From 68a6f2f22015931fcadee13c14d90ed1e89081f3 Mon Sep 17 00:00:00 2001 From: Andy Grove Date: Fri, 17 Jul 2026 13:29:11 -0600 Subject: [PATCH 1/6] Add Comet 1.0.0 release announcement blog post --- .../blog/2026-07-31-datafusion-comet-1.0.0.md | 190 ++++++++++++++++++ 1 file changed, 190 insertions(+) create mode 100644 content/blog/2026-07-31-datafusion-comet-1.0.0.md diff --git a/content/blog/2026-07-31-datafusion-comet-1.0.0.md b/content/blog/2026-07-31-datafusion-comet-1.0.0.md new file mode 100644 index 00000000..c7ff36c2 --- /dev/null +++ b/content/blog/2026-07-31-datafusion-comet-1.0.0.md @@ -0,0 +1,190 @@ +--- +layout: post +title: Apache DataFusion Comet 1.0.0 Release +date: 2026-07-31 +author: pmc +categories: [subprojects] +--- + + + +[TOC] + +The Apache DataFusion PMC is pleased to announce version 1.0.0 of the [Comet](https://datafusion.apache.org/comet/) subproject. + +Comet is an accelerator for Apache Spark that translates Spark physical plans to DataFusion physical plans for +improved performance and efficiency without requiring any code changes. + +This is a major milestone. Comet began as a code donation in early 2024 and has shipped steadily ever since, +one release at a time. Version 1.0.0 marks the point where the project is mature enough to commit to a stable +release line: broad Apache Spark coverage, ANSI SQL semantics, native Parquet and Iceberg scans, and a native +shuffle, all validated continuously against Spark's own test suites. This release covers roughly six weeks of +development since 0.17.0 and is the result of merging over 140 PRs from 19 contributors. See the +[change log] for more information. + +[change log]: https://github.com/apache/datafusion-comet/blob/main/docs/source/changelog/1.0.0.md + +## What 1.0 Means + +Reaching 1.0 is less about any single new feature than about the accumulated maturity of the project: + +- **Broad Spark coverage.** Comet supports Apache Spark 3.4.3, 3.5.8, 4.0.2, and 4.1.1 out of the same + codebase, with dedicated Maven profiles, shim sources, and CI matrices for each. +- **ANSI SQL by default.** Spark 4 enables ANSI semantics by default. Comet implements ANSI behavior for the + expressions it supports natively, including arithmetic overflow checks, ANSI cast behavior, and `try_*` + variants, so ANSI workloads keep accelerating rather than falling back. +- **A correctness harness, not just unit tests.** Comet runs the full Apache Spark SQL test suite through its + native execution path against every supported Spark version. Running Spark's own correctness tests is what + surfaces semantic shifts before they reach user workloads, and it is the foundation of the confidence behind + a 1.0 release. +- **A stable release line going forward.** With 1.0, the project commits to semantic versioning. The very + first deprecations under that policy are announced in this release (see [Deprecation Notice](#deprecation-notice) + below). + +The rest of this post covers what is new since 0.17.0. + +## Native Expression Performance + +A large share of this release is dedicated to making Comet's native scalar expressions faster. These kernels +run per-row or per-batch, so improvements here compound across every query that uses them: + +- **`get_json_object`** ([#4907](https://github.com/apache/datafusion-comet/pull/4907)): roughly 4x faster. +- **`regexp_extract`** ([#4894](https://github.com/apache/datafusion-comet/pull/4894)) and **`parse_url`** + ([#4893](https://github.com/apache/datafusion-comet/pull/4893)): optimized regex and URL parsing paths. +- **Casts**: a faster floating-point-to-decimal cast + ([#4940](https://github.com/apache/datafusion-comet/pull/4940)), an optimized integer-to-integer cast + ([#4920](https://github.com/apache/datafusion-comet/pull/4920)), and shared no-overflow fast paths in + `CheckOverflow` ([#4937](https://github.com/apache/datafusion-comet/pull/4937)) and + `DecimalRescaleCheckOverflow` ([#4938](https://github.com/apache/datafusion-comet/pull/4938)). +- **String and array kernels**: `lpad` ([#4919](https://github.com/apache/datafusion-comet/pull/4919)), + `unhex` ([#4876](https://github.com/apache/datafusion-comet/pull/4876)), + `size` ([#4877](https://github.com/apache/datafusion-comet/pull/4877)), + `arrays_overlap` ([#4906](https://github.com/apache/datafusion-comet/pull/4906)), + `escape_string` ([#4902](https://github.com/apache/datafusion-comet/pull/4902)), and the + `try_*` arithmetic kernel ([#4910](https://github.com/apache/datafusion-comet/pull/4910)). + +To make this kind of work repeatable, the release also adds a +[scalar expression optimization guide](https://github.com/apache/datafusion-comet/pull/4933) documenting how +to benchmark a kernel, keep its output bit-identical to Spark, and gate changes on a no-regression check. + +On the scan side, Parquet reads pick up several improvements: full Parquet metadata (including the page index) +is now cached via DataFusion's `CachedParquetFileReaderFactory` +([#4707](https://github.com/apache/datafusion-comet/pull/4707)), identity casts are unwrapped in the schema +adapter so Parquet statistics pruning can engage +([#4730](https://github.com/apache/datafusion-comet/pull/4730)), and filter pushdown configuration has been +revised ([#4722](https://github.com/apache/datafusion-comet/pull/4722)). + +## New Expression and Aggregate Support + +This release expands the set of Spark expressions and aggregates that run natively: + +- **Aggregates**: `approx_percentile` / `percentile_approx` + ([#4801](https://github.com/apache/datafusion-comet/pull/4801)) and exact `percentile` / `median` + ([#4542](https://github.com/apache/datafusion-comet/pull/4542)). +- **Grouping**: `grouping()` and `grouping_id()` indicator functions + ([#4815](https://github.com/apache/datafusion-comet/pull/4815)). +- **Intervals**: interval types with `make_ym_interval` and `make_dt_interval` + ([#4541](https://github.com/apache/datafusion-comet/pull/4541)). +- **String**: `base64` ([#4778](https://github.com/apache/datafusion-comet/pull/4778)) and `split_part` via + `StringSplitSQL` ([#4592](https://github.com/apache/datafusion-comet/pull/4592)). +- **Array / map**: `array_prepend` ([#4716](https://github.com/apache/datafusion-comet/pull/4716)), + `size()` for `MapType` ([#4580](https://github.com/apache/datafusion-comet/pull/4580)), `ElementAt` over + `MapType` ([#4697](https://github.com/apache/datafusion-comet/pull/4697)), and removal of the constraint on + arrays of nested elements ([#4714](https://github.com/apache/datafusion-comet/pull/4714)). +- **Date/time**: native `TimestampNTZ` inputs for `hour` / `minute` / `second` + ([#4753](https://github.com/apache/datafusion-comet/pull/4753)) and + `PreciseTimestampConversion` for native time-window grouping + ([#4784](https://github.com/apache/datafusion-comet/pull/4784)). +- **Windows**: extended native window function support + ([#4209](https://github.com/apache/datafusion-comet/pull/4209)) and Spark 4 decimal window average + ([#4749](https://github.com/apache/datafusion-comet/pull/4749)). + +## Experimental PyArrow UDF Support + +This release adds experimental support for accelerated PyArrow UDFs +([#4234](https://github.com/apache/datafusion-comet/pull/4234)), allowing PyArrow-based user-defined functions +to participate in native execution instead of forcing a fallback to Spark. When the feature is disabled, Comet +now hints at the native PyArrow UDF path in its fallback reasons +([#4892](https://github.com/apache/datafusion-comet/pull/4892)) so users know the option exists. This is an +early-stage feature and we welcome feedback from users experimenting with it. + +## Iceberg + +- **Iceberg 1.11 support** ([#4840](https://github.com/apache/datafusion-comet/pull/4840)): adds support for + Iceberg 1.11, audits the existing Iceberg diffs, bumps the iceberg-rust dependency, and adds a + `run-iceberg-tests` CI trigger. +- **Delete-file correctness** ([#4760](https://github.com/apache/datafusion-comet/pull/4760)): the native scan + now sizes Iceberg delete files correctly, avoiding dropped deletes. +- **Exchange-reuse correctness** ([#4812](https://github.com/apache/datafusion-comet/pull/4812)): fixed a case + where Iceberg native scan exchange reuse with different pushed filters could produce wrong results. + +## Native Parquet I/O and Cloud + +- **gzip Parquet writes** ([#4930](https://github.com/apache/datafusion-comet/pull/4930)): native Parquet + writes now support gzip compression. +- **Azure authentication** ([#4783](https://github.com/apache/datafusion-comet/pull/4783)): the native + Parquet scan now supports Azure authentication, complementing the existing S3 credential support. + +## Execution and Fallback Improvements + +- **Mixed partial/final aggregation** ([#4861](https://github.com/apache/datafusion-comet/pull/4861)): + `sum` and non-decimal `avg` can now run with the partial and final aggregation stages split across Spark and + Comet, keeping more aggregation plans partially native. +- **Stage-based fallback** ([#4519](https://github.com/apache/datafusion-comet/pull/4519)): a more granular + fallback mechanism that operates at the stage level. +- **Codegen dispatch for opt-in serdes** ([#4728](https://github.com/apache/datafusion-comet/pull/4728)): + expressions reported as `Unsupported` can now route through JVM codegen dispatch for opt-in serdes, and + native opt-in expressions surface as compatible-by-default with a `COMET-INFO` plan hint + ([#4721](https://github.com/apache/datafusion-comet/pull/4721)). +- **Runtime cleanup** ([#4734](https://github.com/apache/datafusion-comet/pull/4734)): the Tokio runtime is + now released on driver and executor exit. +- **Contrib scan SPI** ([#4700](https://github.com/apache/datafusion-comet/pull/4700)): a core SPI for + contrib leaf scans (`CometScanWithPlanData`), the first part of splitting the Delta integration into a + contrib module. + +## Deprecation Notice + +With the move to a stable 1.0 release line, Comet begins deprecating older platforms under semantic versioning +([#4857](https://github.com/apache/datafusion-comet/pull/4857)): + +- **JDK 11** is deprecated and scheduled for removal in Comet 1.1.0. +- **Apache Spark 3.4** is deprecated and scheduled for removal in Comet 1.1.0. + +Users on these platforms should plan to move to JDK 17+ and Spark 3.5 or later before upgrading to 1.1.0. + +## Compatibility + +Supported platforms include: + +- **Spark 3.4.3** with Java 11/17 and Scala 2.12/2.13 (deprecated, removal in 1.1.0) +- **Spark 3.5.8** with Java 11/17 and Scala 2.12/2.13 +- **Spark 4.0.2** with Java 17 and Scala 2.13 +- **Spark 4.1.1** with Java 17 and Scala 2.13 + +See the [Spark Version Compatibility] page for known limitations specific to each version. + +[Spark Version Compatibility]: https://datafusion.apache.org/comet/user-guide/latest/compatibility/spark-versions.html + +This release upgrades to **DataFusion 54** and **Arrow 58.3**. + +## Get Started with Comet 1.0.0 + +Ready to try it out? Follow the [Comet 1.0.0 Installation Guide](https://datafusion.apache.org/comet/user-guide/1.0/installation.html) +to get up and running, then point Comet at your existing Spark workloads and see the speedup for yourself. From 0ce09ac53fd4e77c1a49bea8e2aa6f1b7730af76 Mon Sep 17 00:00:00 2001 From: Andy Grove Date: Mon, 27 Jul 2026 14:39:06 -0600 Subject: [PATCH 2/6] Update Comet 1.0.0 blog post with commits merged since draft Incorporates PRs merged to datafusion-comet main between 2026-07-17 and 2026-07-27, updates PR/contributor counts, and bumps DataFusion version to 54.1. Adds a new Shuffle Improvements section. --- .../blog/2026-07-31-datafusion-comet-1.0.0.md | 60 ++++++++++++++++--- 1 file changed, 51 insertions(+), 9 deletions(-) diff --git a/content/blog/2026-07-31-datafusion-comet-1.0.0.md b/content/blog/2026-07-31-datafusion-comet-1.0.0.md index c7ff36c2..81a9f5e6 100644 --- a/content/blog/2026-07-31-datafusion-comet-1.0.0.md +++ b/content/blog/2026-07-31-datafusion-comet-1.0.0.md @@ -36,7 +36,7 @@ This is a major milestone. Comet began as a code donation in early 2024 and has one release at a time. Version 1.0.0 marks the point where the project is mature enough to commit to a stable release line: broad Apache Spark coverage, ANSI SQL semantics, native Parquet and Iceberg scans, and a native shuffle, all validated continuously against Spark's own test suites. This release covers roughly six weeks of -development since 0.17.0 and is the result of merging over 140 PRs from 19 contributors. See the +development since 0.17.0 and is the result of merging over 190 PRs from 22 contributors. See the [change log] for more information. [change log]: https://github.com/apache/datafusion-comet/blob/main/docs/source/changelog/1.0.0.md @@ -70,9 +70,19 @@ run per-row or per-batch, so improvements here compound across every query that ([#4893](https://github.com/apache/datafusion-comet/pull/4893)): optimized regex and URL parsing paths. - **Casts**: a faster floating-point-to-decimal cast ([#4940](https://github.com/apache/datafusion-comet/pull/4940)), an optimized integer-to-integer cast - ([#4920](https://github.com/apache/datafusion-comet/pull/4920)), and shared no-overflow fast paths in + ([#4920](https://github.com/apache/datafusion-comet/pull/4920)), shared no-overflow fast paths in `CheckOverflow` ([#4937](https://github.com/apache/datafusion-comet/pull/4937)) and - `DecimalRescaleCheckOverflow` ([#4938](https://github.com/apache/datafusion-comet/pull/4938)). + `DecimalRescaleCheckOverflow` ([#4938](https://github.com/apache/datafusion-comet/pull/4938)), a ~40% faster + `float64`-to-`utf8` cast ([#4918](https://github.com/apache/datafusion-comet/pull/4918)), an optimized + `decimal128`-to-`utf8` cast ([#4924](https://github.com/apache/datafusion-comet/pull/4924)), string-to-date + parsing up to 2x faster ([#4917](https://github.com/apache/datafusion-comet/pull/4917)), + `parse_string_to_decimal` 30-40% faster ([#4916](https://github.com/apache/datafusion-comet/pull/4916)), and + `cast_binary_to_string` up to 27x faster on binary-format styles + ([#4912](https://github.com/apache/datafusion-comet/pull/4912)). +- **Decimal and date/time kernels**: `date_trunc` more than 2x faster + ([#4915](https://github.com/apache/datafusion-comet/pull/4915)), + `spark_ceil` 3x faster ([#4926](https://github.com/apache/datafusion-comet/pull/4926)), and a vectorized + `spark_unscaled_value` 9x faster ([#4972](https://github.com/apache/datafusion-comet/pull/4972)). - **String and array kernels**: `lpad` ([#4919](https://github.com/apache/datafusion-comet/pull/4919)), `unhex` ([#4876](https://github.com/apache/datafusion-comet/pull/4876)), `size` ([#4877](https://github.com/apache/datafusion-comet/pull/4877)), @@ -96,14 +106,21 @@ revised ([#4722](https://github.com/apache/datafusion-comet/pull/4722)). This release expands the set of Spark expressions and aggregates that run natively: - **Aggregates**: `approx_percentile` / `percentile_approx` - ([#4801](https://github.com/apache/datafusion-comet/pull/4801)) and exact `percentile` / `median` - ([#4542](https://github.com/apache/datafusion-comet/pull/4542)). + ([#4801](https://github.com/apache/datafusion-comet/pull/4801)), exact `percentile` / `median` + ([#4542](https://github.com/apache/datafusion-comet/pull/4542)), + `approx_count_distinct` ([#4819](https://github.com/apache/datafusion-comet/pull/4819)), and native + `collect_list` / `array_agg` ([#4720](https://github.com/apache/datafusion-comet/pull/4720)). - **Grouping**: `grouping()` and `grouping_id()` indicator functions ([#4815](https://github.com/apache/datafusion-comet/pull/4815)). - **Intervals**: interval types with `make_ym_interval` and `make_dt_interval` - ([#4541](https://github.com/apache/datafusion-comet/pull/4541)). -- **String**: `base64` ([#4778](https://github.com/apache/datafusion-comet/pull/4778)) and `split_part` via - `StringSplitSQL` ([#4592](https://github.com/apache/datafusion-comet/pull/4592)). + ([#4541](https://github.com/apache/datafusion-comet/pull/4541)), + `CalendarIntervalType` support ([#4898](https://github.com/apache/datafusion-comet/pull/4898)), and + `multiply_dt_interval` via codegen dispatch + ([#4900](https://github.com/apache/datafusion-comet/pull/4900)). +- **String**: `base64` ([#4778](https://github.com/apache/datafusion-comet/pull/4778)), + `split_part` via `StringSplitSQL` ([#4592](https://github.com/apache/datafusion-comet/pull/4592)), + native `levenshtein` ([#4105](https://github.com/apache/datafusion-comet/pull/4105)), and native + `randstr` compatible with Spark ([#5035](https://github.com/apache/datafusion-comet/pull/5035)). - **Array / map**: `array_prepend` ([#4716](https://github.com/apache/datafusion-comet/pull/4716)), `size()` for `MapType` ([#4580](https://github.com/apache/datafusion-comet/pull/4580)), `ElementAt` over `MapType` ([#4697](https://github.com/apache/datafusion-comet/pull/4697)), and removal of the constraint on @@ -130,10 +147,16 @@ early-stage feature and we welcome feedback from users experimenting with it. - **Iceberg 1.11 support** ([#4840](https://github.com/apache/datafusion-comet/pull/4840)): adds support for Iceberg 1.11, audits the existing Iceberg diffs, bumps the iceberg-rust dependency, and adds a `run-iceberg-tests` CI trigger. +- **Iceberg table format V3** ([#4991](https://github.com/apache/datafusion-comet/pull/4991)): native table + decryption for V3 tables, with fallback to Spark for other V3 features. Follow-up + [#5020](https://github.com/apache/datafusion-comet/pull/5020) applies the same diff changes across other + Iceberg versions. - **Delete-file correctness** ([#4760](https://github.com/apache/datafusion-comet/pull/4760)): the native scan now sizes Iceberg delete files correctly, avoiding dropped deletes. - **Exchange-reuse correctness** ([#4812](https://github.com/apache/datafusion-comet/pull/4812)): fixed a case where Iceberg native scan exchange reuse with different pushed filters could produce wrong results. +- **Native serde dedup** ([#4982](https://github.com/apache/datafusion-comet/pull/4982)): dedupes Iceberg + residuals and delete files in the native scan serde, reducing planning overhead. ## Native Parquet I/O and Cloud @@ -153,12 +176,31 @@ early-stage feature and we welcome feedback from users experimenting with it. expressions reported as `Unsupported` can now route through JVM codegen dispatch for opt-in serdes, and native opt-in expressions surface as compatible-by-default with a `COMET-INFO` plan hint ([#4721](https://github.com/apache/datafusion-comet/pull/4721)). +- **Collated predicate fallback** ([#4948](https://github.com/apache/datafusion-comet/pull/4948)): Comet + now falls back to Spark for predicates whose operands use non-default collations, avoiding subtle + correctness issues. +- **ANSI cast to date** ([#5014](https://github.com/apache/datafusion-comet/pull/5014)): invalid calendar + dates now raise `CAST_INVALID_INPUT` under ANSI mode, matching Spark's behavior. +- **Local scan nullability** ([#4843](https://github.com/apache/datafusion-comet/pull/4843)): local table scan + child nullability is now widened to match the native kernels, fixing a class of nullability mismatches. +- **Config aliases** ([#4979](https://github.com/apache/datafusion-comet/pull/4979)): a `withAlternative` + alias mechanism lets `CometConf` entries carry old names during renames without breaking existing + configurations. - **Runtime cleanup** ([#4734](https://github.com/apache/datafusion-comet/pull/4734)): the Tokio runtime is now released on driver and executor exit. - **Contrib scan SPI** ([#4700](https://github.com/apache/datafusion-comet/pull/4700)): a core SPI for contrib leaf scans (`CometScanWithPlanData`), the first part of splitting the Delta integration into a contrib module. +## Shuffle Improvements + +- **Native shuffle memory cap** ([#4989](https://github.com/apache/datafusion-comet/pull/4989)): a new + `spark.comet.shuffle.maxBufferBytes` config caps native shuffle writer memory to bound worst-case usage. +- **Shuffle IPC schema encoding** ([#5006](https://github.com/apache/datafusion-comet/pull/5006)): the IPC + schema is now encoded once per writer instead of per block, cutting per-batch shuffle overhead. +- **BatchCoalescer bypass** ([#5003](https://github.com/apache/datafusion-comet/pull/5003)): shuffle bypasses + the `BatchCoalescer` for batches that are already appropriately sized. + ## Deprecation Notice With the move to a stable 1.0 release line, Comet begins deprecating older platforms under semantic versioning @@ -182,7 +224,7 @@ See the [Spark Version Compatibility] page for known limitations specific to eac [Spark Version Compatibility]: https://datafusion.apache.org/comet/user-guide/latest/compatibility/spark-versions.html -This release upgrades to **DataFusion 54** and **Arrow 58.3**. +This release upgrades to **DataFusion 54.1** and **Arrow 58.3**. ## Get Started with Comet 1.0.0 From c26ea16843583f11d4f22227db71f5abee3a808d Mon Sep 17 00:00:00 2001 From: Andy Grove Date: Fri, 31 Jul 2026 15:11:49 -0600 Subject: [PATCH 3/6] Update Comet 1.0.0 blog post with commits merged since last update --- .../blog/2026-07-31-datafusion-comet-1.0.0.md | 183 ++++++++++++++++-- 1 file changed, 165 insertions(+), 18 deletions(-) diff --git a/content/blog/2026-07-31-datafusion-comet-1.0.0.md b/content/blog/2026-07-31-datafusion-comet-1.0.0.md index 81a9f5e6..1578f085 100644 --- a/content/blog/2026-07-31-datafusion-comet-1.0.0.md +++ b/content/blog/2026-07-31-datafusion-comet-1.0.0.md @@ -36,7 +36,7 @@ This is a major milestone. Comet began as a code donation in early 2024 and has one release at a time. Version 1.0.0 marks the point where the project is mature enough to commit to a stable release line: broad Apache Spark coverage, ANSI SQL semantics, native Parquet and Iceberg scans, and a native shuffle, all validated continuously against Spark's own test suites. This release covers roughly six weeks of -development since 0.17.0 and is the result of merging over 190 PRs from 22 contributors. See the +development since 0.17.0 and is the result of merging over 220 PRs from 21 contributors. See the [change log] for more information. [change log]: https://github.com/apache/datafusion-comet/blob/main/docs/source/changelog/1.0.0.md @@ -45,8 +45,11 @@ development since 0.17.0 and is the result of merging over 190 PRs from 22 contr Reaching 1.0 is less about any single new feature than about the accumulated maturity of the project: -- **Broad Spark coverage.** Comet supports Apache Spark 3.4.3, 3.5.8, 4.0.2, and 4.1.1 out of the same - codebase, with dedicated Maven profiles, shim sources, and CI matrices for each. +- **Broad Spark coverage.** Comet supports Apache Spark 3.4.3, 3.5.8, 4.0.2, and 4.1.2 out of the same + codebase, with dedicated Maven profiles, shim sources, and CI matrices for each, plus an experimental + Spark 4.2 profile for early evaluation. The + [Spark version adoption and support-lifetime policy](https://github.com/apache/datafusion-comet/pull/4977) + now states how long each Spark minor stays supported. - **ANSI SQL by default.** Spark 4 enables ANSI semantics by default. Comet implements ANSI behavior for the expressions it supports natively, including arithmetic overflow checks, ANSI cast behavior, and `try_*` variants, so ANSI workloads keep accelerating rather than falling back. @@ -54,9 +57,18 @@ Reaching 1.0 is less about any single new feature than about the accumulated mat native execution path against every supported Spark version. Running Spark's own correctness tests is what surfaces semantic shifts before they reach user workloads, and it is the foundation of the confidence behind a 1.0 release. -- **A stable release line going forward.** With 1.0, the project commits to semantic versioning. The very - first deprecations under that policy are announced in this release (see [Deprecation Notice](#deprecation-notice) - below). +- **A stable release line going forward.** With 1.0, the project commits to semantic versioning. The + [versioning policy](https://github.com/apache/datafusion-comet/pull/5056) spells out what that covers: + because Comet is a plugin rather than a library, its configuration is its primary API surface, so + `spark.comet.*` keys, an explicitly enumerated public Java and Scala API, and query results documented as + Compatible are all part of the compatibility surface. Behavior changes in a minor release now require a + `spark.comet.legacy.*` key that restores the previous behavior, and each release records its behavior changes + in a user-facing upgrade guide. Correctness fixes are exempt from being treated as breaking changes. The very + first deprecations under that policy are announced in this release (see + [Deprecation Notice](#deprecation-notice) below). +- **Documented limitations.** Every open correctness issue is now surfaced in the generated compatibility guide + ([#5085](https://github.com/apache/datafusion-comet/pull/5085)), down to the expression level, so you can see + where Comet is known to diverge from Spark before you hit it in production rather than after. The rest of this post covers what is new since 0.17.0. @@ -98,8 +110,71 @@ On the scan side, Parquet reads pick up several improvements: full Parquet metad is now cached via DataFusion's `CachedParquetFileReaderFactory` ([#4707](https://github.com/apache/datafusion-comet/pull/4707)), identity casts are unwrapped in the schema adapter so Parquet statistics pruning can engage -([#4730](https://github.com/apache/datafusion-comet/pull/4730)), and filter pushdown configuration has been -revised ([#4722](https://github.com/apache/datafusion-comet/pull/4722)). +([#4730](https://github.com/apache/datafusion-comet/pull/4730)), filter pushdown configuration has been +revised ([#4722](https://github.com/apache/datafusion-comet/pull/4722)), the native scan now passes a metadata +size hint so a single read usually captures the footer, matching the Iceberg path +([#4717](https://github.com/apache/datafusion-comet/pull/4717)), and the native Parquet scan seeds its reader +options from the session config so Parquet settings you already set take effect +([#5107](https://github.com/apache/datafusion-comet/pull/5107)). + +Two lower-level costs were also removed: Comet no longer makes excessive timer calls in the native execution +loop ([#4739](https://github.com/apache/datafusion-comet/pull/4739)), and plan-data injection is now an O(1) +lookup by operator kind rather than a scan ([#4535](https://github.com/apache/datafusion-comet/pull/4535)). + +## Correctness + +A 1.0 release is only as good as its results. This release fixes a broad set of divergences from Spark, most of +them found by running Spark's own SQL test suites through Comet's native path: + +- **ANSI error semantics**: `round` on a `Long` with a large negative scale now overflows instead of silently + returning zero ([#5082](https://github.com/apache/datafusion-comet/pull/5082)), `Long.MinValue / -1` raises + `ARITHMETIC_OVERFLOW` ([#5084](https://github.com/apache/datafusion-comet/pull/5084)), floating-point + `x % 0.0` raises `REMAINDER_BY_ZERO` instead of returning `NaN` + ([#5081](https://github.com/apache/datafusion-comet/pull/5081)), `make_decimal` honors its + fail-on-error flag and throws `NUMERIC_VALUE_OUT_OF_RANGE` rather than returning null + ([#5080](https://github.com/apache/datafusion-comet/pull/5080)), invalid calendar dates raise + `CAST_INVALID_INPUT` ([#5014](https://github.com/apache/datafusion-comet/pull/5014)), and errors from nested + casts now propagate ([#4675](https://github.com/apache/datafusion-comet/pull/4675)). +- **Wrong results**: `count` no longer returns zero when the native scan is disabled + ([#4795](https://github.com/apache/datafusion-comet/pull/4795)), Spark's legacy `null IN ()` behavior is + honored ([#5127](https://github.com/apache/datafusion-comet/pull/5127)), `pow` matches Java's `Math.pow` on + the edge cases where C and Java disagree ([#5033](https://github.com/apache/datafusion-comet/pull/5033)), + `percentile` matches Spark's interpolation precision + ([#4792](https://github.com/apache/datafusion-comet/pull/4792)), `date_trunc` handles DST boundaries in + non-UTC sessions and no longer produces schema mismatches through shuffle and sort + ([#4761](https://github.com/apache/datafusion-comet/pull/4761)), `flatten` handles null sub-arrays + ([#4822](https://github.com/apache/datafusion-comet/pull/4822)), `ArrayInsert` handles null source arrays + ([#4726](https://github.com/apache/datafusion-comet/pull/4726)), and the `array_filter` / `array_compact` + fast path is restricted to the lambda variable + ([#4848](https://github.com/apache/datafusion-comet/pull/4848)). +- **Strings and encodings**: `CAST(binary AS string)` now decodes exactly as the JVM's + `new String(bytes, UTF_8)` does, including the surrogate cases where Rust's lossy decoder diverges, instead + of reinterpreting bytes unchecked ([#4763](https://github.com/apache/datafusion-comet/pull/4763)), and + shuffle tolerates non-UTF-8 bytes rather than failing + ([#4524](https://github.com/apache/datafusion-comet/pull/4524)). +- **Timestamps**: reading a Parquet timestamp-with-timezone column as `TimestampNTZ` is now rejected on Spark + 3.x, matching Spark's own refusal, instead of silently returning the UTC instant + ([#4357](https://github.com/apache/datafusion-comet/pull/4357)). +- **Deep expression trees**: long chains of associative bitwise, `Add`, and `Multiply` operators are rebalanced + so plans no longer hit protobuf's recursion limit + ([#4588](https://github.com/apache/datafusion-comet/pull/4588)). +- **Native shuffle**: scalar subqueries used in partitioning expressions are now registered, fixing + "Subquery N not found" failures ([#4869](https://github.com/apache/datafusion-comet/pull/4869)), and constant + column vectors are materialized on Comet's serialize and export paths + ([#4532](https://github.com/apache/datafusion-comet/pull/4532)). +- **Collations**: Comet falls back for predicates whose operands use non-default collations + ([#4948](https://github.com/apache/datafusion-comet/pull/4948)), for Spark 4 datetime expressions under + non-default collations ([#4693](https://github.com/apache/datafusion-comet/pull/4693)), and for + `str_to_map` ([#4701](https://github.com/apache/datafusion-comet/pull/4701)). +- **Conservative fallbacks** where native behavior could not be made to match: decimal `SUM` / `AVG` over + sliding window frames ([#4732](https://github.com/apache/datafusion-comet/pull/4732)), + `FromUnixTime` with a non-default format ([#4847](https://github.com/apache/datafusion-comet/pull/4847)), + `CreateArray` with struct-nullability-divergent children + ([#4533](https://github.com/apache/datafusion-comet/pull/4533)), and native V1 scans on filesystem schemes + that `object_store` does not support ([#4525](https://github.com/apache/datafusion-comet/pull/4525)). +- **Error reporting**: native Parquet read failures now surface as Spark's `FAILED_READ_FILE` + ([#4536](https://github.com/apache/datafusion-comet/pull/4536)), and a DataFusion 54.1.0 Parquet page-index + regression is worked around ([#5132](https://github.com/apache/datafusion-comet/pull/5132)). ## New Expression and Aggregate Support @@ -114,14 +189,16 @@ This release expands the set of Spark expressions and aggregates that run native ([#4815](https://github.com/apache/datafusion-comet/pull/4815)). - **Intervals**: interval types with `make_ym_interval` and `make_dt_interval` ([#4541](https://github.com/apache/datafusion-comet/pull/4541)), - `CalendarIntervalType` support ([#4898](https://github.com/apache/datafusion-comet/pull/4898)), and + `CalendarIntervalType` support ([#4898](https://github.com/apache/datafusion-comet/pull/4898)), `multiply_dt_interval` via codegen dispatch - ([#4900](https://github.com/apache/datafusion-comet/pull/4900)). + ([#4900](https://github.com/apache/datafusion-comet/pull/4900)), and interval codegen dispatch for nested + values and native shuffle ([#4976](https://github.com/apache/datafusion-comet/pull/4976)). - **String**: `base64` ([#4778](https://github.com/apache/datafusion-comet/pull/4778)), `split_part` via `StringSplitSQL` ([#4592](https://github.com/apache/datafusion-comet/pull/4592)), native `levenshtein` ([#4105](https://github.com/apache/datafusion-comet/pull/4105)), and native `randstr` compatible with Spark ([#5035](https://github.com/apache/datafusion-comet/pull/5035)). - **Array / map**: `array_prepend` ([#4716](https://github.com/apache/datafusion-comet/pull/4716)), + the `shuffle()` array function ([#4797](https://github.com/apache/datafusion-comet/pull/4797)), `size()` for `MapType` ([#4580](https://github.com/apache/datafusion-comet/pull/4580)), `ElementAt` over `MapType` ([#4697](https://github.com/apache/datafusion-comet/pull/4697)), and removal of the constraint on arrays of nested elements ([#4714](https://github.com/apache/datafusion-comet/pull/4714)). @@ -132,6 +209,19 @@ This release expands the set of Spark expressions and aggregates that run native - **Windows**: extended native window function support ([#4209](https://github.com/apache/datafusion-comet/pull/4209)) and Spark 4 decimal window average ([#4749](https://github.com/apache/datafusion-comet/pull/4749)). +- **Internal expressions**: native `empty2null` + ([#4683](https://github.com/apache/datafusion-comet/pull/4683)), which appears in write paths, and the + shuffle-side infrastructure for Spark's `TimeType` + ([#4398](https://github.com/apache/datafusion-comet/pull/4398)). + +## Native Sampling + +Spark's `SampleExec` now runs natively when sampling without replacement +([#5110](https://github.com/apache/datafusion-comet/pull/5110)), covering `DataFrame.sample`, SQL +`TABLESAMPLE`, and `DataFrame.randomSplit`. The native operator ports Spark's `BernoulliCellSampler` on top of +`XorShiftRandom` and seeds per partition exactly as Spark does, so it selects the same rows for a given seed — +sampling stays reproducible whether or not Comet is enabled. Sampling with replacement still falls back to +Spark. ## Experimental PyArrow UDF Support @@ -175,12 +265,24 @@ early-stage feature and we welcome feedback from users experimenting with it. - **Codegen dispatch for opt-in serdes** ([#4728](https://github.com/apache/datafusion-comet/pull/4728)): expressions reported as `Unsupported` can now route through JVM codegen dispatch for opt-in serdes, and native opt-in expressions surface as compatible-by-default with a `COMET-INFO` plan hint - ([#4721](https://github.com/apache/datafusion-comet/pull/4721)). -- **Collated predicate fallback** ([#4948](https://github.com/apache/datafusion-comet/pull/4948)): Comet - now falls back to Spark for predicates whose operands use non-default collations, avoiding subtle - correctness issues. -- **ANSI cast to date** ([#5014](https://github.com/apache/datafusion-comet/pull/5014)): invalid calendar - dates now raise `CAST_INVALID_INPUT` under ANSI mode, matching Spark's behavior. + ([#4721](https://github.com/apache/datafusion-comet/pull/4721)). Casts join this path + ([#5079](https://github.com/apache/datafusion-comet/pull/5079)): an incompatible or unsupported cast now runs + Spark's own generated code inside the Comet pipeline instead of pulling the whole operator back to Spark, so + one awkward cast no longer de-accelerates a query. `sort_array` under strict floating-point mode + ([#4637](https://github.com/apache/datafusion-comet/pull/4637)) and `concat` under non-`UTF8_BINARY` + collations ([#4640](https://github.com/apache/datafusion-comet/pull/4640)) opt in the same way. +- **JVM columnar-to-row by default** ([#5114](https://github.com/apache/datafusion-comet/pull/5114)): isolated + benchmarking showed the native columnar-to-row converter is roughly 3.7x slower per row than the JVM + implementation at the default batch size — and up to 15.7x slower for small batches — because of a fixed + per-batch JNI and FFI cost, with no end-to-end benefit in TPC-DS runs. It is now disabled by default. +- **Clearer fallback reasons**: unsupported scalar serde cases are reported in the support level + ([#4745](https://github.com/apache/datafusion-comet/pull/4745)), mixed-execution fallback messages name the + incompatible aggregate functions ([#4750](https://github.com/apache/datafusion-comet/pull/4750)), unsupported + metadata column names appear in fallback reasons + ([#4758](https://github.com/apache/datafusion-comet/pull/4758)), the compatibility guide URLs in fallback + messages are corrected ([#4854](https://github.com/apache/datafusion-comet/pull/4854)), and the spurious + "WriteFilesExec is not supported" message is suppressed + ([#4928](https://github.com/apache/datafusion-comet/pull/4928)). - **Local scan nullability** ([#4843](https://github.com/apache/datafusion-comet/pull/4843)): local table scan child nullability is now widened to match the native kernels, fixing a class of nullability mismatches. - **Config aliases** ([#4979](https://github.com/apache/datafusion-comet/pull/4979)): a `withAlternative` @@ -188,6 +290,8 @@ early-stage feature and we welcome feedback from users experimenting with it. configurations. - **Runtime cleanup** ([#4734](https://github.com/apache/datafusion-comet/pull/4734)): the Tokio runtime is now released on driver and executor exit. +- **Quieter logs** ([#5155](https://github.com/apache/datafusion-comet/pull/5155)): logging levels on frequent + call sites have been lowered so Comet no longer floods executor logs at default settings. - **Contrib scan SPI** ([#4700](https://github.com/apache/datafusion-comet/pull/4700)): a core SPI for contrib leaf scans (`CometScanWithPlanData`), the first part of splitting the Delta integration into a contrib module. @@ -200,6 +304,48 @@ early-stage feature and we welcome feedback from users experimenting with it. schema is now encoded once per writer instead of per block, cutting per-batch shuffle overhead. - **BatchCoalescer bypass** ([#5003](https://github.com/apache/datafusion-comet/pull/5003)): shuffle bypasses the `BatchCoalescer` for batches that are already appropriately sized. +- **Single-partition shuffle** ([#5004](https://github.com/apache/datafusion-comet/pull/5004)): a redundant + concatenation layer has been removed from the single-partition path. + +## Configuration Changes + +Because configuration is Comet's primary API surface, 1.0 is the release where the naming gets cleaned up. +Every rename below registers the old key as an alias, so existing configurations keep working and log a +deprecation warning when an old key is read: + +- **Unified shuffle prefix** ([#4986](https://github.com/apache/datafusion-comet/pull/4986)): shuffle configs + were spread across four disjoint prefixes (`spark.comet.exec.shuffle.*`, `spark.comet.columnar.shuffle.*`, + `spark.comet.native.shuffle.*`, `spark.comet.shuffle.*`). They now all live under `spark.comet.shuffle.*`, + with `.jvm.` and `.native.` sub-namespaces matching the `spark.comet.shuffle.mode` value you already set. +- **Grouped explain configs** ([#5026](https://github.com/apache/datafusion-comet/pull/5026)): the orphan + explain-related configs are collected under a single `spark.comet.explain.*` prefix. +- **`spark.comet.version`** ([#5049](https://github.com/apache/datafusion-comet/pull/5049)): the loaded Comet + build version is now exposed as a runtime config, so you can confirm which Comet a cluster is actually + running with `spark.conf.get` or `SET`. +- **Removed dead configs**: the Parquet parallel-IO knobs + ([#4981](https://github.com/apache/datafusion-comet/pull/4981)) and + `spark.comet.use.lazyMaterialization` ([#4998](https://github.com/apache/datafusion-comet/pull/4998)) had no + remaining effect and were misleading anyone tuning against them. +- **Removed async columnar shuffle** ([#4985](https://github.com/apache/datafusion-comet/pull/4985)): the + `spark.comet.columnar.shuffle.async.*` path was off by default and untested, and has been removed rather than + carried into a stable release line. + +## Documentation + +The documentation received a substantial overhaul for 1.0: + +- A [design refresh](https://github.com/apache/datafusion-comet/pull/4353) of the docs site, with versioned + user guides gaining captioned sidebar sections + ([#4699](https://github.com/apache/datafusion-comet/pull/4699)) and a set of accessibility and navigation + fixes ([#4858](https://github.com/apache/datafusion-comet/pull/4858)). +- The supported-expressions page now shows how each expression is implemented — native, codegen dispatch, or + hybrid ([#5028](https://github.com/apache/datafusion-comet/pull/5028)), generated from the planner itself + rather than maintained by hand. +- An expanded [tuning guide](https://github.com/apache/datafusion-comet/pull/4908) covering the performance and + memory configs that matter in practice, plus guidance on `spark.sql.files.maxPartitionBytes` + ([#4931](https://github.com/apache/datafusion-comet/pull/4931)). +- An [updated roadmap](https://github.com/apache/datafusion-comet/pull/5064) for the post-1.0 work, and a new + [blog posts and talks page](https://github.com/apache/datafusion-comet/pull/5043). ## Deprecation Notice @@ -218,13 +364,14 @@ Supported platforms include: - **Spark 3.4.3** with Java 11/17 and Scala 2.12/2.13 (deprecated, removal in 1.1.0) - **Spark 3.5.8** with Java 11/17 and Scala 2.12/2.13 - **Spark 4.0.2** with Java 17 and Scala 2.13 -- **Spark 4.1.1** with Java 17 and Scala 2.13 +- **Spark 4.1.2** with Java 17/21 and Scala 2.13 +- **Spark 4.2** with Java 17 and Scala 2.13 (experimental, for early evaluation only) See the [Spark Version Compatibility] page for known limitations specific to each version. [Spark Version Compatibility]: https://datafusion.apache.org/comet/user-guide/latest/compatibility/spark-versions.html -This release upgrades to **DataFusion 54.1** and **Arrow 58.3**. +This release upgrades to **DataFusion 54.1** and **Arrow 58.4**. ## Get Started with Comet 1.0.0 From 0a28ee3565383b1ab3b5a630dc766e1b51f9c22a Mon Sep 17 00:00:00 2001 From: Andy Grove Date: Tue, 4 Aug 2026 09:23:27 -0600 Subject: [PATCH 4/6] Update Comet 1.0.0 blog post for final branch-1.0 state --- .../blog/2026-07-31-datafusion-comet-1.0.0.md | 66 ++++++++++++++----- 1 file changed, 51 insertions(+), 15 deletions(-) diff --git a/content/blog/2026-07-31-datafusion-comet-1.0.0.md b/content/blog/2026-07-31-datafusion-comet-1.0.0.md index 1578f085..15e577ea 100644 --- a/content/blog/2026-07-31-datafusion-comet-1.0.0.md +++ b/content/blog/2026-07-31-datafusion-comet-1.0.0.md @@ -36,7 +36,7 @@ This is a major milestone. Comet began as a code donation in early 2024 and has one release at a time. Version 1.0.0 marks the point where the project is mature enough to commit to a stable release line: broad Apache Spark coverage, ANSI SQL semantics, native Parquet and Iceberg scans, and a native shuffle, all validated continuously against Spark's own test suites. This release covers roughly six weeks of -development since 0.17.0 and is the result of merging over 220 PRs from 21 contributors. See the +development since 0.17.0 and consists of 244 commits from 23 contributors. See the [change log] for more information. [change log]: https://github.com/apache/datafusion-comet/blob/main/docs/source/changelog/1.0.0.md @@ -45,7 +45,7 @@ development since 0.17.0 and is the result of merging over 220 PRs from 21 contr Reaching 1.0 is less about any single new feature than about the accumulated maturity of the project: -- **Broad Spark coverage.** Comet supports Apache Spark 3.4.3, 3.5.8, 4.0.2, and 4.1.2 out of the same +- **Broad Spark coverage.** Comet supports Apache Spark 3.4.3, 3.5.9, 4.0.4, and 4.1.3 out of the same codebase, with dedicated Maven profiles, shim sources, and CI matrices for each, plus an experimental Spark 4.2 profile for early evaluation. The [Spark version adoption and support-lifetime policy](https://github.com/apache/datafusion-comet/pull/4977) @@ -68,7 +68,9 @@ Reaching 1.0 is less about any single new feature than about the accumulated mat [Deprecation Notice](#deprecation-notice) below). - **Documented limitations.** Every open correctness issue is now surfaced in the generated compatibility guide ([#5085](https://github.com/apache/datafusion-comet/pull/5085)), down to the expression level, so you can see - where Comet is known to diverge from Spark before you hit it in production rather than after. + where Comet is known to diverge from Spark before you hit it in production rather than after. Notes for bugs + that have since been fixed were removed in the same pass + ([#5154](https://github.com/apache/datafusion-comet/pull/5154)). The rest of this post covers what is new since 0.17.0. @@ -117,17 +119,36 @@ size hint so a single read usually captures the footer, matching the Iceberg pat options from the session config so Parquet settings you already set take effect ([#5107](https://github.com/apache/datafusion-comet/pull/5107)). -Two lower-level costs were also removed: Comet no longer makes excessive timer calls in the native execution -loop ([#4739](https://github.com/apache/datafusion-comet/pull/4739)), and plan-data injection is now an O(1) -lookup by operator kind rather than a scan ([#4535](https://github.com/apache/datafusion-comet/pull/4535)). +Query planning and plan serialization also got cheaper. `QueryContext` SQL text is now interned into a per-plan +pool ([#5204](https://github.com/apache/datafusion-comet/pull/5204)), which makes serialized plans up to 20x +smaller on TPC-DS — every plan crosses the JNI boundary, so this shrinks per-query overhead across the board. +Plan-data injection is now an O(1) lookup by operator kind +([#4535](https://github.com/apache/datafusion-comet/pull/4535)) and no longer rebuilds operators it does not +touch ([#5220](https://github.com/apache/datafusion-comet/pull/5220)). Comet also makes far fewer timer calls +in the native execution loop ([#4739](https://github.com/apache/datafusion-comet/pull/4739)), and nested array +equality now uses Arrow's comparator ([#5176](https://github.com/apache/datafusion-comet/pull/5176)). ## Correctness A 1.0 release is only as good as its results. This release fixes a broad set of divergences from Spark, most of them found by running Spark's own SQL test suites through Comet's native path: -- **ANSI error semantics**: `round` on a `Long` with a large negative scale now overflows instead of silently - returning zero ([#5082](https://github.com/apache/datafusion-comet/pull/5082)), `Long.MinValue / -1` raises +- **Casts**: casting a string to `boolean`, an integral type, `float`/`double`, or `decimal` now uses Spark's + exact whitespace-trimming rules ([#5150](https://github.com/apache/datafusion-comet/pull/5150)). Comet's + kernels had used four different trim sets, three of them wrong, so results diverged in both directions — + returning null where Spark parses a value, and returning a value where Spark returns null. Casting + `float`/`double` to `decimal` now rounds the shortest decimal string form as Spark does, rather than the + binary value ([#5136](https://github.com/apache/datafusion-comet/pull/5136)), which matters for values such + as `0.5153125` whose binary form sits just below the rounding tie the string form lands on; `NaN` and infinity + now return null even in ANSI mode, matching Spark. Decimal promotion also uses the per-expression eval mode + ([#5171](https://github.com/apache/datafusion-comet/pull/5171)). +- **ANSI error semantics**: the codegen dispatcher's null short-circuit no longer swallows errors Spark raises + ([#5219](https://github.com/apache/datafusion-comet/pull/5219)) — Spark evaluates null-intolerant expressions + per node and left to right, so short-circuiting on the union of input ordinals skipped subtrees Spark would + have evaluated, losing their errors. Roughly 70 built-in expressions route through this dispatcher and ANSI is + on by default in Spark 4, so this affected a wide surface. Also: `round` on a `Long` with a large negative + scale now overflows instead of silently returning zero + ([#5082](https://github.com/apache/datafusion-comet/pull/5082)), `Long.MinValue / -1` raises `ARITHMETIC_OVERFLOW` ([#5084](https://github.com/apache/datafusion-comet/pull/5084)), floating-point `x % 0.0` raises `REMAINDER_BY_ZERO` instead of returning `NaN` ([#5081](https://github.com/apache/datafusion-comet/pull/5081)), `make_decimal` honors its @@ -196,7 +217,9 @@ This release expands the set of Spark expressions and aggregates that run native - **String**: `base64` ([#4778](https://github.com/apache/datafusion-comet/pull/4778)), `split_part` via `StringSplitSQL` ([#4592](https://github.com/apache/datafusion-comet/pull/4592)), native `levenshtein` ([#4105](https://github.com/apache/datafusion-comet/pull/4105)), and native - `randstr` compatible with Spark ([#5035](https://github.com/apache/datafusion-comet/pull/5035)). + `randstr` ([#5035](https://github.com/apache/datafusion-comet/pull/5035)) and `uuid` + ([#5034](https://github.com/apache/datafusion-comet/pull/5034)), both bit-for-bit compatible with Spark for a + given seed. - **Array / map**: `array_prepend` ([#4716](https://github.com/apache/datafusion-comet/pull/4716)), the `shuffle()` array function ([#4797](https://github.com/apache/datafusion-comet/pull/4797)), `size()` for `MapType` ([#4580](https://github.com/apache/datafusion-comet/pull/4580)), `ElementAt` over @@ -241,10 +264,14 @@ early-stage feature and we welcome feedback from users experimenting with it. decryption for V3 tables, with fallback to Spark for other V3 features. Follow-up [#5020](https://github.com/apache/datafusion-comet/pull/5020) applies the same diff changes across other Iceberg versions. +- **Metadata columns** ([#4752](https://github.com/apache/datafusion-comet/pull/4752)): the native Iceberg scan + now supports the `_pos`, `_spec`, `_file`, and `_partition` metadata columns. - **Delete-file correctness** ([#4760](https://github.com/apache/datafusion-comet/pull/4760)): the native scan now sizes Iceberg delete files correctly, avoiding dropped deletes. - **Exchange-reuse correctness** ([#4812](https://github.com/apache/datafusion-comet/pull/4812)): fixed a case where Iceberg native scan exchange reuse with different pushed filters could produce wrong results. +- **Scan disambiguation** ([#5180](https://github.com/apache/datafusion-comet/pull/5180)): Iceberg scans that + share a `metadata_location` are now told apart, rather than being conflated. - **Native serde dedup** ([#4982](https://github.com/apache/datafusion-comet/pull/4982)): dedupes Iceberg residuals and delete files in the native scan serde, reducing planning overhead. @@ -283,6 +310,12 @@ early-stage feature and we welcome feedback from users experimenting with it. messages are corrected ([#4854](https://github.com/apache/datafusion-comet/pull/4854)), and the spurious "WriteFilesExec is not supported" message is suppressed ([#4928](https://github.com/apache/datafusion-comet/pull/4928)). +- **Expression coverage in extended explain** ([#5201](https://github.com/apache/datafusion-comet/pull/5201)): + the extended explain summary previously reported operator coverage but said nothing about expressions. It now + ends with a line such as `Comet accelerated 14 expressions (14 native, 1 codegen dispatch)`, so you can see + how much of a plan's expression evaluation runs in native DataFusion kernels versus Spark's generated code + inside the dispatcher. Operator counts also handle `ReusedSubquery` and `CometSubqueryBroadcast` correctly + ([#5206](https://github.com/apache/datafusion-comet/pull/5206)). - **Local scan nullability** ([#4843](https://github.com/apache/datafusion-comet/pull/4843)): local table scan child nullability is now widened to match the native kernels, fixing a class of nullability mismatches. - **Config aliases** ([#4979](https://github.com/apache/datafusion-comet/pull/4979)): a `withAlternative` @@ -318,13 +351,16 @@ deprecation warning when an old key is read: `spark.comet.native.shuffle.*`, `spark.comet.shuffle.*`). They now all live under `spark.comet.shuffle.*`, with `.jvm.` and `.native.` sub-namespaces matching the `spark.comet.shuffle.mode` value you already set. - **Grouped explain configs** ([#5026](https://github.com/apache/datafusion-comet/pull/5026)): the orphan - explain-related configs are collected under a single `spark.comet.explain.*` prefix. + explain-related configs are collected under a single `spark.comet.explain.*` prefix, and the PyArrow UDF + config is renamed to `pyarrowUDF` for consistency + ([#5197](https://github.com/apache/datafusion-comet/pull/5197)). - **`spark.comet.version`** ([#5049](https://github.com/apache/datafusion-comet/pull/5049)): the loaded Comet build version is now exposed as a runtime config, so you can confirm which Comet a cluster is actually running with `spark.conf.get` or `SET`. - **Removed dead configs**: the Parquet parallel-IO knobs - ([#4981](https://github.com/apache/datafusion-comet/pull/4981)) and - `spark.comet.use.lazyMaterialization` ([#4998](https://github.com/apache/datafusion-comet/pull/4998)) had no + ([#4981](https://github.com/apache/datafusion-comet/pull/4981)), + `spark.comet.use.lazyMaterialization` ([#4998](https://github.com/apache/datafusion-comet/pull/4998)), and + `spark.comet.exceptionOnDatetimeRebase` ([#5221](https://github.com/apache/datafusion-comet/pull/5221)) had no remaining effect and were misleading anyone tuning against them. - **Removed async columnar shuffle** ([#4985](https://github.com/apache/datafusion-comet/pull/4985)): the `spark.comet.columnar.shuffle.async.*` path was off by default and untested, and has been removed rather than @@ -362,9 +398,9 @@ Users on these platforms should plan to move to JDK 17+ and Spark 3.5 or later b Supported platforms include: - **Spark 3.4.3** with Java 11/17 and Scala 2.12/2.13 (deprecated, removal in 1.1.0) -- **Spark 3.5.8** with Java 11/17 and Scala 2.12/2.13 -- **Spark 4.0.2** with Java 17 and Scala 2.13 -- **Spark 4.1.2** with Java 17/21 and Scala 2.13 +- **Spark 3.5.9** with Java 11/17 and Scala 2.12/2.13 +- **Spark 4.0.4** with Java 17 and Scala 2.13 +- **Spark 4.1.3** with Java 17/21 and Scala 2.13 - **Spark 4.2** with Java 17 and Scala 2.13 (experimental, for early evaluation only) See the [Spark Version Compatibility] page for known limitations specific to each version. From d8a539a9ed50d1b33956a65f87e85876b88d2b2c Mon Sep 17 00:00:00 2001 From: Andy Grove Date: Tue, 4 Aug 2026 09:39:08 -0600 Subject: [PATCH 5/6] fixes --- content/blog/2026-07-31-datafusion-comet-1.0.0.md | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-) diff --git a/content/blog/2026-07-31-datafusion-comet-1.0.0.md b/content/blog/2026-07-31-datafusion-comet-1.0.0.md index 15e577ea..2f3ba677 100644 --- a/content/blog/2026-07-31-datafusion-comet-1.0.0.md +++ b/content/blog/2026-07-31-datafusion-comet-1.0.0.md @@ -32,18 +32,17 @@ The Apache DataFusion PMC is pleased to announce version 1.0.0 of the [Comet](ht Comet is an accelerator for Apache Spark that translates Spark physical plans to DataFusion physical plans for improved performance and efficiency without requiring any code changes. -This is a major milestone. Comet began as a code donation in early 2024 and has shipped steadily ever since, -one release at a time. Version 1.0.0 marks the point where the project is mature enough to commit to a stable -release line: broad Apache Spark coverage, ANSI SQL semantics, native Parquet and Iceberg scans, and a native -shuffle, all validated continuously against Spark's own test suites. This release covers roughly six weeks of -development since 0.17.0 and consists of 244 commits from 23 contributors. See the -[change log] for more information. +This release covers roughly six weeks of +development since 0.17.0 and consists of 244 commits from 23 contributors. See the [change log] for more information. + [change log]: https://github.com/apache/datafusion-comet/blob/main/docs/source/changelog/1.0.0.md ## What 1.0 Means -Reaching 1.0 is less about any single new feature than about the accumulated maturity of the project: +The 1.0.0 release is the culmination of more than two year's work since the project was [donated] as an Apache DataFusion subproject in March, 2024, and is less about any single new feature than about the accumulated maturity of the project: + +[donated]: https://datafusion.apache.org/blog/2024/03/06/comet-donation/ - **Broad Spark coverage.** Comet supports Apache Spark 3.4.3, 3.5.9, 4.0.4, and 4.1.3 out of the same codebase, with dedicated Maven profiles, shim sources, and CI matrices for each, plus an experimental @@ -131,7 +130,7 @@ equality now uses Arrow's comparator ([#5176](https://github.com/apache/datafusi ## Correctness A 1.0 release is only as good as its results. This release fixes a broad set of divergences from Spark, most of -them found by running Spark's own SQL test suites through Comet's native path: +them found by running extensive AI-assisted audit sweeps of the code base, comparing Comet's expression implementations with all supported versions of Spark. - **Casts**: casting a string to `boolean`, an integral type, `float`/`double`, or `decimal` now uses Spark's exact whitespace-trimming rules ([#5150](https://github.com/apache/datafusion-comet/pull/5150)). Comet's From 769cee76e1d5bfd9d5d074fc84da3973e57978f7 Mon Sep 17 00:00:00 2001 From: Andy Grove Date: Tue, 4 Aug 2026 09:41:03 -0600 Subject: [PATCH 6/6] change date --- ...sion-comet-1.0.0.md => 2026-08-07-datafusion-comet-1.0.0.md} | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) rename content/blog/{2026-07-31-datafusion-comet-1.0.0.md => 2026-08-07-datafusion-comet-1.0.0.md} (99%) diff --git a/content/blog/2026-07-31-datafusion-comet-1.0.0.md b/content/blog/2026-08-07-datafusion-comet-1.0.0.md similarity index 99% rename from content/blog/2026-07-31-datafusion-comet-1.0.0.md rename to content/blog/2026-08-07-datafusion-comet-1.0.0.md index 2f3ba677..dc8475da 100644 --- a/content/blog/2026-07-31-datafusion-comet-1.0.0.md +++ b/content/blog/2026-08-07-datafusion-comet-1.0.0.md @@ -1,7 +1,7 @@ --- layout: post title: Apache DataFusion Comet 1.0.0 Release -date: 2026-07-31 +date: 2026-08-07 author: pmc categories: [subprojects] ---