What happened
PR #3424 (merged Jul 31) introduced default rejection of past --effective-time values as a security hardening measure. Both benchmarks (simple and stress) pin --effective-time 2024-12-10T00:00:00Z for reproducibility and needed the new --allow-past-effective-time flag.
PR #3490 fixed the simple benchmark (merged Aug 12 12:26 UTC). PR #3426 was supposed to add the flag to the stress benchmark (its diff clearly shows adding --allow-past-effective-time at line 168 of benchmark/stress/stress.go), but it merged 2 minutes later at 12:28 UTC and the line was lost during the merge — the final state on main at lines 166–168 of benchmark/stress/stress.go shows --effective-time, 2024-12-10T00:00:00Z, and immediately closes the argument slice with no --allow-past-effective-time.
The review agent on PR #3490 approved correctly — the diff it reviewed was valid. PR #3490's body stated that the stress benchmark already had the flag via EC-1818, which was accurate for #3426's diff but not for what actually landed on main.
What could go better
The line was lost during the merge of PR #3426, likely due to a rebase conflict resolution when #3490 merged moments earlier and touched a neighboring file area. This is a standard merge-ordering hazard for PRs modifying related code that land in quick succession.
Confidence: High. I directly verified that benchmark/stress/stress.go on main at lines 155–172 does not contain --allow-past-effective-time, while PR #3426's diff shows it being added. The flag is present in benchmark/simple/simple.go (from PR #3490) and in 5 other locations across the codebase, confirming it is the expected pattern.
The stress benchmark will fail at runtime with: "effective time ... is in the past; use --allow-past-effective-time to override."
Proposed change
Add "--allow-past-effective-time", after line 167 ("2024-12-10T00:00:00Z",) in benchmark/stress/stress.go, mirroring the fix already applied to benchmark/simple/simple.go by PR #3490.
This is a one-line fix identical in nature to the change in PR #3490.
Validation criteria
- After the fix,
grep -rn 'allow-past-effective-time' benchmark/ shows the flag present in both simple/simple.go and stress/stress.go.
- The stress benchmark runs to completion without the "effective time is in the past" error.
- Every call site in
benchmark/ that passes --effective-time with a past date also passes --allow-past-effective-time.
Generated by retro agent from #3490
What happened
PR #3424 (merged Jul 31) introduced default rejection of past
--effective-timevalues as a security hardening measure. Both benchmarks (simpleandstress) pin--effective-time 2024-12-10T00:00:00Zfor reproducibility and needed the new--allow-past-effective-timeflag.PR #3490 fixed the simple benchmark (merged Aug 12 12:26 UTC). PR #3426 was supposed to add the flag to the stress benchmark (its diff clearly shows adding
--allow-past-effective-timeat line 168 ofbenchmark/stress/stress.go), but it merged 2 minutes later at 12:28 UTC and the line was lost during the merge — the final state onmainat lines 166–168 ofbenchmark/stress/stress.goshows--effective-time,2024-12-10T00:00:00Z, and immediately closes the argument slice with no--allow-past-effective-time.The review agent on PR #3490 approved correctly — the diff it reviewed was valid. PR #3490's body stated that the stress benchmark already had the flag via EC-1818, which was accurate for #3426's diff but not for what actually landed on
main.What could go better
The line was lost during the merge of PR #3426, likely due to a rebase conflict resolution when #3490 merged moments earlier and touched a neighboring file area. This is a standard merge-ordering hazard for PRs modifying related code that land in quick succession.
Confidence: High. I directly verified that
benchmark/stress/stress.goonmainat lines 155–172 does not contain--allow-past-effective-time, while PR #3426's diff shows it being added. The flag is present inbenchmark/simple/simple.go(from PR #3490) and in 5 other locations across the codebase, confirming it is the expected pattern.The stress benchmark will fail at runtime with: "effective time ... is in the past; use --allow-past-effective-time to override."
Proposed change
Add
"--allow-past-effective-time",after line 167 ("2024-12-10T00:00:00Z",) inbenchmark/stress/stress.go, mirroring the fix already applied tobenchmark/simple/simple.goby PR #3490.This is a one-line fix identical in nature to the change in PR #3490.
Validation criteria
grep -rn 'allow-past-effective-time' benchmark/shows the flag present in bothsimple/simple.goandstress/stress.go.benchmark/that passes--effective-timewith a past date also passes--allow-past-effective-time.Generated by retro agent from #3490