Add vectorized filtering to constraint checking #7972
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
antekresic
added
the
enhancement
antekresic
force-pushed
the
antekresic/add_vectorized_filtering
branch
from
3357ca6 to
2b51915
Compare
akuzm
reviewed
Apr 16, 2025
akuzm
reviewed
Apr 16, 2025
akuzm
reviewed
Apr 16, 2025
akuzm
approved these changes
Apr 30, 2025
There was a problem hiding this comment.
We discussed this on a call, generally this can be OK as the first step, because it depends on ArrowArray and decompression functions which are low level but generally safe interfaces (no major changes planned). Ideally would be good to refactor this to use vector_qual_compute to avoid re-implementing the logic. The main obstacle is lack of the Postgres Expr quals, the DML code builds the ScanKeys right away from the unique constraint.
akuzm
reviewed
Apr 30, 2025
antekresic
force-pushed
the
antekresic/add_vectorized_filtering
branch
2 times, most recently
from
ae59e15 to
ab74e60
Compare
svenklemm
approved these changes
Apr 30, 2025
When backfilling compressed chunks with constraint checking, we need to read and filter compressed tuples in order to confirm constraints are validated. This change adds the ability to do this operation using vectorized decompression and vectorized filtering if the types allow us to do so.
antekresic
force-pushed
the
antekresic/add_vectorized_filtering
branch
from
ab74e60 to
6bb27de
Compare
philkra
added a commit
that referenced
this pull request
May 15, 2025
## 2.20.0 (2025-05-15) This release contains performance improvements and bug fixes since the 2.19.3 release. We recommend that you upgrade at the next available opportunity. **Highlighted features in TimescaleDB v2.20.0** * The columnstore now leverages *bloom filters* to deliver up to 6x faster point queries on columns with high cardinality values, such as UUIDs. * Major *improvements to the columnstores' backfill process* enable `UPSERTS` with strict constraints to execute up to 10x faster. * *SkipScan is now supported in the columnstore*, including for DISTINCT queries. This enhancement leads to dramatic query performance improvements of 2000x to 2500x, especially for selective queries. * SIMD vectorization for the bool data type is now enabled by default. This change results in a 30–45% increase in performance for analytical queries with bool clauses on the columnstore. * *Continuous aggregates* now include experimental support for *window functions and non-immutable functions*, extending the analytics use cases they can solve. * Several quality-of-life improvements have been introduced: job names for continuous aggregates are now more descriptive, you can assign custom names to them, and it is now possible to add unique constraints along with `ADD COLUMN` operations in the columnstore. * Improved management and optimization of chunks with the ability to split large uncompressed chunks at a specified point in time using the `split_chunk` function. This new function complements the existing `merge_chunk` function that can be used to merge two small chunks into one larger chunk. * Enhancements to the default behavior of the columnstore now provide better *automatic assessments* of `segment by` and `order by` columns, reducing the need for manual configuration and simplifying initial setup. **PostgreSQL 14 support removal announcement** Following the deprecation announcement for PostgreSQL 14 in TimescaleDB v2.19.0, PostgreSQL 14 is no longer supported in TimescaleDB v2.20.0. The currently supported PostgreSQL major versions are 15, 16, and 17. **Features** * [#7638](#7638) Bloom filter sparse indexes for compressed columns. Can be disabled with the GUC `timescaledb.enable_sparse_index_bloom` * [#7756](#7756) Add warning for poor compression ratio * [#7762](#7762) Speed up the queries that use minmax sparse indexes on compressed tables by changing the index TOAST storage type to `MAIN`. This applies to newly compressed chunks * [#7785](#7785) Do `DELETE` instead of `TRUNCATE` when locks aren't acquired * [#7852](#7852) Allow creating foreign key constraints on compressed tables * [#7854](#7854) Remove support for PG14 * [#7864](#7854) Allow adding CHECK constraints to compressed chunks * [#7868](#7868) Allow adding columns with `CHECK` constraints to compressed chunks * [#7874](#7874) Support for SkipScan for distinct aggregates over the same column * [#7877](#7877) Remove blocker for unique constraints with `ADD COLUMN` * [#7878](#7878) Don't block non-immutable functions in continuous aggregates * [#7880](#7880) Add experimental support for window functions in continuous aggregates * [#7899](#7899) Vectorized decompression and filtering for boolean columns * [#7915](#7915) New option `refresh_newest_first` to continuous aggregate refresh policy API * [#7917](#7917) Remove `_timescaledb_functions.create_chunk_table` function * [#7929](#7929) Add `CREATE TABLE ... WITH` API for creating hypertables * [#7946](#7946) Add support for splitting a chunk * [#7958](#7958) Allow custom names for jobs * [#7972](#7972) Add vectorized filtering for constraint checking while backfilling into compressed chunks * [#7976](#7976) Include continuous aggregate name in jobs informational view * [#7977](#7977) Replace references to compression with columnstore * [#7981](#7981) Add columnstore as alias for `enable_columnstore `in `ALTER TABLE` * [#7983](#7983) Support for SkipScan over compressed data * [#7991](#7991) Improves default `segmentby` options * [#7992](#7992) Add API into hypertable invalidation log * [#8000](#8000) Add primary dimension info to information schema * [#8005](#8005) Support `ALTER TABLE SET (timescaledb.chunk_time_interval='1 day')` * [#8012](#8012) Add event triggers support on chunk creation * [#8014](#8014) Enable bool compression by default by setting `timescaledb.enable_bool_compression=true`. Note: for downgrading to `2.18` or earlier version, use [this downgrade script](https://github.com/timescale/timescaledb-extras/blob/master/utils/2.19.0-downgrade_new_compression_algorithms.sql) * [#8018](#8018) Add spin-lock during recompression on unique constraints * [#8026](#8026) Allow `WHERE` conditions that use nonvolatile functions to be pushed down to the compressed scan level. For example, conditions like `time > now()`, where `time` is a columnstore `orderby` column, will evaluate `now()` and use the sparse index on `time` to filter out the entire compressed batches that cannot contain matching rows. * [#8027](#8027) Add materialization invalidations API * [#8047](#8027) Support SkipScan for `SELECT DISTINCT` with multiple distincts when all but one distinct is pinned * [#8115](#8115) Add batch size limiting during compression **Bugfixes** * [#7862](#7862) Release cache pin when checking for `NOT NULL` * [#7909](#7909) Update compression stats when merging chunks * [#7928](#7928) Don't create a hypertable for implicitly published tables * [#7982](#7982) Fix crash in batch sort merge over eligible expressions * [#8008](#8008) Fix compression policy error message that shows number of successes * [#8031](#8031) Fix reporting of deleted tuples for direct batch delete * [#8033](#8033) Skip default `segmentby` if `orderby` is explicitly set * [#8061](#8061) Ensure settings for a compressed relation are found * [#7515](#7515) Add missing lock to Constraint-aware append * [#8067](#8067) Make sure hypercore TAM parent is vacuumed * [#8074](#8074) Fix memory leak in row compressor flush * [#8099](#8099) Block chunk merging on multi-dimensional hypertables * [#8106](#8106) Fix segfault when adding unique compression indexes to compressed chunks * [#8127](#8127) Read bit-packed version of booleans **GUCs** * `timescaledb.enable_sparse_index_bloom`: Enable creation of the bloom1 sparse index on compressed chunks; Default: `ON` * `timescaledb.compress_truncate_behaviour`: Defines how truncate behaves at the end of compression; Default: `truncate_only` * `timescaledb.enable_compression_ratio_warnings`: Enable warnings for poor compression ratio; Default: `ON` * `timescaledb.enable_event_triggers`: Enable event triggers for chunks creation; Default: `OFF` * `timescaledb.enable_cagg_window_functions`: Enable window functions in continuous aggregates; Default: `OFF` **Thanks** * @arajkumar for reporting that implicitly published tables were still able to create hypertables * @thotokraa for reporting an issue with unique expression indexes on compressed chunks --------- Signed-off-by: Philip Krauss <[email protected]> Signed-off-by: Ramon Guiu <[email protected]> Co-authored-by: Anastasiia Tovpeko <[email protected]> Co-authored-by: Ramon Guiu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When backfilling compressed chunks with constraint checking, we need to read and filter compressed tuples in order to confirm constraints are validated. This change adds the ability to do this operation using vectorized decompression and vectorized filtering if the types allow us to do so.