Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Browse files
Browse the repository at this point in the history
…85329 84975: storage: add `MVCCRangeKeyStack` for range keys r=nicktrav,jbowens a=erikgrinaker **storage: add `MVCCRangeKeyStack` for range keys** This patch adds `MVCCRangeKeyStack` and `MVCCRangeKeyVersion`, a new range key representation that will be returned by `SimpleMVCCIterator`. It is more compact, for efficiency, and comes with a set of convenience methods to simplify common range key processing. Resolves #83895. Release note: None **storage: return `MVCCRangeKeyStack` from `SimpleMVCCIterator`** This patch changes `SimpleMVCCIterator.RangeKeys()` to return `MVCCRangeKeyStack` instead of `[]MVCCRangeKeyValue`. Callers have not been migrated to properly make use of this -- instead, they call `AsRangeKeyValues()` and construct and use the old data structure. The MVCC range tombstones tech note is also updated to reflect this. Release note: None **storage: migrate MVCC code to `MVCCRangeKeyStack`** Release note: None ***: migrate higher-level code to `MVCCRangeKeyStack`** Release note: None **kvserver/gc: partially migrate to `MVCCRangeKeyStack`** Some parts require invasive changes to MVCC stats helpers. These will shortly be consolidated with other MVCC stats logic elsewhere, so the existing logic is retained for now by using `AsRangeKeyValues()`. Release note: None **storage: remove `FirstRangeKeyAbove()` and `HasRangeKeyBetween()`** Release note: None 85017: Revert "sql: Add database ID to sampled query log" r=THardy98 a=THardy98 Reverts: #84195 This reverts commit 307817e. Removes the DatabaseID field from the `SampledQuery` telemetry log due to the potential of indefinite blocking in the case of a lease acquisition failure. Protobuf field not reserved as no official build was released with these changes yet. Release note (sql change): Removes the DatabaseID field from the `SampledQuery` telemetry log due to the potential of indefinite blocking in the case of a lease acquisition failure. 85024: cloud/gcp: add custom retryer for gcs storage, retry on stream INTERNAL_ERROR r=rhu713 a=rhu713 Currently, errors like `stream error: stream ID <x>; INTERNAL_ERROR; received from peer` are not being retried. Create a custom retryer to retry these errors as suggested by: googleapis/google-cloud-go#3735 googleapis/google-cloud-go#784 Fixes: #85217, #85216, #85204, #84162 Release note: None 85069: optbuilder: handle unnest returning a tuple r=DrewKimball a=DrewKimball Currently, the return types of SRFs that return multiple columns are represented as tuples with labels. The tuple labels are used to decide whether or not to create a single output column for the SRF, or multiple. The `unnest` function can return a single column if it has a single argument, and the type of that column can be a tuple with labels. This could cause the old logic to mistakenly create multiple output columns for `unnest`, which could lead to panics down the line and incorrect behavior otherwise. This commit adds a special case for `unnest` in the `optbuilder` to only expand tuple return types if there is more than one argument (implying more than one output column). Other SRFs do not have the same problem because they either always return the same number of columns, cannot return tuples, or both. Fixes #58438 Release note (bug fix): Fixed a bug existing since release 20.1 that could cause a panic in rare cases when the unnest function was used with a tuple return type. 85100: opt: perf improvements for large queries r=DrewKimball a=DrewKimball **opt: add bench test for slow queries** This commit adds two slow-planning queries pulled from #64793 to be used in benchmarking the optimizer. In addition, the `ReorderJoinsLimit` has been set to the default 8 for benchmarking tests. **opt: add struct for tracking column equivalence sets** Previously, the `JoinOrderBuilder` would construct a `FuncDepSet` from scratch on each call to `addJoins` in order to eliminate redundant join filters. This led to unnecessary large allocations because `addJoins` is called an exponential number of times in query size. This commit adds a struct `EquivSet` that efficiently stores equivalence relations as `ColSets` in a slice. Rather than being constructed on each call to `addJoins`, a `Reset` method is called that maintains slice memory. In the future, `EquivSet` can be used to handle equivalencies within `FuncDepSet` structs as well. This well avoid a significant number of allocations in cases with many equivalent columns, as outlined in #83963. **opt: avoid usage of FastIntMap in optimizer hot paths** Previously, `computeHashJoinCost` would use a `FastIntMap` to represent join equality filters to pass to `computeFiltersCost`. In addition, `GenerateMergeJoins` used a `FastIntMap` to look up columns among its join equality columns. This lead to unnecessary allocations since column IDs are often large enough to exceed the small field of `FastIntMap`. This commit modifies `computeFiltersCost` to take an anonymous function that is used to decide whether to skip an equality condition, removing the need for a mapping between columns. This commit also refactors `GenerateMergeJoins` to simply perform a linear scan of its equality columns; this avoids the allocation issue, and should be fast in practice because the number of equalities will not generally be large. Release note: None 85146: [backupccl] Use Expr for backup's Detached and Revision History options r=benbardin a=benbardin This will allow us to set them to null, which will be helpful for ALTER commands. Release note: None 85234: dev: add rewritable paths for norm tests r=mgartner a=mgartner Tests in `pkg/sql/opt/norm` are similar to tests in `pkg/sql/opt/xform` and `pkg/sql/opt/memo` in that they rely on fixtures in `pkg/sql/opt/testutils/opttester/testfixtures`. This commit adds these fixtures as rewritable paths for norm tests so that `./dev test pkg/sql/opt/xform --rewrite` does not fail with errors like: open pkg/sql/opt/testutils/opttester/testfixtures/tpcc_schema: operation not permitted Release note: None 85325: sql: fix explain gist output to show number of scan span constraints r=cucaroach a=cucaroach If there were span constraints we would always print 1, need to actually append them to get the count right. Fixes: #85324 Release note: None 85327: sql: fix udf logic test r=chengxiong-ruan a=chengxiong-ruan Fixes: #85303 Release note: None 85329: colexec: fix recent concat fix r=yuzefovich a=yuzefovich The recent fix of the Concat operator in the vectorized engine doesn't handle the array concatenation correctly and this is now fixed. Fixes: #85295. Release note: None Co-authored-by: Erik Grinaker <[email protected]> Co-authored-by: Thomas Hardy <[email protected]> Co-authored-by: Rui Hu <[email protected]> Co-authored-by: DrewKimball <[email protected]> Co-authored-by: Andrew Kimball <[email protected]> Co-authored-by: Ben Bardin <[email protected]> Co-authored-by: Marcus Gartner <[email protected]> Co-authored-by: Tommy Reilly <[email protected]> Co-authored-by: Chengxiong Ruan <[email protected]> Co-authored-by: Yahor Yuzefovich <[email protected]>
- Loading branch information
