fix(ci): correct Forgejo exporter status semantics #14

Open

john wants to merge 1 commit from feat/forgejo-ci-exporter-status into main

john commented

2026-07-31 22:29:58 +00:00

Owner

Outcome

Establishes the previously host-only Forgejo CI exporter as versioned source and corrects status interpretation for Forgejo 15.0.3.

Changes

defines and tests the verified enum 0..7 (unknown, success, failure, cancelled, skipped, waiting, running, blocked)
selects the current task via job.task_id, falling back to latest attempt, and treats failed step 2 as strongest retained terminal evidence
removes timestamp-derived success and all status-4-as-failure/status-3-as-success logic
mirrors Forgejo's exact aggregate ordering for parent runs
excludes cancelled/skipped from success/failure-rate denominators
adds explicit historical trust-boundary and parent-disagreement metrics
preserves existing metric names/labels consumed by Grafana and Operator
requires the database DSN at startup, uses a PostgreSQL read-only session, and moves secrets to an environment file
adds a locked uv package, systemd unit, tests, and path-scoped Forgejo CI workflow

Validation

uv run ruff check .
uv run ruff format --check .
uv run pytest — 22 passed, including every status 0–7 and representative Prometheus labels
generated CTE smoke test completed in a production READ ONLY transaction
full collector completed through an SSH tunnel with PostgreSQL read-only mode: last_collection_success = 1
current Prometheus target forgejo-ci verified up, one-minute scrape, no error
public CI Run Analytics dashboard verified with 11 panels and existing label contracts; Operator only embeds its public token

Representative corrected values were success_rate_7d=0.8602356, runs_today=670, and runs_today_failed=92. The deployed incorrect exporter had shortly before reported approximately 0.453844, 648, and 359.

Production boundary

This PR does not install or restart the exporter and does not edit Grafana. Deployment/restart and the dashboard caveat/threshold update require an explicit production checkpoint. The large metric shift is expected semantic blast radius, not proof of fully reconstructed historical truth.

Spec Drift Callouts

The live exporter had no authoritative repository copy; it was created directly on the production host. This PR establishes the source under haskytech-ops.
The July 30 claim that step-derived history represented true success was too strong. Pre-containment data is explicitly labeled historically corrupted and unrepaired.

## Outcome Establishes the previously host-only Forgejo CI exporter as versioned source and corrects status interpretation for Forgejo 15.0.3. ## Changes - defines and tests the verified enum `0..7` (unknown, success, failure, cancelled, skipped, waiting, running, blocked) - selects the current task via `job.task_id`, falling back to latest attempt, and treats failed step `2` as strongest retained terminal evidence - removes timestamp-derived success and all status-4-as-failure/status-3-as-success logic - mirrors Forgejo's exact aggregate ordering for parent runs - excludes cancelled/skipped from success/failure-rate denominators - adds explicit historical trust-boundary and parent-disagreement metrics - preserves existing metric names/labels consumed by Grafana and Operator - requires the database DSN at startup, uses a PostgreSQL read-only session, and moves secrets to an environment file - adds a locked uv package, systemd unit, tests, and path-scoped Forgejo CI workflow ## Validation - `uv run ruff check .` - `uv run ruff format --check .` - `uv run pytest` — 22 passed, including every status 0–7 and representative Prometheus labels - generated CTE smoke test completed in a production `READ ONLY` transaction - full collector completed through an SSH tunnel with PostgreSQL read-only mode: `last_collection_success = 1` - current Prometheus target `forgejo-ci` verified `up`, one-minute scrape, no error - public `CI Run Analytics` dashboard verified with 11 panels and existing label contracts; Operator only embeds its public token Representative corrected values were `success_rate_7d=0.8602356`, `runs_today=670`, and `runs_today_failed=92`. The deployed incorrect exporter had shortly before reported approximately `0.453844`, `648`, and `359`. ## Production boundary This PR does not install or restart the exporter and does not edit Grafana. Deployment/restart and the dashboard caveat/threshold update require an explicit production checkpoint. The large metric shift is expected semantic blast radius, not proof of fully reconstructed historical truth. ## Spec Drift Callouts - The live exporter had no authoritative repository copy; it was created directly on the production host. This PR establishes the source under `haskytech-ops`. - The July 30 claim that step-derived history represented true success was too strong. Pre-containment data is explicitly labeled historically corrupted and unrepaired.

john added 1 commit

2026-07-31 22:29:58 +00:00

fix(ci): correct Forgejo exporter status semantics

Forgejo CI Exporter / test (pull_request) Successful in 9s

Details

3b6182d2a5

Forgejo CI Exporter / test (pull_request) Successful in 9s

Details

This pull request can be merged automatically.

This branch is out-of-date with the base branch

You are not authorized to merge this pull request.

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin feat/forgejo-ci-exporter-status:feat/forgejo-ci-exporter-status

git switch feat/forgejo-ci-exporter-status

No reviewers

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

haskytech/haskytech-ops!14

No description provided.

Rows
Columns