fix(data-warehouse): stop a repartition retry abandoning the table

[Gilbert09]

Problem

• A large warehouse table stops repartitioning for good the first time its final sync run is retried, so it keeps syncing on the layout that made it fail.
• The retry cap counts sync runs, not the Temporal retries inside one run, and an attempt is charged before the rewrite starts.
• On the run that charges the third attempt, the stored count already reads the cap, so that run's own retry reads the cap as spent.
• The retry then abandons the rewrite instead of running it. It clears the pending target, engages the daily cooldown, and reports RepartitionAttemptsExhausted with final=true.
• A heartbeat timeout is enough to trigger this, and that is the ordinary reason Temporal retries this activity.

Changes

• A table now keeps its rewrite when the run holding its last attempt is retried, so a heartbeat blip on the final chance no longer abandons it.
• The cap check discounts a charge the current run made itself, by matching the stored charged_job_id against this run's job id.
• The give-up at the cap is unchanged for a retry that fails cleanly, because the failure path counts the charge itself, and for a later sync run, which sees a charge from another job.
• Nothing in the UI changes.

How did you test this code?

• New test test_a_retry_inside_the_run_that_spent_the_cap_still_rewrites: it fails when the cap check ignores the current run's charge, which is the defect fixed here. Its sibling test_retries_inside_one_sync_run_burn_a_single_attempt starts from zero attempts, so no existing case reaches the cap inside one run.
• Ran the cap check over its inputs directly in this sandbox: no marker, a count under the cap, a count at the cap, a charge from another job, and a charge from this job. Only the last one is newly exempt.
• Not run: test_repartition_controller.py itself. Its session setup needs ClickHouse, which this sandbox has no server for. The file collects, and CI runs it.
• No manual or production verification was performed.

Automatic notifications

• Publish to changelog?

Docs update

None. The README already describes the cap as consecutive failed sync runs, which is what this restores.

🤖 Agent context

Autonomy: Fully autonomous

• Written by Claude Opus 5 in a PostHog cloud task, triaging a day of warehouse_repartition_failed events. The terminal one was an RepartitionAttemptsExhausted give-up, and reading the charge and give-up bookkeeping found the retry that never rewrites.
• Skills invoked: /writing-tests, /writing-code-comments, /writing-pr-descriptions.
• No duplicate: no open PR touches the attempt cap. #95861 fixes the other failure mode from the same day, budget-exceeded rewrites that cannot converge, by holding imports while a checkpoint is live. It leaves the give-up path alone, and the two diffs do not overlap.
• Public artifact: the diff carries no material from the agent session. The test uses the fixture helpers already in the file and invented ids.

[trunk-io]

😎 Merged successfully - details.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[github-actions]

🤖 CI report

⚠️ Trunk lane — backend Python lane (py:product:warehouse_sources)

This PR is assigned to the backend Python lane (py:product:warehouse_sources). It runs backend Python tests and may merge in parallel with PRs in other lanes.

⚠️ Backend coverage — 95.0% of changed backend lines covered — 1 uncovered

🧪 Backend test coverage

Patch coverage — changed backend lines (products + core): ███████████████████░ 95.0% (19 / 20)

File	Patch	Uncovered changed lines
products/warehouse_sources/backend/temporal/data_imports/workflow_activities/repartition_table.py	87.5%	713

🤖 Agents: add a test covering the lines above, or note why under "How did you test this code?". Machine-readable gap list: the patch-coverage artifact on this run (gh run download 34109754203 -n patch-coverage), or the coverage-data block at the end of this comment.

Per-product line coverage (touched products)

Product	Coverage	Lines
warehouse_sources	███████████████████░ 97.2%	409,547 / 421,449

Report-only. Patch coverage = changed backend lines covered vs origin/master. Sorted lowest first.
Known gaps: lines covered only by Temporal tests show as uncovered; core line numbers may drift if master changed the same file.

[stamphog]

Approved.

Small, focused bugfix by an author on the owning team, with a new test covering the exact scenario described; no schema/API/security surface touched, and the only review comment is a bot that couldn't actually review (quota limit), not a substantive concern.

• Author wrote 0% of the modified lines and has 108 merged PRs in these paths (familiarity MODERATE).
• copilot-pull-request-reviewer[bot] reviewed the current head.

Gate mechanics and policy version

Gate		Result
prerequisites	✓	all clear
deny-list	✓	no deny categories matched
size	✓	19L, 1F substantive, 49L/2F incl. docs/generated/snapshots — within ceiling
tier	✓	T1-agent / T1b-small (49L, 2F, single-area, fix)
stamphog 2.0.0b4		.stamphog/policy.yml @ 1820958 · reviewed head 1820958

[trunk-io]

   

View Full Report ↗︎ ⋅ Docs

[talyn-app]

/trunk merge

[deployment-status-posthog]

Deploy status

Environment	Status	Deployed At	Workflow
dev	✅ Deployed	2026-09-07 12:32 UTC	Run
prod-us	✅ Deployed	2026-09-07 12:42 UTC	Run
prod-eu	✅ Deployed	2026-09-07 12:45 UTC	Run