fix(smb): serve every share on a host from the one store registered for it (fixes #14550) by @grokspice in #14689
  • fix(runtime): discard cached plans and results when a dataset is unloaded (refs #14251) by @claudespice in #14365
  • feat(snapshots): reload snapshot-mode datasets from S3 event notifications (SQS) by @lukekim in #14335
  • fix(cayenne): report runs per size tier when protected-snapshot compaction declines (fixes #13622) by @claudespice in #14476
  • fix(cayenne): re-run a post-write compaction pass a concurrent append asked for (fixes #13906) by @claudespice in #14479
  • fix(http): keep the request URL out of HTTP connector errors (fixes #13534) by @claudespice in #14490
  • fix(runtime): load every localpod dataset that reads from one parent at startup (fixes #13087) by @claudespice in #14532
  • fix(build): share one cargo metadata helper across the lint guards, so a broken cargo is never a violation (fixes #13121) by @claudespice in #14534
  • fix(duckdb): push regexp_replace down again for a one-digit group reference, and refresh the Postgres ClickBench q35 plan (refs #13966) by @claudespice in #14552
  • fix(duckdb): keep a cast into binary out of the federated plan (refs #14397) by @claudespice in #14633
  • fix(runtime): keep finished async query jobs finished, and delete expired ones by @lukekim in #14585
  • fix(federation): keep DataFusion's cast built-ins out of every federated plan (fixes #14444) by @claudespice in #14484
  • test(cayenne): cover a swapped join filter through the sort-merge rewrite end to end (refs #14235) by @claudespice in #14551
  • fix(runtime): decide every Spark/built-in function collision by name, and refuse an undecided one (fixes #14361) by @grokspice in #14400
  • fix(ci): run every bin target's unit tests in the sign-off gate, and give spice connect its own --cloud-region refusal (fixes #13426) by @grokspice in #14406
  • test(data_components): compile the federation unparser guards in every scoped test run (fixes #13625) by @grokspice in #14447
  • fix(runtime-component): skip an inferred Cayenne index on a floating-point column instead of failing the load (fixes #14590) by @grokspice in #14599
  • fix(cache): serve stale cached results during frequent table updates (stale_while_revalidate_ttl) by @sgrebnov in #14708
  • Update Turso to 0.8.1 and retry write conflicts a metastore statement raises by @lukekim in #14680
  • fix(snapshots): stop a snapshot dataset's load when a reload replaces it by @lukekim in #14673
  • fix(cayenne): hand non-partition filters to every partition scan (fixes #12959) by @claudespice in #14370
  • fix(postgres-accel): resolve secret references in the sidecar connection parameters (fixes #13296) by @claudespice in #14544
  • fix(cayenne): drain the in-memory CDC tier before a full rewrite scans it (fixes #14450) by @claudespice in #14486
  • feat(caching): warm SQL results cache on first refresh from persisted plan shapes by @lukekim in #14178
  • fix(ci): retry only the container startup in the zero-retry MySQL CDC tests by @grokspice in #14727
  • feat(key-index): immutable secondary index runs over compound Arrow keys by @bjchambers in #14592
  • fix(federation): keep a fractional-to-integer cast out of the plans pushed to DuckDB, PostgreSQL, MySQL and BigQuery (fixes #14482) by @grokspice in #14601
  • fix(testoperator): give accelerated bench configs a 300s ready_wait and name unready datasets on timeout (fixes #13973) by @claudespice in #14728
  • fix(federation): keep arrow_typeof and its plan-introspection siblings local on every backend (fixes #14334) by @grokspice in #14695
  • test(bench): refresh the tpch_q16 explain snapshots for the null-aware NOT IN plan (fixes #13977) by @claudespice in #14731
  • ci: build trunk-push macOS legs on spiceai-macos-large again, keeping one-at-a-time coalescing by @grokspice in #14735
  • ci(verus): derive the verified crates from cargo metadata; pin vstd once by @bjchambers in #14709
  • docs: raise the test and evidence bar: differential first, exact assertions, performance always measured by @lukekim in #14723
  • test(vortex): wait for a dropped segment cache to be freed before asserting it is gone (fixes #13295) by @claudespice in #14470
  • ci(integration): report each integration part's own result in its required check by @grokspice in #14743
  • fix(cayenne): back off futile bakes under a violated query goal too by @lukekim in #14736
  • feat(object_store_occ): add transactional WAL and MVCC snapshots by @lukekim in #14732
  • fix(snapshots): bootstrap readers from late snapshot publications by @phillipleblanc in #14468
  • fix(cayenne): apply retention_sql to a mode: memory acceleration (fixes #14045) by @claudespice in #14344
  • fix(duckdb, sqlite): keep the first copy of a key a write repeats under on_conflict: drop (fixes #14629) by @claudespice in #14748
  • fix(cache): release a removed dataset's memory when the plan cache discards its plans (fixes #14251) by @claudespice in #14760
  • build(deps): bump the aws-sdk group with 2 updates by @dependabot in #14764
  • fix(llms): explain an Anthropic model's refusal of a sampling control instead of passing the bare 400 through (fixes #13564) by @grokspice in #14700
  • build(deps-dev): bump dompurify by @dependabot in #14613
  • fix(snapshots): allow cayenne_file_path and cayenne_metadata_dir on file_format: snapshot datasets (fixes #14696) by @sgrebnov in #14698
  • test(snapshots): make snapshot integration tests robust under parallel CI runs by @sgrebnov in #14765
  • feat(cayenne): keep the secondary index current across every write by @bjchambers in #14593
  • fix: Update tpch benchmark snapshots for accelerated/file[parquet]-duckdb[memory].yaml by @app/github-actions in #14674
  • Isolate Docker integration fixtures across concurrent processes by @phillipleblanc in #14639
  • ci: run Substrait Mode A TPC-H at SF 1 on spiceai-macos in every merge-queue entry by @lukekim in #14522
  • ci: stop repeating scheduled benchmarks: one source per accelerator, hosted sources weekly, a release commit once by @lukekim in #14681
  • build(deps): bump nvidia/cuda by @dependabot in #14763
  • fix(http): stop a non-2xx response body replacing an accelerated table's rows (fixes #13515) by @claudespice in #13538
  • fix(chat-api): answer a tool-call-only assistant turn instead of panicking (fixes #13207) by @grokspice in #14232
  • fix(sqlite, duckdb): keep a decimal AVG, and on SQLite a decimal SUM, out of the federated plan (fixes #14492) by @claudespice in #14670
  • fix(spiceai, duckdb, sharepoint): name the Spicepod key a missing-parameter error asks for (refs #14446) by @claudespice in #14769
  • Add table-bound ChangeSink ownership and backends by @phillipleblanc in #14704
  • fix(ci): serialize rustup installs on shared macOS runners by @lukekim in #14717
  • fix(cayenne): keep the per-shard PK index when the table-wide index is discarded by @lukekim in #14604
  • Upgrade to DataFusion 55.1 and Arrow 59.3 by @krinart in #14612
  • Update openapi.json by @app/github-actions in #14742
  • build(deps): bump rustls from 0.23.40 to 0.23.45 by @dependabot in #14773
  • Upgrade to iceberg-rust to 0.11.0 and DF 55.2 by @sgrebnov in #14771
  • fix(cayenne): write an append into an empty unkeyed table as one load by @phillipleblanc in #14784
  • fix(mysql_replication): don't re-send a live member's delivered commits after a reconnect by @lukekim in #14751
  • ci(codeql): don't fail the SARIF upload when the merge queue already deleted its ref by @grokspice in #14793
  • fix(acceleration): accept time_format iso8601 when the time_column is already a timestamp by @lukekim in #14777
  • fix(cache): store SQL results that go stale mid-query by @lukekim in #14710
  • fix(cayenne): serve filtered and global maintained aggregates by @lukekim in #14761
  • perf(cayenne): build the checkpoint's tombstone union after releasing the capture locks by @lukekim in #14759
  • ci: keep no artifacts or caches from PR and merge-queue checks, and run gating macOS jobs on spiceai-macos-large by @lukekim in #14804
  • fix(cayenne): keep maintenance from deleting files a snapshot is archiving by @sgrebnov in #14789
  • fix(snapshots): never overwrite shared snapshot metadata, and publish more than once to a file:// location by @lukekim in #14582
  • perf(cdc): build deferred change rows in the reader while the apply runs by @lukekim in #14746
  • fix: Update tpch benchmark snapshots for accelerated/on_zero_results/file[parquet]-cayenne[file]-on_zero_results.yaml by @app/github-actions in #14786
  • test(testoperator): add a cold-start time-to-ready regression test by @phillipleblanc in #14785
  • test(testoperator): make append tests more robust by @sgrebnov in #14817
  • fix(runtime,data_components): wait for change-data-capture sources to record their positions before shutdown closes the accelerations (fixes #14523) by @grokspice in #14702
  • fix(graphql): detect non-JSON responses and treat gateway errors as retryable by @lukekim in #14781
  • fix(runtime): require keys for s3_auth: key, honor gs:// state location params, and leave newer rate-control state alone by @lukekim in #14584
  • build(deps): bump hickory-resolver from 0.26.1 to 0.26.2 by @dependabot in #14774
  • fix(listing): skip zero-byte objects (S3 folder markers) in format-selected listings by @sgrebnov in #14822
  • Avoid unnecessary repartitioning in indexed dynamic-filter joins by @bjchambers in #14799
  • feat(cayenne): keep one row per primary key automatically and deprecate on_conflict by @bjchambers in #14726
  • fix(object-store): stop reading a response body at its first error by @phillipleblanc in #14831
  • fix(ci): use byte-order collation in the benchmark Postgres container (refs #14815) by @sgrebnov in #14836
  • fix(listing): keep a partition predicate as a residual filter on the _location fast path by @grokspice in #14790
  • fix(ci): search only the unixodbc keg, not all of Homebrew's lib, in the macOS release builds by @grokspice in #14846
  • fix(federation): keep date, timestamp and interval values local on SQLite reached through ADBC or ODBC (fixes #14753) by @claudespice in #14840
  • chore: pin the fork at spiceai/datafusion#249 and guard the DataFusion fixes backported from 54 by @krinart in #14800
  • Revert "fix(sqlite, duckdb): keep a decimal AVG, and on SQLite a decimal SUM, out of the federated plan (fixes #14492) (#14670)" by @krinart in #14825
  • perf(runtime): skip EnsureRequirements while planning a point lookup by @lukekim in #14807
  • fix(testoperator): report CH-benCH queries with no rows on either side as vacuous, not as matches by @lukekim in #14810
  • feat(acceleration): make Cayenne the default acceleration engine by @phillipleblanc in #14837
  • fix(cayenne): keep a join's LIMIT when the oversized-join rewrite makes it a sort-merge join by @lukekim in #14805
  • test(bigquery): exclude subtrees an empty join build side never runs from the corpus job count (fixes #14848) by @sgrebnov in #14849
  • fix(cayenne): fail contradictory write settings once, and order NULL times below every time by @bjchambers in #14847
  • fix(runtime): serve an existing acceleration while its source is unavailable (fixes #14610) by @bjchambers in #14624
  • Prune object-store file listing by metadata columns (#14264) by @Jeadie in #14303
  • fix(vortex): fold partition values into the file-pruning predicate; test null-equal joins under mode:file by @lukekim in #14797
  • feat(rate-control): adaptive, bounded and cluster-coordinated HTTP rate controls by @Jeadie in #14143
  • fix(sql): align date_part('dow') with EXTRACT(dow) Sunday=0 by @lukekim in #14796
  • fix(runtime): carry API-key principal into MCP tools/call by @lukekim in #14828
  • perf(cayenne): hold maintained-aggregate retraction state in a compact shared index by @lukekim in #14762
  • fix(deps): keep a hash join's order once it has a dynamic filter, so vector_search plans again by @Jeadie in #14857
  • fix(cayenne): fence schema statistics and control maintenance tests by @bjchambers in #14856
  • fix(github): retry nested GraphQL pages in place and fail incomplete nested connections by @sgrebnov in #14862
  • fix(cayenne): load an append dataset that has retention_sql and orders versions by time by @phillipleblanc in #14878
  • fix(cdc): Cayenne replication needs only a primary key, not on_conflict by @phillipleblanc in #14880
  • feat(connector-huggingface): Hugging Face datasets data connector (hf://datasets/...) by @lukekim in #14877
  • fix(iceberg): Iceberg REST clients read the real table or get an error, never an empty one by @lukekim in #14588

Full Changelog: https://github.com/spiceai/spiceai/compare/v2.3.2...v2.4.0-rc.1

Spice v2.1.4 (Aug 5, 2026)

ยท 2 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Spice v2.1.4 is now available! ๐Ÿ› ๏ธ

Spice v2.1.4 is a patch release that improves DuckDB acceleration and runtime stability: datasets with UTC timestamp columns, such as Iceberg tables, now load successfully when accelerated with DuckDB, retention policies run reliably, and the runtime is more resilient under heavy query load.

What's New in v2.1.4โ€‹

DuckDB Acceleration Works with Iceberg Timestampsโ€‹

A dataset with a UTC timestamp column โ€” for example, any Iceberg timestamptz column โ€” could previously fail to load when accelerated with DuckDB, leaving the dataset unhealthy and unqueryable. These datasets now load and become ready normally, with no configuration changes needed.

Reliable Retention Policies on DuckDBโ€‹

Retention policies now apply cleanly on DuckDB-accelerated datasets, including policies that combine a time window with an additional condition. Expired rows are evicted on every retention interval, keeping accelerated data fresh and storage bounded.

Improved Runtime Stabilityโ€‹

The runtime is now more robust when queries are cancelled โ€” whether by a client disconnecting, a timeout, or a new refresh superseding an in-flight read. A rare crash in this path has been eliminated.

Contributorsโ€‹

Breaking Changesโ€‹

No breaking changes.

Cookbook Updatesโ€‹

No new cookbook recipes.

The Spice Cookbook includes more than 100 recipes to help you get started with Spice quickly and easily.

Upgradingโ€‹

To upgrade to v2.1.4, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:2.1.4 image:

docker pull spiceai/spiceai:2.1.4

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 2.1.4

AWS Marketplace:

Spice is available in the AWS Marketplace.

What's Changedโ€‹

Changelogโ€‹

  • chore(deps): bump datafusion and table-providers for the DuckDB timezone and retention fixes (2.1 backport) by @phillipleblanc in #12546
  • fix(deps): bump the Vortex pin to pick up the task-cancellation fix by @phillipleblanc in #12547

*Full Changelog: https://github.com/spiceai/spiceai/compare/v2.1.3...v2.1.4

Spice v2.1.3 (Aug 4, 2026)

ยท 5 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Spice v2.1.3 is now available! ๐Ÿ› ๏ธ

Spice v2.1.3 is a patch release focused on resource efficiency: explicit CPU sizing with the new runtime.cpu.cores setting, more query memory for Cayenne deployments, and improved memory and crash diagnostics. It also fixes Cayenne acceleration of Iceberg datasets with timestamptz columns and restores WHERE filters on federated LEFT/RIGHT JOIN queries.

What's New in v2.1.3โ€‹

CPU Sizing with runtime.cpu.coresโ€‹

The new runtime.cpu.cores setting controls how many cores the runtime targets. Thread pools, query partitioning, and accelerator concurrency are all derived from it.

runtime:
cpu:
cores: 4 # `auto` (default) detects. Accepts 4, 3.5, 3500m

Also available as --cpu-cores and SPICE_CPU_CORES (precedence: flag > environment > Spicepod).

This is most useful on large, shared nodes. A pod that sets resources.requests.cpu without a CPU limit exposes no cgroup quota, so the runtime sizes itself for every core on the node rather than its allocated share. Setting the entitlement aligns parallelism and memory footprint with the CPU the pod actually receives.

The effective value, its source, and the derived sizing are logged at startup and exported as the spiced_cpu_budget_cores gauge.

More Query Memory for Cayenne Deploymentsโ€‹

The Cayenne compaction memory pool is now reserved only for accelerations that can compact into it: file mode with a small-write refresh profile. Other deployments, including refresh_mode: full, keep the full memory limit available to queries โ€” up to 6.4 GiB on a 32 GiB limit, with no configuration change.

Memory budgets are now derived from the process's own cgroup limit rather than total host memory.

Diagnosticsโ€‹

Three new gauges report memory in use: query_memory_pool_used_bytes, cayenne_compaction_memory_pool_used_bytes, and process_resident_memory_bytes.

Memory pool refusals now return ResourcesExhausted and HTTP 503, distinguishing them from query errors.

Fatal native signals (SIGSEGV, SIGBUS, SIGILL, SIGFPE) report the signal, faulting address, and thread before exit, so a crash can be diagnosed from logs.

Fixed a bug where setting runtime.task_history.enabled: false also disabled every query metric โ€” query_duration_ms, query_execution_duration_ms, query_executions, query_failures, query_returned_rows, and query_returned_bytes. These are now reported regardless of the task history setting.

Cayenne Acceleration of Iceberg timestamptz Columnsโ€‹

Accelerating an Iceberg dataset with a timestamptz column using the Cayenne engine previously failed during the refresh write with an error resolving the time zone +00:00. Iceberg maps every timestamptz column to the fixed-offset Arrow time zone +00:00, which the file writer could not resolve when building column statistics. Fixed-offset time zones (ยฑHH:MM, ยฑHHMM, and ยฑHH) are now resolved wherever time zones are handled, so these datasets accelerate correctly.

Federated Outer Join Filtersโ€‹

A federated query combining a LEFT JOIN with a WHERE filter on the left table previously returned all rows instead of the filtered rows: when the query was pushed down to the data source or accelerator, the filter was folded into the JOIN ON clause, where it no longer filters the left side (RIGHT JOIN was affected symmetrically). Filters now stay on the side of the join they came from, so these queries return the correct rows.

Contributorsโ€‹

Breaking Changesโ€‹

No breaking changes.

Cookbook Updatesโ€‹

No new cookbook recipes.

The Spice Cookbook includes more than 100 recipes to help you get started with Spice quickly and easily.

Upgradingโ€‹

To upgrade to v2.1.3, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:2.1.3 image:

docker pull spiceai/spiceai:2.1.3

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 2.1.3

AWS Marketplace:

Spice is available in the AWS Marketplace.

What's Changedโ€‹

Changelogโ€‹

  • feat(runtime): size every CPU-derived pool from the CPU entitlement by @bjchambers in #12276
  • fix(runtime): carve the Cayenne compaction memory pool only when a dataset can compact into it by @sgrebnov in #12326
  • fix(runtime): size memory budgets from the process's own cgroup limit by @lukekim in #12263
  • fix(telemetry): read the cgroup CPU quota along the whole cgroup path by @sgrebnov in #12300
  • feat(runtime): expose the memory numbers that explain an OOM as gauges by @lukekim in #12195
  • fix(runtime): report a memory-pool refusal as ResourcesExhausted and answer it with 503 by @sgrebnov in #12289
  • fix(cayenne): the write-concurrency raise must respect the memory brake by @lukekim in #12317
  • feat(spiced): report fatal signals before exit by @sgrebnov in #12334
  • fix: report query metrics when task history is disabled by @sgrebnov in #12227
  • chore(deps): repoint vortex at the 2.1 fixed-offset timezone fix by @phillipleblanc in #12455
  • chore(deps): bump datafusion rev for outer-join unparser fix by @Jeadie in #12460

*Full Changelog: https://github.com/spiceai/spiceai/compare/v2.1.2...v2.1.3

Spice v2.1.2 (Jul 28, 2026)

ยท 3 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Spice v2.1.2 is now available! ๐Ÿ› ๏ธ

Spice v2.1.2 is a patch release focused on improving the DuckDB data accelerator. It upgrades the DuckDB engine to v1.5.5 and introduces the on_full_refresh parameter, giving file-mode DuckDB accelerations compact, predictable disk usage across repeated full refreshes.

What's New in v2.1.2โ€‹

Bounded DuckDB Acceleration File Growth with on_full_refreshโ€‹

File-mode DuckDB accelerations using refresh_mode: full now reclaim disk space on every refresh, keeping the database file compact and disk usage predictable for long-running deployments. Each full refresh bulk-loads a fresh copy of the data, and the new on_full_refresh modes ensure the space held by prior copies is returned rather than accumulating in the file.

The new on_full_refresh acceleration parameter controls how disk space is reclaimed after each full refresh:

acceleration:
engine: duckdb
mode: file
refresh_mode: full
params:
duckdb_file: /data/shared.duckdb
on_full_refresh: replace_file # default: reuse_file
  • reuse_file (default): Existing behavior โ€” refresh into the existing database file.
  • replace_file: Each full refresh streams data into a fresh staging database file, carries over every other object sharing the file (other datasets' tables, views, indexes, and Spice metadata), checkpoints it, and atomically replaces the configured file. Queries are never interrupted โ€” in-flight queries drain against the old file while new queries see the new file โ€” and space is fully reclaimed on every refresh.
  • checkpoint_file: After each refresh, run a CHECKPOINT in place, escalating to FORCE CHECKPOINT when concurrent transactions block the plain attempt (waiting for in-flight transactions; never aborting them).

DuckDB 1.5.5โ€‹

The DuckDB engine is upgraded from v1.5.3 to v1.5.5, bringing the latest upstream stability fixes.

Contributorsโ€‹

Breaking Changesโ€‹

No breaking changes.

Cookbook Updatesโ€‹

No new cookbook recipes.

The Spice Cookbook includes more than 100 recipes to help you get started with Spice quickly and easily.

Upgradingโ€‹

To upgrade to v2.1.2, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:2.1.2 image:

docker pull spiceai/spiceai:2.1.2

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 2.1.2

AWS Marketplace:

Spice is available in the AWS Marketplace.

What's Changedโ€‹

Changelogโ€‹

  • feat(duckdb): on_full_refresh: replace_file โ€” full refresh into a new database file, atomically replaced by @lukekim in #12135
  • feat(duckdb): 'on_full_refresh: checkpoint_file' to bound acceleration file growth by @sgrebnov in #12139

*Full Changelog: https://github.com/spiceai/spiceai/compare/v2.1.1...v2.1.2

Spice v1.11.5 (Apr 1, 2026)

ยท 4 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Announcing the release of Spice v1.11.5! ๐Ÿ› ๏ธ

Spice v1.11.5 is a patch release improving on_zero_results: use_source fallback performance, Delta Lake timestamp predicate data skipping, S3 Parquet read performance, PostgreSQL partitioned table support, Cayenne target file size handling, and preparing the CLI for v2.0 runtime upgrades.

What's New in v1.11.5โ€‹

on_zero_results: use_source Fallback Performance Improvementโ€‹

Improved the on_zero_results: use_source fallback path to run DataFusion's physical optimizer on the federated scan plan (#9927). The fallback path now runs SessionState::physical_optimizers() rules on the federated scan plan before execution, enabling parallel file group scanning and other optimizations. This results in significantly faster fallback queries on multi-core machines, particularly for file-based data sources like Delta Lake.

Delta Lake: Improved Data Skipping for >= Timestamp Predicatesโ€‹

Delta Lake table scans with >= timestamp filters now correctly prune files that do not match the predicate (#9932), improving query performance through more effective data skipping (file-level pruning).

PostgreSQL: Partitioned Tables Supportโ€‹

The PostgreSQL data connector now supports partitioned tables (#9997) for both federated and accelerated queries.

S3 Parquet Read Performance Improvementโ€‹

Improved parquet read performance from S3 and other object stores (#10064), particularly for tables with many columns. Column data ranges are now coalesced into fewer, larger requests instead of being fetched individually, reducing the number of HTTP round-trips.

Cayenne: Ensure Target File Size is Respectedโ€‹

The Cayenne accelerator now correctly respects the configured target file size (#10071). Previously, Cayenne could produce many small, fragmented Vortex files; with this fix, files are written at the expected target size, improving storage efficiency and query performance.

CLI: Support for v2.0 Runtime Upgradesโ€‹

The Spice CLI can now upgrade to v2.0 runtime versions. This enables upgrading to v2.0 release candidates and, once released, the v2.0 stable runtime.

spice upgrade v2.0.0-rc.1

Running spice upgrade without a version will upgrade to the latest stable version, including v2.0 once released.

Note: Native Windows runtime builds will no longer be provided in v2.0. Use WSL for local development instead.

Contributorsโ€‹

Breaking Changesโ€‹

No breaking changes.

Cookbook Updatesโ€‹

No new cookbook recipes.

The Spice Cookbook includes 86 recipes to help you get started with Spice quickly and easily.

Upgradingโ€‹

To upgrade to v1.11.5, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:1.11.5 image:

docker pull spiceai/spiceai:1.11.5

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 1.11.5

AWS Marketplace:

Spice is available in the AWS Marketplace.

What's Changedโ€‹

Changelogโ€‹

  • fix(runtime): Run physical optimizer on FallbackOnZeroResultsScanExec fallback plan by @sgrebnov in #9927
  • fix(delta_lake): Fix data skipping for >= timestamp predicates by @sgrebnov in #9932
  • fix(PostgreSQL): Fix schema discovery for PostgreSQL partitioned tables by @sgrebnov in #9997
  • fix(cli): Skip models variant download for v2+ in upgrade/install by @lukekim and @sgrebnov in #10052
  • perf(s3): Improve Parquet read performance by @sgrebnov in #10064
  • fix(cayenne): Ensure Cayenne respects target file size by @krinart in #10071

Full Changelog: https://github.com/spiceai/spiceai/compare/v1.11.4...v1.11.5

Spice v1.11.4 (Mar 12, 2026)

ยท 5 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Announcing the release of Spice v1.11.4! โšก

Spice v1.11.4 is a patch release improving S3 metadata column query robustness and enabling on_zero_results: use_source for accelerated views.

What's New in v1.11.4โ€‹

Accelerated Views: on_zero_results: use_source Supportโ€‹

Accelerated views now support the on_zero_results: use_source configuration (#9699). Previously, accelerated views only supported on_zero_results: return_empty, which returned an empty result set when the accelerated data contained no matching rows. With this change, views can fall back to querying the source data when the accelerated query returns zero results, matching the behavior already available for accelerated datasets.

Example configuration:

views:
- name: sales_summary
sql: |
SELECT region, SUM(amount) as total
FROM sales
GROUP BY region
acceleration:
enabled: true
on_zero_results: use_source

How the Fallback Worksโ€‹

When an accelerated view is configured with on_zero_results: use_source, the following happens at query time:

  1. The accelerated store is queried first. The query runs against the view's accelerated data (e.g., Spice Cayenne, Arrow, DuckDB, or SQLite).

  2. If the accelerated query returns zero rows, the runtime falls back to re-executing the view's SQL query against the datasets it references.

  3. Referenced datasets are queried according to their own configuration. The view's SQL is re-executed against each referenced dataset as it is configured. This means:

    • If a referenced dataset is accelerated, the query hits that dataset's accelerated store โ€” not the raw data source.
    • If a referenced dataset is accelerated with on_zero_results: use_source and its accelerated store also returns zero rows, it will independently fall back to its own federated data source (e.g., Postgres, S3, etc.).
    • If a referenced dataset is federated (not accelerated), the query goes directly to the data source.

This means the fallback can chain through multiple layers: first the view's acceleration, then each referenced dataset's acceleration, and finally the original data source โ€” each layer independently applying its own on_zero_results behavior.

Example: Multi-layer fallback

datasets:
- from: postgres:orders
name: orders
acceleration:
enabled: true
refresh_sql: "SELECT * FROM orders WHERE created_at > now() - interval '7 days'"
on_zero_results: use_source # Falls back to Postgres if accelerated data has no matches

views:
- name: recent_orders_summary
sql: |
SELECT status, COUNT(*) as order_count
FROM orders
GROUP BY status
acceleration:
enabled: true
on_zero_results: use_source # Falls back to re-running the SQL against referenced datasets

In this example, a query like SELECT * FROM recent_orders_summary WHERE status = 'cancelled' follows this path:

  1. Queries recent_orders_summary in the view's accelerated store (DuckDB/SQLite).
  2. If zero rows are returned, re-executes SELECT status, COUNT(*) ... FROM orders GROUP BY status against the orders dataset.
  3. Since orders is accelerated, the query hits the orders accelerated store.
  4. If orders also returns zero rows (e.g., the refresh_sql excluded cancelled orders), it falls back to querying Postgres directly.

S3 Data Connector: More Robust Metadata Column Handlingโ€‹

Improved the robustness of metadata column (location, last_modified, size) handling for S3 datasets. Building on the v1.11.3 release, this update addresses an additional edge case where the query optimizer's projection swap could cause an index out of bounds panic when metadata columns are referenced in projections with filters or scalar functions.

Contributorsโ€‹

Breaking Changesโ€‹

No breaking changes.

Cookbook Updatesโ€‹

No new cookbook recipes.

The Spice Cookbook includes 86 recipes to help you get started with Spice quickly and easily.

Upgradingโ€‹

To upgrade to v1.11.4, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:1.11.4 image:

docker pull spiceai/spiceai:1.11.4

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 1.11.4

AWS Marketplace:

Spice is available in the AWS Marketplace.

What's Changedโ€‹

Changelogโ€‹

  • fix(s3): Make metadata column handling more robust by @sgrebnov in #9714
  • feat(views): Enable on_zero_results: use_source for accelerated views by @krinart in #9699

Full Changelog: https://github.com/spiceai/spiceai/compare/v1.11.3...v1.11.4

Spice v2.0-rc.1 (Mar 4, 2026)

ยท 23 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Announcing the release of Spice v2.0-rc.1! ๐Ÿš€

v2.0.0-rc.1 is the first release candidate for early testing of v2.0.

Highlights in this release candidate include:

  • Active-Active Highly-Available Distributed Query that is object-store-native and built on Apache Ballista, with dynamic cluster sizing, distributed ingestion, and cluster observability
  • Spice Cayenne RC with staged append writes, file-based retention deletes, composite partitioning, and distributed ingestion
  • DataFusion v52.2.0 Upgrade with sort pushdown, a new merge join, and dynamic filters
  • DDL Support for CREATE TABLE and DROP TABLE via SQL for Iceberg and Cayenne catalogs
  • DuckLake Catalog & Data Connector for lakehouse-style data management
  • GCS Data Connector (Alpha) for Google Cloud Storage
  • Rust CLI Rewrite for a unified single-binary experience
  • Dependency upgrades including DuckDB v1.4.4, delta_kernel v0.18.2, and mistral.rs

Spice v2.0 includes several breaking changes. Review the breaking changes section before upgrading.

Distribution Changesโ€‹

AI/ML support including local LLM/ML model and hosted LLM inference is now included in the default Spice build and image. The separate models build variant has been removed.

With models now included by default, the data-only distribution (without AI/ML support) is only published in nightly builds. Official production-ready data-only distributions are available exclusively through Spice Cloud and the Enterprise release.

A new Network Attached Storage (NAS) distribution with built-in SMB and NFS data connector support is also now available in nightly builds and with Spice.ai Enterprise.

Distribution / VariantOpen SourceSpice CloudEnterprise
Defaultโœ…โœ…โœ…
DataNightly onlyโœ…โœ…
NAS (SMB + NFS)Nightly onlyโŒโœ…
Metal (macOS)โœ…โœ…โœ…
CUDA (Linux)Nightly onlyโœ…โœ…
Allocator variantsNightly onlyโœ…โœ…
ODBC connectorLocal build onlyโœ…โœ…

For more details, see the Distributions documentation.

What's New in v2.0.0-rc.1โ€‹

Active-Active HA Distributed Queryโ€‹

Distributed Query exits Beta with active-active highly-available object-store-based distributed query.

Distributed query supports two execution modes:

  • Synchronous: Queries for accelerated datasets are distributed across executors and results are streamed back in real-time. Non-accelerated datasets execute only on the scheduler. Best for interactive queries where low latency is critical.
  • Asynchronous: Queries are submitted via the new HTTP-only /v1/queries API and results are materialized to object storage for later retrieval. Best for long-running analytical workloads, batch processing, and non-accelerated datasets in distributed mode.

Key improvements:

  • Dynamic Cluster Sizing: The query planner automatically adjusts parallelism based on the number of active executors in the cluster, ensuring optimal resource utilization as nodes are added or removed.
  • Distributed Ingestion: Data ingestion for partitioned accelerated tables is now distributed across executor nodes, enabling higher throughput and parallel data loading in cluster mode. Regular (non-partitioned) accelerated tables do not distribute ingestion loads.
  • Synchronous Execution on Scheduler: /v1/sql and FlightSQL queries now execute synchronously on the scheduler when appropriate, reducing inter-node overhead for queries that don't benefit from distribution.
  • Faster Failure Detection: Executor heartbeat timeout reduced from 180s to 30s, enabling the cluster to quickly detect and respond to executor failures.
  • Cluster Observability: New metrics and Grafana dashboard for monitoring distributed query clusters.

Spice Cayenne Improvementsโ€‹

The Spice Cayenne data accelerator exits Beta with significant reliability and performance improvements:

  • Staged Append Writes: WAL-based staged append writes prevent partial writes and data loss on stream errors. Batches are written to a WAL file before being committed, ensuring atomicity.
  • File-Based Retention Deletes: Time-based retention now supports file-level deletes for both position-based and primary-key tables, reducing I/O overhead compared to row-level deletion.
  • Multiple Partition Expressions: Support for composite partitioning with partition_by: [col1, col2] using hierarchical path-like keys (e.g., 2025/10/15).
  • Distributed Ingestion: Cayenne catalog now supports distributed ingestion across executor nodes in cluster mode, including UPDATE operations.
  • Improved Robustness: Fixed CDC edge case where DELETE + UPSERT sequences could produce duplicate primary keys across protected snapshots. Improved upsert handling during runtime restarts.

DataFusion v52.2.0 Upgradeโ€‹

Apache DataFusion has been upgraded to v52.2.0, bringing significant performance improvements, new query features, and enhanced extensibility.

Performance Improvements:

  • Faster CASE Expressions: Lookup-table-based evaluation for certain CASE expressions avoids repeated evaluation, accelerating common ETL patterns
  • MIN/MAX Aggregate Dynamic Filters: Queries with MIN/MAX aggregates now create dynamic filters during scan to prune files and rows as tighter bounds are discovered during execution
  • New Merge Join: Rewritten sort-merge join (SMJ) operator with speedups of three orders of magnitude in pathological cases (e.g., TPC-H Q21: minutes โ†’ milliseconds)
  • Caching Improvements: New statistics cache for file metadata avoids repeatedly recalculating statistics, significantly improving planning time. A prefix-aware list-files cache accelerates evaluating partition predicates for Hive partitioned tables
  • Improved Hash Join Filter Pushdown: Build-side hash map contents are now passed dynamically to probe-side scans for pruning files, row groups, and individual rows

Major Features:

  • Sort Pushdown to Scans: Sorts are pushed into data sources, enabling ~30x performance improvement on pre-sorted data with top-K queries. Parquet scans now reverse row group order for DESC queries on ASC-sorted files
  • TableProvider supports DELETE and UPDATE: New hooks for DELETE and UPDATE statements in the TableProvider trait, enabling Iceberg and Cayenne connectors to implement SQL DELETE and UPDATE operations
  • More Extensible SQL Planning: New RelationPlanner API for extending SQL planning for FROM clauses, enabling support for vendor-specific SQL dialects

DDL Support for Iceberg and Cayenneโ€‹

SQL Schema Management: Spice now supports CREATE TABLE and DROP TABLE DDL operations for Iceberg and Cayenne catalogs via FlightSQL and the /v1/sql API. DML validation has been updated for catalog-level writability.

DuckLake Catalog & Data Connectorโ€‹

Lakehouse-Style Data Management: New DuckLake catalog and data connector enable lakehouse-style data management with DuckDB as the metadata catalog and object storage for data files. DuckLake provides ACID transactions, time travel, and schema evolution on top of Parquet files.

GCS Data Connector (Alpha)โ€‹

Google Cloud Storage Support: New Google Cloud Storage data connector enables federated queries against data stored in GCS buckets, with Iceberg table support.

Rust CLI Rewriteโ€‹

Unified Single-Binary Experience: The Spice CLI has been completely rewritten from Go to Rust, eliminating the Go dependency and providing a single spice binary built from the same codebase as spiced. This improves startup performance, reduces distribution size, and ensures consistent behavior between CLI and runtime.

Key Features:

  • Full Feature Parity: All 27+ CLI commands re-implemented in Rust with identical behavior
  • New spice query Command: Interactive REPL for async queries via the /v1/queries API with multi-line SQL input, spinner progress indicator, Ctrl+C cancellation, and partial query ID matching
  • --output=json Flag: Machine-readable JSON output for CLI commands, enabling scripting and automation
  • spice login --output: New output modes (env, json, keychain) for flexible credential management
  • spice cloud metrics: New command for Spice Cloud deployment metrics

Models Included by Defaultโ€‹

Local LLM/ML model inference (via mistral.rs) is now included in the default Spice build. The separate models build variant has been removed. This simplifies installation and ensures all users have access to local AI inference capabilities.

Error Propagation for Dataset and Model Status APIsโ€‹

The /v1/datasets and /v1/models APIs now return structured error information when a component is in an Error state. The ?status=true query parameter must be passed to retrieve the real-time component status, including the error state and details. Previously, the status field only indicated Error with no further detail. Now, two new fields are included when ?status=true is specified:

  • error: A structured object with category, type, and code fields for programmatic error handling (e.g. { "category": "dataset", "type": "auth", "code": "dataset.auth" }).
  • error_message: A human-readable description of why the component entered an error state.

These fields are only present when ?status=true is passed and the component is in an error state.

Example /v1/datasets?status=true response:

[
{
"from": "postgres:syncs",
"name": "daily_journal",
"replication_enabled": false,
"acceleration_enabled": true,
"status": "Ready"
},
{
"from": "databricks:hive_metastore.default.messages",
"name": "messages",
"replication_enabled": false,
"acceleration_enabled": true,
"status": "Error",
"error": {
"category": "dataset",
"type": "auth",
"code": "dataset.auth"
},
"error_message": "Unable to authenticate with datasource credentials"
}
]

The spice datasets and spice models CLI commands now include an ERROR column that displays the error message for any component in an error state.

Additional Dependency Upgradesโ€‹

DependencyVersion
Ballistav52.0.0
DuckDBv1.4.4
delta_kernelv0.18.2
mistral.rsv0.7.0 (candle fork removed, now uses candle 0.9.2 from crates.io)
Turso (libsql)v0.4.4
VortexUpgraded with CASE-WHEN support
AWS SDKMultiple crates updated + APN user-agent support

Other Improvementsโ€‹

  • Spicepod v2 Support: Spicepods now support version v2, and spice init generates spicepod.yaml files with version: v2 by default while maintaining backward compatibility for existing v1 spicepods.
  • x.ai Models: x.ai models now exclusively use the /v1/responses endpoint with rate limiting support.
  • HuggingFace Chat Templates: Added support for chat templates in HuggingFace model configurations.
  • Databricks SQL Dialect: Added Databricks SQL dialect for DataFusion unparser, improving federation query generation.
  • Snowflake: Added snowflake_private_key parameter for key-pair authentication.
  • Acceleration Metrics: New rows_written, bytes_written, and dataset_acceleration_size_bytes metrics for acceleration refresh ingestion.
  • Refresh SQL UDFs: Core scalar UDFs are now enabled in refresh SQL expressions.
  • FlightSQL: Fixed TLS connection handling for grpc+tls:// endpoints with custom CA certificate support.
  • FlightSQL: Fixed schema consistency by expanding view types and verifying field names.
  • Hash Index: Fixed query correctness when hash index is used with additional filters.
  • Results Cache: Fixed schema preservation for empty query results.
  • Query Nullability: Reconciled execution stream nullability with logical plan schema.
  • Schema Evolution: Graceful handling of schema evolution mismatch errors during data refresh.
  • Internal YAML Parser: Replaced deprecated serde_yaml with an internal YAML implementation.

Spicepod v1 to v2 Changesโ€‹

Spicepod v2 introduces configuration improvements while maintaining backward compatibility with v1. Existing v1 spicepods continue to work โ€” deprecated fields are automatically migrated at load time.

Version support:

VersionStatus
v2Default. Used by spice init.
v1Supported. Deprecated fields auto-migrate.
v1beta1Removed. No longer accepted.

Configuration changes:

v1 (deprecated)v2 (preferred)Notes
runtime.results_cacheruntime.caching.sql_resultsAll fields migrate automatically. cache_max_size โ†’ max_size.
runtime.memory_limitruntime.query.memory_limitAuto-migrated. query.memory_limit takes priority if both set.
runtime.temp_directoryruntime.query.temp_directoryAuto-migrated. query.temp_directory takes priority if both set.
dataset.invalid_type_actiondataset.unsupported_type_actionAuto-migrated. v2 adds a new string variant.

New v2 fields:

  • runtime.ready_state โ€” Controls when the runtime reports ready (on_load default, or on_registration).
  • runtime.flight.do_put_rate_limit_enabled โ€” Enable/disable FlightSQL DoPut rate limiting (default: true).
  • runtime.query.spill_compression โ€” Compression for query spill files (e.g., lz4_frame).
  • runtime.scheduler.partition_management โ€” Configure partition assignment interval, limits, and timeouts for distributed mode.
  • runtime.caching.sql_results.stale_while_revalidate_ttl โ€” Serve stale cached results while revalidating in the background.
  • runtime.caching.sql_results.encoding โ€” Cache entry compression (e.g., zstd).
  • catalog.access: read_write_create โ€” New access mode for catalogs that support DDL operations.

Migration note: When both the deprecated v1 field and its v2 equivalent are set, the v2 field takes priority.

Contributorsโ€‹

Breaking Changesโ€‹

  • Cayenne and Distributed Query exit Beta: Beta warnings have been removed from documentation and code. Both features are now considered GA-ready.
  • Models included by default: The separate models build variant has been removed. Local LLM inference is now always included.
  • Spicepod version defaults to v2: New spicepods created with spice init now default to version: v2. Existing v1 spicepods remain supported, and v1beta1 is no longer accepted.
  • Windows native builds removed: Native Windows builds are no longer provided. Use WSL for local development instead.
  • Metric renames: accelerated_refresh metrics renamed to acceleration_refresh for consistency. last_refresh_time gauge renamed to include milliseconds unit.
  • Caching config renamed: ResultsCache replaced with SQLResultsCacheConfig in configuration.
  • DuckDB parameter rename: partitioned_write_flush_threshold renamed to partitioned_write_flush_threshold_rows.
  • v1/search API: The /v1/search API now always returns an array in matches, even for single results.
  • x.ai model endpoint: x.ai models now exclusively use the /v1/responses endpoint.
  • Error messages: Error messages across S3 Vectors, ScyllaDB, Snowflake, ClickHouse, and other components have been refactored for clarity and consistency.

Cookbook Updatesโ€‹

New and updated Spice Cookbook recipes:

  • Async Queries: Submit long-running queries asynchronously and retrieve results later.
  • DuckLake Catalog Connector: Use DuckLake for lakehouse-style data management with ACID transactions and time travel.

The Spice Cookbook includes 88 recipes to help you get started with Spice quickly and easily.

Upgradingโ€‹

To upgrade to v2.0.0-rc.1, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:2.0.0-rc.1 image:

docker pull spiceai/spiceai:2.0.0-rc.1

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai --version 2.0.0-rc.1

AWS Marketplace:

Spice is available in the AWS Marketplace.

What's Changedโ€‹

Changelogโ€‹

  • Add TPC-DS integration tests with S3 source and PostgreSQL acceleration by @phillipleblanc in #9006
  • fix(tests): fix flaky/slow/failing unit tests by @phillipleblanc in #9009
  • fix: Update benchmark snapshots for DF51 upgrade by @app/github-actions in #9008
  • fix: add feature gate to rrf TEST_EMBEDDING_MODEL by @phillipleblanc in #9017
  • fix: features check by @phillipleblanc in #9014
  • fix: Enable Cayenne acceleration snapshots by @lukekim in #9020
  • URL table support by @lukekim in #9018
  • ScyllaDB key filter by @lukekim in #8997
  • fix: Schema mismatch when using column projection with HTTP caching by @phillipleblanc in #9021
  • Add more tests for HTTP caching with columns selection by @sgrebnov in #9025
  • HTTP cache snapshots: default to time_interval and fix snapshots_creation_policy: on_change by @sgrebnov in #9026
  • Fix duplicate snapshot creation on startup by @sgrebnov in #9029
  • Add ScyllaDB and SMB to the README table by @krinart in #9034
  • Remove waiting for runtime to be ready before creating snapshot by @krinart in #9033
  • Fix snapshot on_change policy to skip when no writes occurred by @sgrebnov in #9028
  • Release notes for release release/1.11.0-rc.2 by @krinart in #9016
  • ci: use arduino/setup-protoc for official protobuf compiler by @phillipleblanc in #9036
  • ci: install unzip on aarch64 runner for arduino/setup-protoc by @phillipleblanc in #9038
  • fix: don't fail release if upload to minio fails by @phillipleblanc in #9039
  • Add missing protoc step to setup-cc action by @krinart in #9041
  • fix: Update Search integration test snapshots by @app/github-actions in #9013
  • Fix formula_1 and codebase_community in bird-bench by @Jeadie in #9000
  • Cayenne S3 Express One Zone improvements by @lukekim in #9015
  • Add zlib1g-dev to CI by @lukekim in #9052
  • Improve validation and logging for hash indexes by @lukekim in #9047
  • Upgrade Vortex with CASE-WHEN by @lukekim in #9051
  • x.ai models now exclusively use /v1/responses endpoint by @lukekim in #9400
  • Improvements for snapshot schema comparison by @krinart in #9401
  • v2.0 breaking changes by @lukekim in #9233
  • Create PartitionManagementTask for scheduler to update accelerated table partition assignments by @Jeadie in #9378
  • refactor(Cayenne): route all write orchestration through CayenneDataSink by @sgrebnov in #9402
  • Refactor benchmark to use QueryExecutor trait by @Jeadie in #9418
  • feat: Add spidapter build and release workflow by @peasee in #9427
  • Testoperator: add support for api-key when connecting to external spice instance by @sgrebnov in #9421
  • Initial implementation of Ducklake catalog & data connectors by @lukekim in #9083
  • Require aws_lc_rs since jsonwebtoken upgrade by @Jeadie in #9426
  • feat: Add spidapter tool by @peasee in #9425
  • Add release notes for 1.11.2 patch release by @sgrebnov in #9430
  • feat(spidapter): integrate system-adapter-protocol with SCP provisioning by @phillipleblanc in #9434
  • Add DuckLake TPCH E2E workflow and federated Spicepod configuration by @lukekim in #9431
  • fix(spidapter): use Flight handshake auth instead of x-api-key header by @phillipleblanc in #9435
  • [spidapter] Keep only what sparks joy by @Jeadie in #9439
  • Refactor binary operator balancing by @Jeadie in #9424
  • feat: Add Iceberg DDL support (CREATE TABLE / DROP TABLE) for default catalog override by @phillipleblanc in #9440
  • Fix Flight SQL schema consistency: expand view types and verify field names by @sgrebnov in #9438
  • Update spidapter for new system-adapter-protocol by @sgrebnov in #9442
  • docs: fix typos and syntax errors in style guide and error handling docs by @cluster2600 in #9445
  • Add acceleration refresh ingestion metrics (rows_written, bytes_written) by @phillipleblanc in #9461
  • Refactor(Cayenne): Replace CatalogError and string based errors with Snafu errors by @sgrebnov in #9403
  • Replace deprecated claude-3-5-haiku-latest with claude-haiku-4-5 by @Jeadie in #9492
  • Fix #9481: Preserve schema in results cache for empty query results by @phillipleblanc in #9485
  • Fix partition by serializing by @Jeadie in #9474
  • query: reconcile execution stream nullability with logical plan schema by @phillipleblanc in #9486
  • initial spice-cloud-client crate and spice cloud metrics --app <app-name>. by @Jeadie in #9480
  • feat: Return dataset error message in datasets API by @peasee in #9487
  • Spicebench by @lukekim in #9447
  • build(deps): consolidate dependabot dependency updates by @phillipleblanc in #9504
  • fix(cluster): route non-partitioned accelerated tables in distributed mode by @phillipleblanc in #9508
  • Enable core scalar UDFs in refresh SQL by @sgrebnov in #9502
  • Fix metrics in Spidapter again by @Jeadie in #9497
  • fix(cluster): tolerate Completed->status propagation race in distributed query handle by @phillipleblanc in #9510
  • feat: Support distributed ingestion in cayenne catalog by @peasee in #9506
  • Fix Cayenne duplicate primary keys after DELETE + UPSERT CDC sequences by @krinart in #9494
  • fix(cluster): rewrite table scans inside subqueries for distributed execution by @phillipleblanc in #9518
  • fix: Set catalog mode to readwritecreate in spidapter by @peasee in #9519
  • Upgrade AWS SDK crates & set APN user-agent in AWS SDK credential bridge by @lukekim in #8328
  • feat(runtime): add runtime ready_state on_registration semantics by @lukekim in #9522
  • fix: Add spidapter post-setup retries by @peasee in #9526
  • Make partition discovery more robust and make initialization non-blocking by @sgrebnov in #9499
  • Make lint-rust-fix support targeted packages and features by @Jeadie in #9511
  • Handle new Cloud SCP API by @Jeadie in #9532
  • Refactor and simplify streaming benchmarks by @krinart in #9405
  • fix: ensure spidapter only increments attempts on failures by @peasee in #9534
  • feat: Support specifying app resources in spidapter by @peasee in #9536
  • test(runtime): Spice Cayenne DDL integration test by @lukekim in #9535
  • fix: Handle schema evolution mismatch errors during data refresh by @lukekim in #9527
  • fix: resolve clippy lint warnings by @phillipleblanc in #9547
  • pr-builds --tag <TAG> for build_and_release.yml by @Jeadie in #9507
  • Add --output flag to spice login with env/json/keychain modes by @Jeadie in #9541
  • Don't use 'PartitionedTableScanRewrite' in async distributed query by @Jeadie in #9548
  • feat(spidapter): add local backend mode with single executor by @phillipleblanc in #9531
  • support chat template in HF by @Jeadie in #9543
  • fix(cayenne): stream PK retention deletes and run OOM regression in CI by @phillipleblanc in #9533
  • cayenne: Staged append writes to prevent partial writes and data loss on stream error by @sgrebnov in #9491
  • AcceleratedTable::scan use FederatedTable::scan when ClusterRole::Scheduler by @Jeadie in #9550
  • Upgrade to delta-kernel-rs v0.18.2 by @lukekim in #9528
  • Run cayenne tests as part of PR CI by @sgrebnov in #9554
  • Upgrade to DataFusion v52.2.0 by @lukekim in #9419
  • Remove Snapshot Compaction + Add snapshot existence check by @krinart in #9523
  • Update dependencies by @lukekim in #9566
  • fix: Update benchmark snapshots by @app/github-actions in #9565
  • fix: Compare Cayenne table configuration on startup by @peasee in #9529
  • Make Refresh::refresh_sql more robust to alterations over time. by @Jeadie in #9549
  • fix: Update datafusion-table-providers dependency to latest revision by @lukekim in #9574
  • Unset AWS_ENDPOINT_URL when empty by @krinart in #9575
  • fix: allow BytesProcessedExec repartitioning for unordered input by @lukekim in #9540
  • Sanitize DataFusion errors by @lukekim in #9530
  • Add conditional logging for partition assignments by @Jeadie in #9577
  • use 'properly early exit on SIGTERM' by @Jeadie in #9573
  • Update datafusion to 52.2.0 by @phillipleblanc in #9582
  • Ensure we query one and only one partition per request by @Jeadie in #9416
  • feat: Add support for Spicepod version v2 by @lukekim in #9583
  • [SpiceDQ] Improve error messages; Avoid race condition on allocate_initial_partitions. by @Jeadie in #9579
  • Update ballista dependencies to latest 52.0.0 revision by @lukekim in #9581
  • Fix Databricks spark_connect mode always disabled by @phillipleblanc in #9586
  • Support partitioning in Arrow accelerator by @Jeadie in #9571
  • Fix spice query CLI response deserialization by @phillipleblanc in #9588
  • fix: Update benchmark snapshots by @app/github-actions in #9584
  • fix: Share RuntimeEnv across Cayenne read/write/delete paths for targeted list_files_cache invalidation by @sgrebnov in #9589
  • feat: Add file:// state_location support for async queries scheduler by @phillipleblanc in #9590
  • Update endgame links by @krinart in #9598

Full Changelog: https://github.com/spiceai/spiceai/compare/v1.11.2...v2.0.0-rc.1

Spice v1.10.2 (Dec 22, 2025)

ยท 5 min read
Sergei Grebnov
Member of Technical Staff at Spice AI

Announcing the release of Spice v1.10.2! ๐Ÿ”ฅ

v1.10.2 introduces Tiered Caching Acceleration with Localpod for multi-layer acceleration architectures, Periodic Acceleration Snapshots with configurable intervals, DynamoDB JSON Nesting for column consolidation, and Kafka/Debezium Batching for faster data ingestion. This release also includes fixes for SQLite accelerator decimal/date handling and real-time status reporting for the /v1/datasets and /v1/models API endpoints.

What's New in v1.10.2โ€‹

Tiered Caching with Localpodโ€‹

Multi-Layer Acceleration Architecture: The Localpod connector now supports caching refresh mode, enabling tiered acceleration where a persistent cache (e.g., file-mode DuckDB) feeds a fast in-memory cache (e.g., Arrow, memory-mode DuckDB).

Key Features:

  • Automatic Cache Propagation: New cache entries automatically propagate from parent to child accelerators
  • Warm Startup: Child accelerators initialize from existing parent data on startup, eliminating cold-start latency
  • Flexible Tiering: Combine any accelerator engines (DuckDB, SQLite, Cayenne) across tiers

Example spicepod.yaml configuration:

datasets:
# Parent: persistent file-mode cache
- from: https://api.example.com
name: api_cache
acceleration:
enabled: true
refresh_mode: caching
engine: duckdb
mode: file

# Child: fast in-memory cache fed by parent
- from: localpod:api_cache
name: api_cache_memory
acceleration:
enabled: true
refresh_mode: caching
engine: arrow
mode: memory

For more details, refer to the Localpod Data Connector Documentation.

Periodic Acceleration Snapshotsโ€‹

Configurable Snapshot Intervals: A new snapshots_create_interval parameter enables periodic snapshot creation for accelerated datasets across all refresh modes. This provides better control over snapshot frequency and ensures consistent recovery points for accelerated data.

Example spicepod.yaml configuration:

datasets:
- from: s3://my-bucket/data.parquet
name: my_data
acceleration:
enabled: true
engine: duckdb
mode: file
refresh_mode: caching
snapshots: enabled
params:
snapshots_create_interval: 60s # Write a snapshot every 60 seconds

For more details, refer to the Data Acceleration Documentation.

DynamoDB JSON Nestingโ€‹

Consolidate Columns into JSON: The DynamoDB Data Connector now supports consolidating columns into a single JSON column using the json_object: "*" metadata option. This is useful when only a few columns are needed as discrete fields while the rest can be accessed as nested JSON.

Example spicepod.yaml configuration:

datasets:
- from: dynamodb:my_table
name: my_table
columns:
- name: PK
- name: SK
- name: data_json
metadata:
json_object: '*' # Captures all other columns as JSON

Example Output: Given a DynamoDB table with columns PK, SK, name, email, and status, the resulting table schema consolidates all non-specified columns into the data_json column:

PKSKdata_json
pk_1sort_1{"name": "Alice", "email": "[email protected]", "status": "active"}
pk_2sort_2{"name": "Bob", "email": "[email protected]", "status": "inactive"}

For more details, refer to the DynamoDB JSON Nesting Documentation.

Kafka/Debezium Batchingโ€‹

Faster Data Ingestion: Configure message batching for Kafka and Debezium connectors to improve data ingestion throughput. Batching reduces processing overhead by grouping multiple messages together before insertion.

Key Features:

  • Configurable Batch Size: Control the maximum number of records per batch (default: 10,000)
  • Configurable Batch Duration: Set the maximum wait time before flushing a partial batch (default: 1s)

Example spicepod.yaml configuration:

datasets:
- from: debezium:kafka-server.public.my_table
name: my_table
params:
batch_max_size: 10000 # Max records per batch (default: 10000)
batch_max_duration: 1s # Max wait time per batch (default: 1s)

For more details, refer to the Kafka Data Connector Documentation and Debezium Data Connector Documentation.

Additional Improvements & Bug Fixesโ€‹

  • Reliability: Fixed SQLite accelerator decimal and date type handling for improved data type accuracy.
  • Reliability: Fixed real-time status reporting for /v1/datasets and /v1/models API endpoints.
  • Reliability: Fixed Kafka warning when security.protocol is set to PLAINTEXT.

Contributorsโ€‹

Breaking Changesโ€‹

No breaking changes.

Cookbook Updatesโ€‹

New Cayenne Data Accelerator Recipe: New recipe demonstrating how to accelerate a local copy of the taxi trips dataset using Cayenne as the data accelerator engine. See Cayenne Data Accelerator Recipe for details.

New Dataset Partitioning Recipe: New recipe demonstrating how to partition accelerated datasets to improve query performance. See Dataset Partitioning for details.

The Spice Cookbook includes 84 recipes to help you get started with Spice quickly and easily.

Upgradingโ€‹

To upgrade to v1.10.2, use one of the following methods:

CLI:

spice upgrade

Homebrew:

brew upgrade spiceai/spiceai/spice

Docker:

Pull the spiceai/spiceai:1.10.2 image:

docker pull spiceai/spiceai:1.10.2

For available tags, see DockerHub.

Helm:

helm repo update
helm upgrade spiceai spiceai/spiceai

AWS Marketplace:

๐ŸŽ‰ Spice is now available in the AWS Marketplace!

What's Changedโ€‹

Changelogโ€‹