purge-unhealthy ran for four hours on 2026-09-21 and then failed on "push rejected (fetch first)", throwing the whole run away. The data-write concurrency group serialises the WORKFLOWS, but main still moves under a long job: a merged pull request, an /admin approval or override the Worker commits, the extension's push to scripts/temp_info.jsonl. bash/commit_push.sh does what every wrapper did - add, commit only if something changed - and then pushes with up to five attempts, rebasing onto origin/main between them. None of the other writers touch the shards, so the rebase is clean; a genuine conflict still fails the job. All nine data wrappers now call it, with their commit messages unchanged (snapshot.sh keeps its data/snapshots/ pathspec). Tested against a scratch remote: with main moved by another commit, the first push is rejected, the rebase lands the job's commit on top, and the retry succeeds; with nothing to commit it prints "No changes". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
6 lines
210 B
Bash
6 lines
210 B
Bash
#!/bin/bash
|
|
set -euo pipefail
|
|
# No pip install: snapshot.py and scripts/core/* are stdlib-only.
|
|
python scripts/snapshot.py
|
|
bash "$(dirname "$0")/commit_push.sh" "Snapshot [$(date +'%Y-%m-%d')]" data/snapshots/
|