Bug triage results: 2026-04-27

[andygrove]

Triage summary for 2026-04-27

Triaged 19 open issues that carried requires-triage. Labels have already been applied; please spot-check below and close this issue when satisfied. Any individual relabel can be done directly on the affected issue.

Triage criteria come from docs/source/contributor-guide/bug_triage.md. Most issues in this batch are enhancement requests rather than bugs; the priority decision tree is bug-focused, so enhancement priorities reflect judgment about user impact and urgency relative to the priority descriptions in the guide (low = tooling/cosmetic/process; medium = functional gap with workaround; high = crash/major breakage; critical = silent wrong results).

Counts by priority applied

Priority	Count
priority:critical	0
priority:high	1
priority:medium	6
priority:low	12

good first issue applied: 3.

Triaged

Issue	Priority	Areas	GFI	Rationale
#4005	medium	(none)	no	EPIC tracking broad planner improvements; cross-cutting impact on perf/correctness puts this above tooling but it is umbrella scope, so not high.
#4006	low	(none)	no	Internal API enhancement to distinguish info vs fallback messages; no user-visible bug, fits "tooling" tier.
#4007	medium	area:aggregation	no	Functional gap with possible correctness issues, but windowed aggregates are disabled by default so users are not silently affected. See escalation note.
#4020	low	area:scan, native_iceberg_compat	no	Internal refactor / cleanup; no user-visible bug.
#4027	low	(existing spark 4)	no	Forward-port to next Spark 4.0.x patch; routine dependency bump.
#4045	medium	area:scan, area:shuffle	no	Functional bug: Comet+vanilla mismatch breaks ReusedExchangeExec under AQE+DPP. Has workaround (disable AQE), fits medium.
#4050	low	(none)	no	Process / tooling enhancement for audit reporting.
#4054	low	(existing spark 4)	no	Build-profile default change discussion; no user-visible bug.
#4059	low	area:ci, spark 4	yes	Add Java 21 row to CI matrix; well-scoped YAML change, contributor docs update.
#4069	low	area:expressions	no	Exploratory: try DataFusion lambda functions in Comet. Tooling/research scope.
#4074	medium	area:expressions	no	Mix of expression support-level fixes. One sub-item (CometStringRepeat does not fall back for incompatibility) is potentially correctness; see escalation note.
#4081	low	(none)	no	Release planning for 0.16.0; tracking/process.
#4094	low	(none)	no	Release planning for 0.15.1; tracking/process.
#4096	medium	spark 4	no	Functional gap: Spark 4.1 introduces OneRowRelationExec causing ~30 sql-file test fallbacks. Workaround = run on Spark 4.0.
#4098	high	spark 4	no	Tracking issue for four Spark 4.1 failure clusters; includes potential silent-wrong-results bloom filter mismatch. See escalation note.
#4099	low	(none)	no	Cosmetic / branding update; fits the "cosmetic" tier of priority:low.
#4101	low	area:ci, area:Iceberg	yes	CI cost optimization; scoped YAML matrix edit.
#4102	low	area:ci	yes	CI cost optimization; scoped YAML matrix edit.
#4107	low	spark 4	no	Internal refactor of 4.0/4.1 shims after 4.1 stabilizes; no user impact.

Escalations to consider

• Tracking: remaining Spark 4.1 CI failures on #4093 #4098 (Spark 4.1 CI failures) — One cluster (bloom filter result mismatch) produces different might_contain results between Comet and Spark, i.e. silent wrong results. Per the guide's "correctness over crashes" principle this would normally warrant priority:critical, but Spark 4.1 is not yet a supported target so users are not exposed today. Worth splitting that cluster into its own priority:critical issue once Spark 4.1 support lands.
• Fix issues found when documenting current expression incompatibilities #4074 — The CometStringRepeat sub-item ("incompatibility that we do not fall back for") could produce silent wrong results on the relevant inputs; if confirmed, escalate that sub-item to priority:critical and split into its own issue.
• Enable windowed aggregates and fix correctness issues #4007 — Title says "fix correctness issues"; if the underlying correctness issues are present even when windowed aggregates are explicitly enabled by users, this should escalate to priority:high (or priority:critical if results are silently wrong).