In versions before 2.24, the hashing scheme of the bloom filter sparse indexes used to depend on the build options of the TimescaleDB executables. These options are set by the package publishers and might differ between different package sources or even versions. After upgrading to a version with different options, the queries that use the bloom filter lookups could erroneously stop returning the rows that should in fact match the query conditions. The 2.24 release fixes this by using distinct column names for each hashing scheme.

The bloom filter sparse indexes will be disabled on the compressed chunks created before upgrading to 2.24. To re-enable them, you have to decompress and then compress the affected chunks.

If you were running the official APT package on AMD64 architecture, the hashing scheme did not change, and it is safe to use the existing bloom filter sparse indexes. To enable this, set the GUC timescaledb.read_legacy_bloom1_v1 = on in the server configuration.

The chunks compressed after upgrade to 2.24 will use the new index format, and the bloom filter sparse indexes will continue working as usual for these chunks without any intervention.

For more details, refer to the pull request #8761.

Deprecations

• The next release of TimescaleDB will remove the deprecated partial continuous aggregates format. The new format was introduced in 2.7.0 and provides significant improvements in terms of performance and storage efficiency. Please use cagg_migrate(<CONTINUOUS_AGGREGATE_NAME>) to migrate to the new format. Tiger Cloud users are migrated automatically.
• In future releases the deprecated view timescaledb_information.compression_settings will be removed. Please use timescaledb_information.hypertable_columnstore_settings as a replacement.
• The experimental view timescaledb_experimental.policies and the adjacent experimental functions add_policies, alter_policies, show_policies, remove_policies, and remove_all_policies to manage continuous aggregates will be removed in an upcoming release. For replacements, please use the Jobs API.

Backward-Incompatible Changes

• #8761 Fix matching rows in queries using the bloom filter sparse indexes potentially not returned after extension upgrade. The version of the bloom filter sparse indexes is changed. The existing indexes will stop working and will require action to re-enable. See the section above for details.

Features

• #8465 Speed up the filters like x = any(array[...]) using bloom filter sparse indexes.
• #8569 In-memory recompression
• #8754 Add concurrent mode for merging chunks
• #8786 Display chunks view range as timestamps for UUIDv7
• #8819 Refactor chunk compression logic
• #8840 Allow ALTER COLUMN TYPE when compression is enabled but no compressed chunks exist
• #8908 Add time bucketing support for UUIDv7
• #8909 Support direct compress on hypertables with continuous aggregates
• #8939 Support continuous aggregates on UUIDv7-partitioned hypertables
• #8959 Cap continuous aggregate invalidation interval range at chunk boundary
• #8975 Exclude date/time columns from default segmentby
• #8993 Add GUC for in-memory recompression

Bugfixes

• #8839 Improve _timescaledb_functions.cagg_watermark error handling
• #8853 Change log level of continuous aggregate refresh messages to DEBUG1
• #8933 Potential crash or seemingly random errors when querying the compressed chunks created on releases before 2.15 and using the minmax sparse indexes.
• #8942 Fix lateral join handling for compressed chunks
• #8958 Fix if_not_exists behaviour when adding refresh policy
• #8969 Gracefully handle missing job stat in background worker
• #8988 Don't ignore additional filters on same column when building scankeys

GUCs

• direct_compress_copy_tuple_sort_limit: Number of tuples that can be sorted at once in a COPY operation.
• direct_compress_insert_tuple_sort_limit: Number of tuples that can be sorted at once in an INSERT operation.
• read_legacy_bloom1_v1: Enable reading the legacy bloom1 version 1 sparse indexes for SELECT queries.
• enable_in_memory_recompression: Enable in-memory recompression functionality.

Thanks

• @bezpechno for implementing ALTER COLUMN TYPE for hypertable with columnstore when no compressed chunks exist