90097: kv: collect hot ranges per tenant on store r=koorosh a=koorosh The main goal of this change is to provide ability to request hot ranges info per tenant (as part of changes for UA). To track down hot ranges per tenant, `ReplicaRankingMap` struct is added that works almost identically as existing `ReplicaRankings` struct with only difference that hottest replicas are stored in an underlying map structure to keep separately ranges for every tenant. Also, it was decided to not extend existing `ReplicaRankings` functionality because it is used for replica rebalancing that doesn't need to know about tenants. This change might bring small overhead for self-hosted clusters as it will accumulate same hot ranges in both `replRankings` and `replRankingsByTenant` fields in Store. Changes in next PRs are supposed to rely on this change. Release note: None Epic: https://cockroachlabs.atlassian.net/browse/CRDB-19124 92313: roachtest: fix transfer-leases r=erikgrinaker a=tbg `checkNoLeases` verifies that once a node is drained, for each range there is one replica of the other nodes that sees the lease on one of the non-drained stores. The reason it asks justs for *one* replica to have the lease as opposed to all of them is because some followers may be behind. However, even just the assumption that there is one can be violated. The drain succeeds once the draining node has locally seen the lease transfer succeed. It is likely the raft leader at this point, so it will be the first one to see this event. Other nodes will only see it after one additional round-trip (when they learn that the log index has committed, and then go and apply it). So going and looking for a replica that sees the new lease immediately after the drain succeeds may fail. Work around this by sleeping for one second before checking, which ought to be enough, and is also a small enough delay to make sure that if leases are actually not getting transferred, the check will continue to fail (non-cooperative lease failover is on the order of multiple seconds). This commit also improves a debugging file which was previously clobbered over the multiple iterations of the surrounding loop. It also makes it clearer that we're pulling the range data from each node in turn (we were previously hitting otherNodeID but asking it to proxy to the node. Now we're hitting each node directly without relying on the internal redirect, which is less confusing). See: #91801 (comment) Fixes #91801. Epic: none Release note: None 92834: tree: improve type-checking for placeholders with ambiguous type r=jordanlewis a=rafiss fixes #90364 The key fix is to change the typeCheckSplitExprs function so that it marks _all_ placeholder indexes. This then causes the existing type-checking logic in typeCheckOverloadedExprs to check all placeholder expressions, rather than just ones that don't have type hints. Release note (bug fix): Prepared statements that use type hints can now succeed type-checking in more cases when the placeholder type is ambiguous. 93047: base: remove old testing knobs r=andreimatei a=andreimatei The testing knobs for startup migrations don't exist anymore. Release note: None Epic: None 93074: testutils: add missing comment r=andreimatei a=andreimatei Release note: None Epic: None 93079: insights: record more txn insights r=xinhaoz a=xinhaoz Closes #93076 This commit adds the following fields at the txn level when recording insights in the sql insights system: - contention: total txn contention time - start_time: txn start time - end_time: txn end time - auto_retry_reason: last reason for auto txn retry - retry_count - rows_written - rows_read In addition, the following fields have been moved from recording at the stmt level to recording at the txn level for insights: - user - application_name Release note: None 93160: bazci: insert `--config ci` if necessary r=healthy-pod a=rickystewart All builds and tests in CI need this --config argument. Epic: None Release note: None 93177: dev: fix dev build error when cross building r=rickystewart a=healthy-pod This is a temporary fix for the issue. In a future change, we should let beaver hub distinguish between normal and cross builds, and then have an actual fix for this. Release note: None Epic: none 93193: vendor: bump Pebble to 0d6d19018632 r=nicktrav a=itsbilal ``` 0d6d1901 crossversion: don't stream TestMeta output on verbose d05a6f0e vfs: use SequentialReadsOption in vfs.[Limited]Copy b85fc64f merging_iter: don't relative-seek past prefix in isNextEntryDeleted 32ad55f8 *: use github.com/cockroachdb/datadriven 6b644274 sstable: don't fatal if file no longer exists during readahead ``` Fixes #93191. Release note: None. Co-authored-by: Andrii Vorobiov <and.vorobiov@gmail.com> Co-authored-by: Tobias Grieger <tobias.b.grieger@gmail.com> Co-authored-by: Rafi Shamim <rafi@cockroachlabs.com> Co-authored-by: Andrei Matei <andrei@cockroachlabs.com> Co-authored-by: Xin Hao Zhang <xzhang@cockroachlabs.com> Co-authored-by: Ricky Stewart <rickybstewart@gmail.com> Co-authored-by: healthy-pod <ahmad@cockroachlabs.com> Co-authored-by: Bilal Akhtar <bilal@cockroachlabs.com>

insights: record missing txn insights #93076

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions