What's New in v21.1.0-rc.1

May 05, 2021

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

$ docker pull cockroachdb/cockroach-unstable:v21.1.0-rc.1

SQL language changes

CockroachDB no longer allows ADD REGION or DROP REGION statements if a REGIONAL BY ROW table has index changes underway, or if a table is transitioning to or from REGIONAL BY ROW. #64255
CockroachDB now prevents index modification on REGIONAL BY ROW tables and locality to or from REGIONAL BY ROW changes while an ADD REGION or DROP REGION statement is being executed. #64255

Bug fixes

Fixed a scenario in which a rapid sequence of range splits could trigger a storm of Raft snapshots. This would be accompanied by log messages of the form "would have dropped incoming MsgApp, but allowing due to ..." and tended to occur as part of RESTORE/IMPORT operations. #64202
Read-write contention on GLOBAL tables no longer has a potential to thrash without making progress. #64215
Previously, if a DROP INDEX failed during a REGIONAL BY ROW transition, the index could have been re-inserted back into the REGIONAL BY ROW table but would be invalid if it was hash-sharded or partitioned. This bug is now fixed. #64255
Fixed a rare bug present in v21.1 beta versions that could cause rapid range splits and merges on a GLOBAL table to lead to a stuck leaseholder replica. The situation is no longer possible. #64304
Fixed a bug in previous v21.1 beta versions that allowed the store rebalancer to spuriously down-replicate a range during normal operation. #64303
CockroachDB now prevents some out-of-memory conditions caused by schema change validations concurrent with other high-memory-use queries. #64307
Fixed a bug present since v21.1.0-alpha.1 that could cause cascading DELETEs with subqueries to error. #64278
Fixed a bug that caused store information to be incorrectly redacted from the CockroachDB logs, when logging was configured with redaction. #64338
Previously, the remote flows of execution in the vectorized engine could take a long time to shut down if a node participating in the plan died. This bug is now fixed. #64219