Skip to content

feat(bench): the Bulk and Jobs benches run on Oracle behind a provider switch (G7 Oracle 8/12 Part 1/2, #556) - #560

Merged
omercelikdev merged 6 commits into
mainfrom
agent/556-oracle-bulk-jobs-benches
Nov 12, 2026
Merged

omercelikdev merged 6 commits into
mainfrom
agent/556-oracle-bulk-jobs-benches

Conversation

@omercelikdev

@omercelikdev omercelikdev commented Nov 10, 2026 •

Copy link
Copy Markdown
Collaborator

What and why

Part 1/2 of #411 (G7 Oracle 8/12): the Bulk and Jobs benches can run on Oracle. Tests, scripts and bench.yml only; no package changes.

  • Written once. BulkBenchTests, BulkIntakeBenchTests and JobsBenchTests now take a SuiteStore, the store the Campaign, Bulk and Archival suites already use for both providers.
    • The PostgreSQL class keeps its name and its own container.
    • The Oracle twins (BulkBenchOracleTests, BulkIntakeBenchOracleTests, JobsBenchOracleTests) run on a fresh schema of the Oracle collection's pinned container (OracleFixture.Image).
  • The provider switch. GOLDPATH_BENCH_PROVIDER=postgres|oracle on scripts/bench-bulk.sh and scripts/bench-jobs.sh, with postgres as the default.
    • An Oracle bench line ends in provider=oracle.
    • The scripts refuse an Oracle run with no such line, so an Oracle run cannot silently measure PostgreSQL.
  • bench.yml gains a provider dispatch input. oracle runs Bulk (intake at 1,000,000 rows) and Jobs on Oracle, and the step fails if either script did. The monthly schedule stays PostgreSQL-only (Category=Bench&Category!=Oracle).
  • Budgets bind PostgreSQL only. On Oracle the facts assert correctness and print their numbers, as the oracle RFC §6 says ("measured, not promised").

Acceptance criteria of #556, each with its evidence

  • Both scripts take a provider choice, default PostgreSQL; with Oracle the same facts run on OracleFixture's image; PostgreSQL output is unchanged.

    • The PostgreSQL line format is byte-identical: BENCH-BULK ingest10k upload=0.71s total=1.89s (budget 300s) from tonight's run.
    • bench-jobs.sh on PostgreSQL now filters Category!=Oracle, so it runs exactly the set it ran before.
  • No test that is both Bench and Oracle runs in the per-PR lanes. Both lanes' real filters were run against the built assembly, narrowed to the twins (&FullyQualifiedName~BenchOracleTests), and each answered "No test matches": Category=Oracle&Category!=Bench (the oracle lane) and Category!=Bench&Category!=Oracle (the integration lane). So did Category=Bench&Category!=Oracle (bench.yml's PostgreSQL filter).

  • Bulk: the 1M intake and the 10k/100k facts on Oracle, with numbers: NOT MET. Every Bulk fact fails on Oracle on its first save with ORA-01460. That is a product defect, filed as Bulk on Oracle: any upload over 2,000 bytes fails with ORA-01460, because the file chunk column is RAW(2000) and the store writes 256 KB chunks #559:

  • Jobs: every JobsBenchTests fact has an Oracle column in jobs-benchmarks.md. These are three Oracle runs and one PostgreSQL run on the same sandbox, the same night:

    Proof PostgreSQL 17 Oracle Free 23.26.3
    100k no-op run, plan → complete 1.5 s 1.8 / 1.7 / 1.7 s
    Checkpoint cost per chunk 7.6 ms 9.2 / 8.3 / 8.4 ms
    Interactive p95 under a full-tilt job 1.5 → 3.4 ms 1.5 → 5.1 / 3.3 → 7.5 / 1.2 → 6.9 ms

    Nothing is an order of magnitude apart.

  • Every Oracle number states its profile (machine, image tag, build configuration) and date. No budget is promised.

  • bench.yml dispatch input. It is Tier C (.github/**).

  • Ledgers: CHANGELOG [Unreleased]; ga-gate-status.md G7 names the two pages; open-threads T68. coverage-matrix.md has no Bench column, so it is unchanged. docs/rfc/goldpath-ga-readiness.md is not edited.

Reproduce

git fetch origin && git checkout agent/556-oracle-bulk-jobs-benches
# green: Jobs on Oracle (Docker; about 30 s)
GOLDPATH_BENCH_PROVIDER=oracle scripts/bench-jobs.sh          # two BENCH lines ending in provider=oracle, exit 0
# the lanes never run a twin (each: "No test matches")
dotnet build tests/Goldpath.IntegrationTests -c Release
for f in "Category=Oracle&Category!=Bench" "Category!=Bench&Category!=Oracle" "Category=Bench&Category!=Oracle"; do
  dotnet test tests/Goldpath.IntegrationTests -c Release --no-build --filter "$f&FullyQualifiedName~BenchOracleTests"; done
# RED (the issue's proof): the switch says Oracle, the wiring says PostgreSQL. The script must fail, not measure PostgreSQL:
sed -i 's/: BulkBenchTests(SuiteStore.Oracle(oracle, "GP_BBENCH"));/: BulkBenchTests(oracle is null ? null! : SuiteStore.PostgreSql());/' tests/Goldpath.IntegrationTests/BulkBenchTests.cs
GOLDPATH_BENCH_PROVIDER=oracle scripts/bench-bulk.sh card     # prints untagged PostgreSQL lines, then
# "bench-bulk: GOLDPATH_BENCH_PROVIDER=oracle, but the run printed no Oracle bench line; nothing was measured on Oracle", exit 1
git checkout tests/Goldpath.IntegrationTests/BulkBenchTests.cs
# #559 as found: Bulk on Oracle fails on its first save
GOLDPATH_BENCH_PROVIDER=oracle scripts/bench-bulk.sh card     # 2 failed: ORA-01460 at GoldpathBulkFileStore.SaveAsync

Gates run

Gate Result
dotnet build tests/Goldpath.IntegrationTests -c Release 0 warnings, 0 errors
dotnet format whitespace --verify-no-changes on the four touched test files clean
docs-freshness, ledger-check, schema-honesty, skills-parity, template-pins rc 0 ×5
Oracle Jobs bench green ×2 after the budget guard (the first run failed only the PostgreSQL budget it no longer asserts on Oracle)
PostgreSQL Jobs bench, same sandbox Bench_100k… green. Bench_interactive… misses its own budget (1.5 → 3.4 ms against 1.25× + 1 ms) on this 4-CPU co-located profile. Its PostgreSQL code is unchanged by this PR; the page says so.

Pre-PR self-check

  • Acceptance criteria: listed above with evidence. The Bulk numbers are not met (Bulk on Oracle: any upload over 2,000 bytes fails with ORA-01460, because the file chunk column is RAW(2000) and the store writes 256 KB chunks #559, T68), and that is said plainly.
  • Ledgers: CHANGELOG, G7, T68 (T65 and T66 are taken by open feat(gates): the surface freeze gate — CLI verbs and flags, template parameters, manifest schema, additions-only (G4 API-10, #399) #552, T67 by open feat(campaign): a tick's releases go out as one batch through an internal seam member (#285, throughput Part 3/3) #558), both bench pages.
  • XML docs: each new class and constructor is documented. The new members sit before the existing ones' /// blocks, never between a block and its declaration. Re-read.
  • Honesty:
    • The Bulk page publishes no number, and says why.
    • The Jobs page says the PostgreSQL interactive budget also fails on this profile.
    • "PostgreSQL output unchanged" was checked against the line from tonight's run.
  • Wired:
    • The scripts' Oracle path runs the twins on the pinned image, proven by the provider=oracle lines.
    • The refusal is proven red by the plant above.
    • bench.yml's filters: the PostgreSQL step's filter answered "No test matches" for every Oracle twin.
  • Contracts: no package, schema or API change.
  • Benches: they ran alone, after the mutation and conformance runs of another PR had finished, and nothing ran beside them. The numbers carry profile, date, commit and "no run id".
  • Deep self-review:
    • Security: the workflow gains one choice input, interpolated only into if: expressions, never into a shell line. No permission, secret or action change.
    • Shell: each script captures dotnet test's exit code (set +e around it) and fails on it before the Oracle-line check. An Oracle run with zero bench lines is refused, and so is a run where any line lacks the tag.
    • Edge cases: an unknown provider value exits 2 with a message.
    • Compatibility: the default is PostgreSQL everywhere, and PostgreSQL behaviour and output are unchanged.
    • Code quality: the three benches follow BulkTests/BulkOracleTests' existing pattern, and BenchProvider is the one place the tag is spelled.

Not verified

Part of #411.

Refs #556 — not closed by this PR: Bulk's Oracle numbers are still open under T68 / #559, and #556 closes with them (owner decision, 2026-11-11).


Generated by Claude Code

…r switch (G7 Oracle 8/12 Part 1/2, #556)

- BulkBenchTests, BulkIntakeBenchTests and JobsBenchTests are written once on SuiteStore:
  PostgreSQL in the class itself (output unchanged), Oracle in BulkBenchOracleTests,
  BulkIntakeBenchOracleTests and JobsBenchOracleTests on the Oracle collection's pinned
  container, each carrying Category=Bench AND Category=Oracle so neither the integration lane
  (Category!=Bench&Category!=Oracle) nor the oracle lane (Category=Oracle&Category!=Bench)
  runs them. Oracle bench lines end in `provider=oracle`; the intake's size on Oracle is the
  schema's segments.
- scripts/bench-bulk.sh and scripts/bench-jobs.sh take GOLDPATH_BENCH_PROVIDER
  (postgres|oracle, default postgres), and refuse an Oracle run that printed no
  `provider=oracle` line. bench-jobs.sh on PostgreSQL now filters Category!=Oracle, so it runs
  the set it ran before.
- bench.yml: a `provider` dispatch input; the monthly schedule stays PostgreSQL-only
  (Category=Bench&Category!=Oracle).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…bind PostgreSQL only (#556)

- jobs-benchmarks.md: Oracle Free 23.26.3 beside PostgreSQL 17 on the same sandbox (3 Oracle
  runs; checkpoint 8.3-9.2 ms/chunk vs 7.6 ms; plan->complete 1.7-1.8 s vs 1.5 s). The
  interactive proof misses its own budget on PostgreSQL too on this 4-CPU profile.
- bulk-benchmarks.md: no Oracle number - every upload over 2,000 bytes fails with ORA-01460
  (GoldpathBulkFileChunks.Data is RAW(2000) on Oracle, chunks are 256 KB); #559 filed.
- Bench facts assert their budgets on PostgreSQL only (oracle RFC §6: measured, not promised);
  correctness asserts run on both.
- bench.yml's Oracle step runs both scripts and fails if either did; the summary also keeps
  `BENCH items=` lines.
- CHANGELOG, ga-gate-status G7, open-threads T68.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@omercelikdev omercelikdev added the agent-pr PR opened by the overnight agent label Nov 10, 2026
@github-actions github-actions Bot added the tier:C ADR-0014 Tier C: decisions, releases, gates, policy, agent layer; the owner merges label Nov 10, 2026
…yml runs them (#556)

CI gates on 17eecab: merge-tier selftest red, "gate script not Tier C" for both scripts,
because bench.yml's Oracle step now calls them. Listed beside bench-campaign.sh; selftest
green, docs-freshness and ledger-check rc 0.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@omercelikdev

Copy link
Copy Markdown
Collaborator Author

CI on 17eecab: gates was red, and this PR caused it. The merge-tier selftest requires every script a workflow runs to be Tier C, and bench.yml's Oracle step now runs scripts/bench-bulk.sh and scripts/bench-jobs.sh. The gate reported gate script not Tier C for both.

Fixed in 6eadf36: both scripts are now listed in .github/merge-tiers.yml beside bench-campaign.sh. This only makes the classification stricter. Locally, python3 scripts/merge-tier.py selftest is green, and docs-freshness and ledger-check exit 0. I should have run the selftest before opening; it is in the self-check from now on.


Generated by Claude Code

@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Verifier verdict (run 2026-11-11)

Head verified: 6eadf369f1fd4b6405fe1011f31e65c3b06dc2fe (3 commits). It merges cleanly into today's main@2782d3c. Tier C (.github/workflows/bench.yml, .github/merge-tiers.yml, and the two bench scripts now Tier C), so it is not auto-merged.

Verdict: no blocking findings. 0 BLOCKER, 0 MAJOR, 0 MINOR, 2 NIT. The switch, the refusal, the lane isolation and the Jobs-on-Oracle numbers all reproduce here. #556's Bulk criterion is unmet, which the PR says plainly (#559, T68), but it still Closes #556: scope question below, so needs-owner.

What I checked

Step Result
1. Scope vs #556 AC1 (the provider switch, PostgreSQL default, PostgreSQL output unchanged): ✓. The PostgreSQL lines are untagged (BENCH-BULK ingest10k upload=0.69s total=1.86s (budget 300s)), and the PostgreSQL filter is the same set as before (Category=Bench&Category!=Oracle; the base classes keep the Bench trait). AC2 (no Bench+Oracle test in the per-PR lanes): ✓ (P4). AC3 (Bulk numbers on Oracle): ❌ not met and disclosed. I reproduced it: both card facts fail with ORA-01460: unimplemented or unreasonable conversion requested (#559 is open). AC4 (the Jobs Oracle column): ✓. AC5 (a profile on every number): ✓. AC6 (the bench.yml dispatch input): ✓. No package changes, no permission, secret or action change, no test weakened. The Oracle-only budget skip follows the oracle RFC §6 ("measured, not promised").
2. Rules / ledgers CHANGELOG [Unreleased] ✓. G7 row ✓. T68 is unique: main ends at T64, #552 has T65/T66, #558 has T67. T68 has a trigger (#559 merges) and a proof. Both scripts were added to merge-tiers.yml's C list, and merge-tier.py selftest is green (the CI red on 17eecab is fixed). XML docs on each new class and constructor.
3. Confidentiality The diff, the commit messages and the PR body: 0 hits.
4. Build dotnet build tests/Goldpath.IntegrationTests -c Release on the merged tree: 0 warnings, 0 errors. bench.yml parses (5 steps; the two bench steps are mutually exclusive on inputs.provider, and a schedule run, where the input is empty, takes PostgreSQL).
4. Reproduce GOLDPATH_BENCH_PROVIDER=oracle scripts/bench-jobs.sh (Oracle Free in Docker, 1 m 14 s with the container start): two lines ending in provider=oracle (BENCH items=100000 … checkpointed=1.7s … checkpoint-cost-per-chunk=8.6ms; BENCH-INTERACTIVE baseline-p95=1.7ms under-load-p95=6.3ms), in line with the published 1.7 s, 8.3–9.2 ms and 1.2–3.3 → 5.1–7.5 ms.
4. Red proof (the PR's) With the Oracle twin wired to PostgreSQL, bench-bulk.sh card prints untagged PostgreSQL lines, then bench-bulk: GOLDPATH_BENCH_PROVIDER=oracle, but the run printed no Oracle bench line; nothing was measured on Oracle, and exits 1 ✓. Restored, and the tree is clean.
4. Gates docs-freshness, ledger-check, schema-honesty, skills-parity, template-pins, merge-tier.py selftest: rc 0 ×6.
5. CI on 6eadf36 Every check green except merge-policy, which is ✗ by design (Tier C, no owner-approved).
6. Charter n.a.

Deep review

Dimension Result
a Security ✓ The new input is a choice; GitHub validates a dispatched value against the options. No permission, secret or action change. NIT 2: one interpolation does reach a run: line, contrary to the PR's self-review.
b Correctness & edge cases ✓ An unknown provider exits 2 (P1). An Oracle run with zero bench lines is refused. A dotnet test failure is propagated before the tag check. NIT 1 is the one edge where the check could fail open.
c Concurrency n.a. Each twin takes a fresh schema on the shared Oracle collection's container.
d Data & migrations n.a.
e Performance ✓ The Oracle Jobs run takes about 74 s here, container start included. The monthly schedule is unchanged.
f Compatibility ✓ The PostgreSQL default and output are unchanged. JobsBenchTests stops being sealed, but it is a test class.
g Test quality ✓ The refusal is proven red. The lane filters were run against the built assembly. The budget skip is explicit and limited to _store.IsOracle.
h Code quality ✓ It follows the existing BulkTests/BulkOracleTests twin pattern. BenchProvider is the single place the tag is spelled. The comments say why.
i Observability ✓ The job summary names the provider and now also greps BENCH lines (the Jobs bench's own prefix).
j Honesty ✓ The Bulk page publishes no number and says why (reproduced: ORA-01460). The Jobs page admits the PostgreSQL interactive budget also misses on this profile. The two NITs are PR-text and edge-case wording.

Edge-case probes (this run)

# Probe Result
P1 GOLDPATH_BENCH_PROVIDER=mysql scripts/bench-jobs.sh must be postgres or oracle, not 'mysql', exit 2. Pass
P2 The PR's plant: the Oracle twin measures PostgreSQL Refused, exit 1. Pass
P3 The bulk tag check, lifted verbatim into a set -euo pipefail script and fed an output whose first BENCH-BULK line is untagged, followed by 200,000 tagged lines ACCEPTED ×3: it fails open (NIT 1). With 2 lines it is REFUSED, as it should be.
P4 The three lane filters (Category=Oracle&Category!=Bench, Category!=Bench&Category!=Oracle, Category=Bench&Category!=Oracle), each narrowed to ~BenchOracleTests "No test matches" ×3. The twins exist: 5 listed under Category=Bench&Category=Oracle. Pass
P5 Bulk on Oracle, bench-bulk.sh card 2 failed, ORA-01460 ×2, exit 1: #559 as the PR states. Pass (the claim holds)

NITs (optional)

  1. The tag check can fail open under pipefail. grep -a "BENCH-BULK" "$OUT" | grep -avq "provider=oracle" (in scripts/bench-bulk.sh, and the same shape in bench-jobs.sh):
    • grep -q exits at its first match.
    • If the producer is still writing, it dies of SIGPIPE (141). Under set -o pipefail the pipeline is then "false", so an untagged line is accepted (P3).
    • It is unreachable at today's output, which is about 2–4 bench lines and far below a pipe buffer. That is why this is only a NIT.
    • It is a gate's refusal, though, and the fix is one word: drop -q and redirect, grep -a "BENCH-BULK" "$OUT" | grep -av "provider=oracle" >/dev/null, or use ! grep -a … | grep -avc … | grep -qx 0.
  2. The PR body's self-review says the input is "interpolated only into if: expressions, never into a shell line". The summary step does put it in a run: line: echo "## Reference profile …, provider: ${{ inputs.provider || 'postgres' }}".
    • It is harmless, because choice values are validated.
    • The house style for workflow inputs is still env: PROVIDER: ${{ inputs.provider || 'postgres' }} and then $PROVIDER. Either change that, or correct the sentence.

Question for the owner (needs-owner)

Should this PR close #556 although its Bulk criterion (the 1M intake and the 10k/100k facts on Oracle, with numbers) is unmet until #559 lands?

Labels: agent-verified and needs-owner. Tier C, so not auto-merged.

Owner: safe to merge — summary: scripts/bench-bulk.sh and scripts/bench-jobs.sh (and bench.yml by dispatch) can run the Bulk and Jobs benches on Oracle Free, and refuse a run that measured PostgreSQL. Jobs now has Oracle numbers; Bulk on Oracle is blocked by #559 (ORA-01460).


Generated by Claude Code

@omercelikdev omercelikdev added agent-verified Independently verified by the verifier agent: ready for owner merge needs-owner Needs an owner decision; the question is in the last comment labels Nov 11, 2026
@omercelikdev omercelikdev added owner-approved ADR-0014: the owner approves merging this Tier C agent PR and removed needs-owner Needs an owner decision; the question is in the last comment labels Nov 11, 2026
CHANGELOG.md: both sides' [Unreleased] entries kept. open-threads.md: main's
T67 and this branch's T68 both kept, in number order.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018AFdNZbL5UCo9Kkefph8f3
@omercelikdev omercelikdev removed the agent-verified Independently verified by the verifier agent: ready for owner merge label Nov 11, 2026
@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Brought up to date with main@02128c5. The head is now 99a51e3, a merge commit with no rebase and no force-push.

Conflicts resolved in:

ga-gate-status.md auto-merged: main touched G4 and G10c, and this PR touched G7. Diffed against each parent, the merge removes no line of this PR's.

Build, tests and gates:

Check Result
dotnet build tests/Goldpath.IntegrationTests -c Release 0 warnings, 0 errors
The lanes never run a twin: Category=Oracle&Category!=Bench, Category!=Bench&Category!=Oracle and Category=Bench&Category!=Oracle, each narrowed to ~BenchOracleTests "No test matches" ×3
GOLDPATH_BENCH_PROVIDER=oracle scripts/bench-jobs.sh (real Oracle Free container) rc 0, two lines tagged provider=oracle (see below)
docs-freshness (24 packages), schema-honesty, skills-parity, template-pins, merge-tier.py selftest rc 0 each
ledger-check.sh rc 0, but it skipped the GitHub reconciliation because there is no authenticated gh in this session
dotnet format whitespace --verify-no-changes on the PR's .cs files clean

The two provider=oracle lines were 100k / 200 chunks, 8.8 ms per checkpoint, and interactive p95 2.7 → 6.8 ms. They are the same order as the page's numbers. I ran them as a smoke check only, so nothing published changed.

The PR's behaviour and recorded decisions are unchanged by the merge. I removed agent-verified because the verified head changed. Back to the verifier.


Generated by Claude Code

@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Verifier verdict (run 2026-11-11, daytime)

Head verified: 99a51e3b908ec4abff62c783b87f32005006c2bb. My last verdict was on 6eadf36. The delta is one commit: the merge of main@02128c5. Tier C, so this verdict is advice to the owner. Nothing merges it automatically.

This morning's scope question is resolved by the owner. The PR body now says Refs #556 — not closed by this PR … (owner decision, 2026-11-11), and needs-owner was removed at 06:30.

Verdict: no findings in the PR's content. gates is red on this head because of main's defect at 02128c5, so I am not adding agent-verified.

Evidence

  • The PR's own content is identical across the merge. git diff <old-base> 6eadf36 and git diff 02128c5 99a51e3 agree on every file. That includes the PR's own lines in the three resolved documents (CHANGELOG.md, ga-gate-status.md, open-threads.md).
  • The merge kept everything main brought. I checked every line main added between the old base and 02128c5; each one is present in the head (0 missing).
  • Open threads. T67 (main's) and T68 (this PR's) are both present, in number order. No other open PR adds a thread number.
  • Build. dotnet build Goldpath.sln -c Release: 0 warnings, 0 errors.
  • Gates. docs-freshness, ledger-check, schema-honesty, skills-parity and template-pins all PASS. merge-tier.py selftest: green.
  • git merge-tree against today's main@86ab2df: clean.
  • CI on 99a51e3. Everything is green except gates: TickQueryCountTests.One_hungry_campaign_costs_a_fixed_number_of_commands_per_tick (Expected 7, Actual 8). That is the defect main had at 02128c5, which test(campaign): pin 8 commands per steady tick — R3-S8's ceiling read (main red after #529 + #558) #563 (86ab2df) fixed after this merge. I reproduced the same failure on feat(locking): the optional Goldpath.Locking.Oracle package — DBMS_LOCK through DistributedLock.Oracle (G7, O-4, #407) #535's head, which has the same main base. The oracle lane is green on this head.

Finding

  1. MAJOR (CI, a required check red on the head).

Edge-case probes (the provider switch's refusal; read-only runs of the PR's scripts)

# GOLDPATH_BENCH_PROVIDER Expected bench-jobs.sh / bench-bulk.sh
1 Oracle (mixed case) refused ✓ exit 2, "must be postgres or oracle, not 'Oracle'" (both scripts)
2 oracle (leading space) refused ✓ exit 2 (both)
3 ORACLE refused ✓ exit 2 (both)
4 empty the default postgres ✓ the run started (my 20 s timeout stopped it)

Deep review (delta-scoped: merge-only; the full review is in my verdict of 03:53 today)

Dimension Result
a Security n.a.: no code delta
b Correctness & edge cases ✓ probes 1–4, and the merge loses no line
c Concurrency n.a.
d Data & migrations n.a.
e Performance n.a.: bench-only change, content unchanged
f Compatibility n.a.: tests, scripts and docs only
g Test quality ✓ unchanged since my red→green run this morning
h Code quality n.a.: no code delta. This morning's 2 NITs stand as optional
i Observability n.a.
j Honesty ✓ the PR body now matches #556's real state (Refs, not Closes; T68 / #559 hold the Bulk column)

Generated by Claude Code

@omercelikdev omercelikdev added the changes-requested Verifier found problems; the worker fixes these first label Nov 11, 2026
# Conflicts:
#	CHANGELOG.md
#	docs/strategy/ga-gate-status.md
@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Synced with main@440f4a0 (#535, after #563 and #552) in a321d48. This brings #563, which is the fix the daytime verifier asked for: gates was red on TickQueryCountTests only because of main's defect.

Conflicts resolved in:

In CHANGELOG.md and open-threads.md, the PR's own diff is unchanged by the resolution.

These ran locally on the merged tree:

  • ledger-check, docs-freshness, api-review-gate, schema-honesty, skills-parity, template-pins, train-freshness samples/corpay, surface-freeze-gate: rc 0.
  • merge-tier.py selftest: green.
  • Goldpath.sln build with -warnaserror: 0 warnings.

The oracle job can still go red from #571, a main defect. It is not this PR.

Owner's session (Claude Code).

@omercelikdev omercelikdev removed the changes-requested Verifier found problems; the worker fixes these first label Nov 11, 2026
@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Synced with main@c9d23ae (#573, Aspire 13.6.5 for GHSA-f7vh-h5g3-7jrh) in ed7fcb2, with no conflicts.

On a321d48 the required golden manifests (PR subset) check was red only on that advisory: NU1903 on Aspire.Hosting 13.4.6 in GmCleanFeatures, GmEverything, GmGrownRest and GmWorkerInSolution. On #573 all 11 shapes are green.

The merge brings only #573's 9 files, and this PR touches none of them. Its own lines in CHANGELOG.md, ga-gate-status.md and open-threads.md are unchanged. ledger-check, docs-freshness and template-pins give rc 0.

Owner's session (Claude Code).

@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Verifier verdict (run 2026-11-12)

Head verified: ed7fcb21714f7cd0ef0d86c8258c82e6ca40adaf · Base: main @ c9d23ae5020cbd4a87540dd58e9c20253764d52b (merge-base = main head, so git diff origin/main HEAD is exactly the PR's change: 13 files, +290/−60).

Verdict: verified. No blocking findings, 3 optional NITs.

Delta since the last verdict (99a51e3)

Steps 1–6

# Step Evidence Result
1 Scope #556 criteria against the diff. Provider switch, lane exclusion, Jobs column, profile and date on every number, bench.yml dispatch input, ledgers: met. The Bulk numbers are not met and say so (#559, T68). The owner's answer in the body is "Refs #556, not closed". No package, dependency, pin or threshold change, and no test deleted. PostgreSQL budgets unchanged; Oracle budgets skipped by design (RFC §6). ✓
2 Rules CHANGELOG [Unreleased] ✓. G7 row ✓. T68 is new: it is not on main, and the only other open PR adding a thread (#575) uses T69. XML docs on the new classes and constructors are placed before existing /// blocks. merge-tiers.yml puts both scripts in Tier C. No PublicAPI impact (tests only). ✓
3 Confidentiality grep -inE over the diff, every commit message and the PR body, for the brief's list plus \bBTK\b, credential patterns and internal hostnames 0 hits ✓
4 Reproduce dotnet build tests/Goldpath.IntegrationTests -c Release: 0 warnings, 0 errors. docs-freshness, ledger-check, schema-honesty, skills-parity, template-pins: rc 0 ×5. dotnet format whitespace --verify-no-changes on the 4 .cs files: clean. Lane filters: each of the three, narrowed with &FullyQualifiedName~BenchOracleTests, gives "No test matches". GOLDPATH_BENCH_PROVIDER=oracle scripts/bench-jobs.sh: 2 lines tagged provider=oracle, exit 0 (also the positive control that the twins are selectable). RED: the recipe's plant (Oracle twin wired to PostgreSQL) printed untagged lines, then the refusal, exit 1. Restored: bench-bulk.sh card on Oracle fails both facts with ORA-01460 at GoldpathBulkFileStore.SaveAsync, as #559 describes. scripts/bench-bulk.sh card on PostgreSQL: green, lines untagged and in the old format. ✓
5 CI on ed7fcb2 Required (merge-policy, gates, integration, kit, golden manifests (PR subset)): all success. gm/*, CodeQL and review: success. oracle (not required): failure. Details below. ✓ (see below)
6 Charter n.a. —

The red oracle check comes from main and the environment, not this PR:

  • Failing test: NotificationOracleTests.The_renewal_mail_actually_lands_and_the_evidence_stamps failed with ORA-12537: TNS:connection closed. This PR does not touch Notification.
  • Main is red too: the oracle lane failed on main itself at c9d23ae, on a different test (CampaignOracleTests.Thousands_release…, TaskCanceledException). It was also red at 440f4a0 and 9311ccd.
  • Same failure on main locally: in this sandbox the same Notification fact failed identically (TaskCanceledException in WaitForStateAsync) at the PR head and at main c9d23ae.
  • The twins are not in that lane: the PR's Oracle twins carry Category=Bench, and the lane's filter Category=Oracle&Category!=Bench matches none of them, as shown in step 4.

Deep review

Area Result
a Security ✓ The workflow adds one choice input with no permission, secret or action change. NIT 2 covers one interpolation into an echo. It is not exploitable: a choice input, and dispatch needs write access. No new dependencies.
b Correctness & edge cases ✓ The scripts capture the dotnet test exit code under set +e and fail on it before the tag check. They refuse an Oracle run with zero lines or with any untagged line. An unknown provider exits 2, and so does the case variant Oracle. The filters' Integration.<Class>. form cannot match the derived twin. The user_segments size query is per schema. Under tr-TR the tag still holds.
c Concurrency ✓ The twins run in OracleCollection, which runs alone. SuiteStore schemas take an Interlocked counter, and the pool is cleared on stop.
d Data & migrations n.a. Tests only, with a fresh schema per suite.
e Performance ✓ Bench-only code with no hot path. The PostgreSQL wiring is equivalent to the old one: SuiteStore.PostgreSql() starts its own postgres:17-alpine container and runs EnsureCreated.
f Compatibility ✓ No package or API change. PostgreSQL defaults and line format are unchanged; I checked this from my own PostgreSQL card run. bench-jobs.sh on PostgreSQL = Category=Bench&Category!=Oracle, which is the old set.
g Test quality ✓ The refusal goes red under two plants: the recipe's, and my mixed-tag mutation. On Oracle the Jobs fact still asserts Completed and Processed == items before skipping the budgets.
h Code quality ✓ BenchProvider is the one place the tag is spelled, and the twins follow the BulkTests/BulkOracleTests pattern. NIT 1 is about a message's wording.
i Observability & operability ✓ The job summary names the provider and now also catches BENCH items=.
j Honesty ✓ Bulk publishes no number and says why, which I reproduced. The Jobs page's profile, image tag, build configuration and date are stated. My Oracle Jobs runs (this sandbox, co-located with another session's containers) gave 12.7–17.7 ms a chunk and 2.5–3.5 s: the same order as the page. NIT 2 is the one PR-text inaccuracy.

Edge-case probes (throwaway, not pushed)

# Probe Expected Result
P1 GOLDPATH_BENCH_PROVIDER=mysql scripts/bench-jobs.sh / =Oracle scripts/bench-bulk.sh exit 2 with a message exit 2 ×2 ✓
P2 Jobs BENCH-INTERACTIVE line with its tag removed (one tagged line and one untagged) refused exit 1 ✓ (see NIT 1 for the message)
P3 Oracle Jobs bench under DOTNET_CULTURE=tr-TR tag present, exit 0 ✓ (numbers print as 1,8ms and %435; see NIT 3)
P4 Docker unavailable (it was down at first) the script fails, not passes exit 1 ✓
P5 scripts/bench-bulk.sh card with the default provider PostgreSQL only, untagged, green ✓

NITs (optional, none blocks)

  1. Refusal message: scripts/bench-bulk.sh L122 / scripts/bench-jobs.sh L166. On a mixed run (P2) the message says "printed no Oracle bench line" when one was printed. Suggested wording: "printed a bench line without provider=oracle".
  2. PR body claim: the self-review says the input is "interpolated only into if: expressions, never into a shell line". In fact bench.yml L65 puts ${{ inputs.provider || 'postgres' }} into an echo. Not exploitable (a choice input). Correct the claim, or pass the value through env:.
  3. Culture-dependent bench lines (pre-existing, not this PR): the bench lines format numbers in the current culture. Pinning InvariantCulture, as BulkIntakeBenchTests already does, would keep pasted numbers comparable.

Could not run: a bench.yml dispatch run; Bulk on Oracle beyond its first save (#559); the full oracle lane (I ran only the failing class, on both PR and main).

Labels: agent-verified added. Tier C: the owner merges.

Owner: safe to merge — summary: scripts/bench-jobs.sh/bench-bulk.sh and bench.yml (by dispatch) can run the Jobs and Bulk benches on Oracle, and refuse a run that did not actually measure Oracle; the Jobs page gains its Oracle column.
Bulk's Oracle numbers stay open under T68/#559 (#556 is not closed by this PR); nothing changes for PostgreSQL or for any package.


Generated by Claude Code

@omercelikdev omercelikdev added the agent-verified Independently verified by the verifier agent: ready for owner merge label Nov 12, 2026
@omercelikdev
omercelikdev merged commit 90f78d3 into main Nov 12, 2026
26 of 27 checks passed
@omercelikdev
omercelikdev deleted the agent/556-oracle-bulk-jobs-benches branch November 12, 2026 09:03
omercelikdev added a commit that referenced this pull request Nov 12, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

agent-pr PR opened by the overnight agent agent-verified Independently verified by the verifier agent: ready for owner merge owner-approved ADR-0014: the owner approves merging this Tier C agent PR tier:C ADR-0014 Tier C: decisions, releases, gates, policy, agent layer; the owner merges

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants