fix: fallback UPDATE planner produces nondeterministic project layout - #24794
Merged
mergify[bot] merged 6 commits intoJun 3, 2026
Conversation
Qodo reviews are paused for this user.Troubleshooting steps vary by plan Learn more → On a Teams plan? Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center? |
Qodo reviews are paused for this user.Troubleshooting steps vary by plan Learn more → On a Teams plan? Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center? |
XuPeng-SH
approved these changes
Jun 3, 2026
XuPeng-SH
left a comment
XuPeng-SH
left a comment
Contributor
There was a problem hiding this comment.
LGTM after reviewing the current head from 3 angles.
- Logic/correctness: the fix removes both nondeterministic table-block ordering and nondeterministic per-column block ordering in fallback UPDATE planning, and the downstream consumers bind updated values by column name via updateColPosMap rather than depending on the prior random layout.
- Behavior/compatibility: this is a layout-stabilization change, not a semantic rewrite of UPDATE results; it makes planner output deterministic across builds without changing the intended column-to-value mapping.
- Tests: the new deterministic planner unit test directly covers the reported regression path. A SQL-level BVT could be a nice follow-up hardening, but I do not see a blocking gap for this fix.
Approve.
Contributor
Merge Queue Status
This pull request spent 1 hour 6 minutes 14 seconds in the queue, including 1 hour 4 minutes 30 seconds running CI. Required conditions to merge
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What type of PR is this?
Which issue(s) this PR fixes:
issue ##24718
What this PR does / why we need it:
Root cause
In pkg/sql/plan/build_constraint_util.go inside getUpdateTableInfo (around line 218):
// remove unused table
newTblInfo := &dmlTableInfo{...}
for alias, columns := range usedTbl { // ← non-deterministic map iteration
idx := tblInfo.alias[alias]
...
newTblInfo.tableDefs = append(newTblInfo.tableDefs, tblDef)
newTblInfo.alias[alias] = len(newTblInfo.tableDefs) - 1
...
}
usedTbl is a map[string]map[string]tree.Expr populated from the SET list. For UPDATE emp, dept SET emp.comm=1, dept.loc=... it has two keys ("emp", "dept"). Iterating it with range randomizes
the order per Go map semantics, so newTblInfo.tableDefs ends up as either [emp, dept] or [dept, emp].
The downstream fallback UPDATE builder (selectUpdateTables / rewriteGeneratedColumnsForUpdate in build_update.go) faithfully follows whatever order it receives — placing each target's columns,
SET expressions, and recomputed generated-column projections in a contiguous block. The plan it produces is internally correct in both orderings.
The test TestUpdateFallbackGeneratedColumnsUseOnUpdateAfterRewrite then asserts a fixed position in the project list:
generatedExpr := requireFallbackSourceProjectExpr(t, logicPlan.GetQuery(),
len(...emp.Cols)+2+len(...dept.Cols)+1, // total projectLen=16
len(...emp.Cols)+1, "on-update-marker") // pos=10 (expects emp first)
Position 10 holds the recomputed emp.ename only when emp is the first block. When the map iteration yields [dept, emp], the emp block lives at offsets 5..14, the recomputed ename lands at index
15, and position 10 is emp.sal — exactly what the failure trace shows.
Verdict
Real underlying bug in upstream main (introduced by 7ff125d): newTblInfo.tableDefs ordering depends on Go map iteration randomness. Should iterate tblInfo.tableDefs in index order (or sort
by tblInfo.alias[alias]) instead of ranging over usedTbl.
The new test added in the same commit only happens to pass most of the time because [emp, dept] is randomly chosen ~50%.
Not related to your gpu_plugin_all branch — neither getUpdateTableInfo nor the test exists there independently; both came in from main via merge.
Suggested upstream fix: replace the range usedTbl loop with an iteration over tblInfo.tableDefs by index, looking up usedTbl[alias] per table to preserve source order.