[LeftHandCold]

What type of PR is this?

• API-change
• BUG
• Improvement
• Documentation
• Feature
• Test and CI
• Code Refactoring

Which issue(s) this PR fixes:

Fixes #31005.

What this PR does / why we need it:

Fixed MIN/MAX capacity preflight repeatedly simulates metadata winners even when ordinary input and destination vectors cannot create metadata sidecars. Skip that simulation only when both sides have default prepare/string-source metadata and nil sidecars. Retain input bounds, original malformed-target errors, active/standby target checks and the existing fallback for any metadata-bearing vector. BatchFill, merge, capacity admission, spill, cleanup and value comparisons are unchanged. The 39-line production change adds no allocation or retained state.

On original TSBS CPU-only data (8.64 million measurements, 1,000 tags), A/B/B/A with ten measured samples per version/query gives:

Query	Baseline median	Candidate median	Whole-unit CPU change
Q1 hourly MAX	298.39 ms	225.25 ms	-30.72%
Q5 minute MAX	355.63 ms	239.70 ms	-34.73%

All five original queries retain complete source-checked outputs and identical physical objects. Q2/Q3/Q4 control wall changes are -1.03%/+15.60%/+3.65%, with CPU +0.47%/+2.68%/-0.02%. Q3 is 39.01→45.10 ms with intermittent 63–80 ms samples in both versions; this run does not establish that every control is regression-free. Raw samples are retained and Q3 remains a control in the combined-patch validation. Whole engine/client/validator resources are four CPU cores and 12 GiB without swap. Whole-block memory peaks are 1.70–1.88 GiB, with zero OOM/limit events. This causal A/B isolates only the preflight patch on b78794175e65; it does not combine DateTrunc or Top-1 gains.

Validation:

• Baseline and candidate pass 16 targeted tests covering real fill results, NULL/constant/offset/256-row boundaries, malformed target shape with unchanged error/no-mutation behavior, ordinary seen metadata, active/standby state, exhausted budgets, target-only metadata and spill decoding. A deliberately disabled metadata guard fails the capacity assertion.
• Complete owning aggexec package and dependent Group capacity/spill/cancellation/reuse tests pass. Scoped vet, configured incremental golangci-lint, formatting, diff and native-provenance checks pass on the causal candidate.
• Thirteen paired micro controls: ordinary preflight -79.2% to -92.1%; preflight plus fill -58.1% to -69.0%; metadata fallback -0.83%/-0.67%; unchanged zero allocations.
• Delivery commit 08cf92dcbb55 is based on verified main 1651eec948; all four changed source/test files are byte-identical to the validated causal candidate. The full owning package, seven current-main Group capacity/spill/cancellation/reuse consumers, scoped vet/lint, formatting/diff and native-provenance checks pass on this clean delivery head.

This is a draft for review. CPU-only results establish this improvement; broader benchmark and combined-patch validation remain separate work.

[qodo-code-review]

Qodo reviews are paused for this user.

Troubleshooting steps vary by plan Learn more →

On a Teams plan?
Reviews resume once this user has a paid seat and their Git account is linked in Qodo.
Link Git account →

Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center?
These require an Enterprise plan - Contact us
Contact us →

[LeftHandCold]

代码复核：建议通过，未发现需要 Request Changes 的确定性问题。审查 HEAD：08cf92dcbb550611677f9e835b845c5028d758e9。

设计判断：在现有容量预检 owner 内消除普通值的重复元数据模拟；没有新增容量账本，也没有绕过 GroupGrow/BatchFill/spill 的既有责任。

逐项核对：

• 快路径同时要求输入与每个实际触及的目标没有非默认 prepare/string-source 元数据，且 sidecar 为 nil；“非 nil 但长度为 0”的 sidecar 不会误入快路径。
• 原输入范围/常量广播、GroupNotMatched、NULL、active/standby 容量检查保留。目标形状异常或任一目标带元数据会回退原预检。
• 对照了实际 BatchFill 的 metadata setter：普通值对普通目标不会创建这些 sidecar；不是仅检查输入就跳过目标的潜在分配。
• 新测试覆盖拒绝分配后的不发布、目标-only 元数据、后续 chunk、空 sidecar、standby、spill/merge 和消费侧容量边界。

非阻断的性能说明：作者报告 Q3 对照 wall time 39.01→45.10ms（+15.60%），CPU +2.68%，两边均有慢样本。现有证据不足以认定本 PR 引入稳定性能回归，也不足以宣称所有对照完全无退化。组合版本验证应保留 Q3 对照，不能把单补丁收益直接相加；不因此添加新的 merge blocker。

验证边界：当前 HEAD 的 CI 40034727403 已成功；diff --check 通过。本次未重跑重型 UT/TSBS，性能数字是对作者报告的复核而非独立复测。

当前认证账号也是 PR 作者，故以 COMMENT 记录建议通过；这不是 GitHub 正式 Approval。