What's New in v21.1.7

August 09, 2021

This page lists additions and changes in v21.1.7 since v21.1.6.

Get future release notes emailed to you:

Downloads

Warning:
The CockroachDB executable for Windows is experimental and not suitable for production deployments. Windows 8 or higher is required.

Docker image

icon/buttons/copy
$ docker pull cockroachdb/cockroach:v21.1.7

Security updates

  • The --cert-principal-map flag passed to cockroach commands now allows the certificate principal name to contain colons. #67810

General changes

  • Added a new cluster setting (kv.transaction.reject_over_max_intents_budget) that controls the behavior of CockroachDB when a transaction exceeds the locks-tracking memory budget set by the kv.transaction.max_intents_bytes cluster setting. If kv.transaction.reject_over_max_intents_budget is set to true, CockroachDB rejects the query that would push its transaction over the memory budget with an error (error code 53400 - "configuration limit exceeded"). Transactions that don't track their locks precisely are potentially destabilizing for the cluster since cleaning up locks can take up considerable resources. Transactions that change many rows have the potential to run into this memory budget issue. #67967

SQL language changes

  • The remote DistSQL flows are now eagerly canceled if they were queued up and the query was canceled. #66331
  • Added a new cluster setting (changefeed.slow_span_log_threshold) that allows setting a cluster-wide default for slow span logging. #68106
  • Added a new session variable (enable_copying_partitioning_when_deinterleaving_table) which will change the behavior of ALTER PRIMARY KEY when performing a change which retains the same primary key but removes an INTERLEAVE INTO clause. When this variable is set to true and an ALTER PRIMARY KEY is run that only removes an INTERLEAVE INTO clause, the partitioning and zone configuration which applied to the root of the interleave will not be applied to the new primary index. The default value for enable_copying_partitioning_when_deinterleaving_table is equal to the value set for the new cluster setting sql.defaults.copy_partitioning_when_deinterleaving_table.enabled. #68114

Operational changes

DB Console changes

Bug fixes

  • Fixed a bug causing the ST_GeneratePoints built-in function to return a garbage value or an error if an empty geometry or negative nPoints input is given. #67580
  • Fixed a bug where DROP DATABASE could return errors if the database contained temporary views in use in an another session. #67172
  • Fixed a storage-level bug where Pebble would occasionally create excessively large SSTables, causing poor compaction performance and high read-amplification. This was especially likely after a manual offline compaction. #67610
  • Correlated subqueries that couldn't be decorrelated and that have their own subqueries are now executed correctly when supported. Note that it is an edge case of an edge case, so it's unlikely that users have hit this bug (it was found by the randomized testing). #67570
  • Fixed very rare, unexpected "index out of bounds" error from the vectorized engine when evaluating a CASE operator. #67779
  • Catching up Raft followers on the Raft log is now more efficient in the presence of many large Raft log entries. This helps avoid situations where Raft leaders struggle to retain leadership while catching up their followers. #67127
  • Fixed a bug that allowed rows to be inserted into a table with a CHECK constraint that always evaluated to false (e.g., CHECK (false)). This bug was present since version 21.1.0. #67341
  • Fixed a bug causing changefeeds to sometimes get stuck. #67968
  • Previously, CockroachDB nodes would crash whenever the cluster setting sql.trace.txn.enable_threshold was changed to a non-zero value. The bug was introduced in 21.1.0. #68027
  • Fixed a deadlock that could occur when many replicas were rapidly queued for removal. #65859
  • Fixed two bugs which affected geospatial queries with the st_distance function. The first bug caused errors for filters of the form st_distance(g1, g2, use_spheroid) = 0. The second could cause incorrect results in some cases; it incorrectly transformed filters of the form st_distance(g1, g2) = 0 when g1 and g2 were geographies to st_instersects(g1, g2). This is not a valid transformation because st_distance makes spheroid-based calculations by default while st_intersects only makes sphere-based calculations. #67392
  • Fixed a bug causing a prepared statement to incorrectly reuse the query plan of a different prepared statement that had similar, but not identical type hints. #67688
  • Fixed an issue with statistics estimation in the optimizer that could cause it to over-estimate the number of rows for some expressions and thus choose a sub-optimal plan. This issue could happen when multi-column statistics were used in combination with histograms, the query contained a predicate on two or more columns where the columns were highly correlated, and the selected values were very common according to the histograms. #67998
  • Previously, CockroachDB could encounter an internal error or crash when performing a cast of NULL JSON value to Geography or Geometry types. Now this is fixed. #67902
  • INSERT and UPDATE statements which operate on larger rows can now be split into batches using the sql.mutations.mutation_batch_byte_size setting. #67958
  • A rare bug that could result in a crash while creating a debug.zip file has been fixed. The bug was only possible to hit if a debug.zip file was captured during a period of rapid lease movement. #67728
  • Previously the GRANT and REVOKE commands would incorrectly handle role names. CockroachDB treats role names as case-insensitive, but these commands were incorrectly handling the names. Now, GRANT and REVOKE normalize the names and are case-insensitive. #67901

Performance improvements

  • Vectorized flows can use less memory when sending and receiving data to the network. #67609
  • Range merges are no longer considered if a range has seen significant load over the previous 5 minutes, instead of being considered as long as a range had low load over the previous second. This change improves stability, as load-based splits will no longer rapidly disappear during transient throughput dips. #65362
  • A new cluster setting sql.defaults.optimizer_improve_disjunction_selectivity.enabled enables more accurate selectivity estimation of query filters with OR expressions. This improves query plans in some cases. This cluster setting is disabled by default. #67730

Contributors

This release includes 47 merged PRs by 26 authors.

YesYes NoNo