sql: vectorized engine incorrectly formats interval output #83094
Assignees
Labels
C-bug
Code not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
T-sql-queries
SQL Queries Team
Comments
rytaft
added
the
C-bug
yuzefovich
added a commit
to yuzefovich/cockroach
that referenced
this issue
Jul 7, 2022
This commit adds the native cast from strings to ints, floats, and timestamps. I was inspired to do this because the combination of this commit and the vectorized rendering on top of the wrapped row-by-row processors would expose some bugs (e.g. cockroachdb#83094). Release note: None
This was referenced Jul 7, 2022
yuzefovich
added a commit
to yuzefovich/cockroach
that referenced
this issue
Jul 8, 2022
This commit adds the native casts from strings to all remaining natively-supported types (dates, decimals, floats, ints, intervals, timestamps, jsons). I was inspired to do this because the combination of this commit and the vectorized rendering on top of the wrapped row-by-row processors would expose some bugs (e.g. cockroachdb#83094). Release note: None
craig bot
pushed a commit
that referenced
this issue
Jul 11, 2022
83719: storage: remove dependency to sql/catalog/bootstrap r=Xiang-Gu a=Xiang-Gu Previously, tests in `pkg/storage` depended on `sql/catalog/bootstrap`. This was inadequate/weird because `storage` is a much lower layer in the architectural stack. It will also help prevent dependency cycles in other PRs when we introduce depencies (see #82172 if interested). Release note: None 83958: colexecbase: add all remaining casts from strings r=yuzefovich a=yuzefovich **tree: minor cleanup** This commit does a few minor things: - actually uses the error in a few places when constructing a ParseError - refactors some of the interval-parsing functions to expose them to be used in the follow-up commit - extracts a helper method to construct an error when timestamp exceeds bounds. Release note: None **colexecbase: sort native cast info lexicographically** This commit sorts the information about natively supported casts lexicographically so that it is easier to see what is actually supported. This is simply a mechanical change. Release note: None **colexecbase: add all remaining casts from strings** This commit adds the native casts from strings to all remaining natively-supported types (dates, decimals, floats, ints, intervals, timestamps, jsons). I was inspired to do this because the combination of this commit and the vectorized rendering on top of the wrapped row-by-row processors would expose some bugs (e.g. #83094). Addresses: #48135. Release note: None 83984: storageccl: use NewPebbleIterator in restore data processor r=erikgrinaker a=msbutler This PR replaces the multiIterator used in the restore data processor with the PebbleSSTIterator, which has baked in range key support. This patch is apart of a larger effort to teach backup and restore about MVCC bulk operations. Next, the readAsOfIterator will need to learn how to deal with range keys. Informs #71155 Release note: none 84049: sql: fix memory accounting of prepared statements and portals in error cases r=yuzefovich a=yuzefovich **sql: make sure to close mem acc of prepared stmt in case of an error** Previously, it was possible that we would not close the memory account created for a prepared statement when an error is encountered. This was the case because we would not include the prepared stmt into the prep stmts namespace, so it would just get lost. However, up until recently this was not an issue since we mistakenly cleared that memory account when creating the prepared statement. Release note: None **sql: only increment ref count of prep stmt in non-error case of portals** Previously, it was possible to "leak" a reference to a prepared statement if we made a portal from it (i.e. took a reference to the prepared statement) and the memory reservation was denied. This could lead to "leftover bytes" errors when stopping the "session" monitors. However, the impact is minor because on release builds we'd still return those "leftover bytes" and would just file a sentry issue. This is now fixed. Fixes: #83935 Release note: None 84097: DOC-4899: Remove linking on subdirectory from show_backup diagram r=RichardJCai a=nickvigilante Release note: None 84143: Revert "kvstreamer: reuse incomplete Get requests on resume batches" r=yuzefovich a=yuzefovich This reverts commit 21f2390. Previously, I didn't realize that the KV layer would modify all requests included into the BatchRequest in `txnSeqNumAllocator`, so we cannot reuse even incomplete GetRequests. It is unfortunate, but not a big deal. Fixes: #83974. Release note: None 84169: opt: fix incorrect column indexing in index recommendations r=mgartner a=mgartner #### opt: fix incorrect column indexing in index recommendations Fixes #83965 Release note (bug fix): A minor bug has been fixed that caused internal errors and poor index recommendations when running `EXPLAIN` statements. #### opt: clarify logic in Metadata.UpdateTableMeta Release note: None 84188: roachtest: update supported tag for gorm r=ZhouXing19 a=ZhouXing19 fixes #83794 fixes #83797 fixes #83885 Release note: None Co-authored-by: Xiang Gu <[email protected]> Co-authored-by: Yahor Yuzefovich <[email protected]> Co-authored-by: Michael Butler <[email protected]> Co-authored-by: Nick Vigilante <[email protected]> Co-authored-by: Marcus Gartner <[email protected]> Co-authored-by: Jane Xing <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Consider the following difference between the vectorized engine and the row engine:
The result with
vectorize=offis what is also returned by Postgres.Jira issue: CRDB-16860
The text was updated successfully, but these errors were encountered: