update engineering/0017/SESSION-0017.0-TRANSCRIPT-2026-06-10T21-33--INPROGRESS.md

This commit is contained in:
2026-06-10 21:47:58 -07:00
parent 92cf912d5d
commit 2b353258eb
@@ -35,8 +35,43 @@ spec interval amendment) → PR → merge.
Session runs isolated in worktree `gitea-backups-slice-2` (live concurrent
SESSION-0009 in flight).
## Execution record (checkpoint, ~21:50 PST)
- **Repo artifacts** (PR #42, merged): `infra/gitea-backup/gitea-backup.sh` +
`gitea-backup.{service,timer}`; spec v0.2.1 (Q2 amended 8h→4h, §6.4/§6.5
pure-create wording); runbook Backups build-state + operating verbs; plan
archived in-repo at `plans/2026-06-10-gitea-backups-slice-2.md`.
- **Scratch gate (T-1, spec §6.8)**: scratch bucket
`gs://wiggleverse-gitea-backups-scratch` (objectCreator for the VM SA; kept
for SLICE-3's T-4) + throwaway `gitea-scratch` container on the VM (torn
down after). Two real bugs caught and fixed before live:
1. staging dir must be created in-container and chowned to `git` (root
host-mkdir → dump can't write; then /data is root-owned so git can't
mkdir either);
2. fresh scratch gitea lacks `/data/git/repositories` (env setup, not a
script bug).
T-1 verification independent from the laptop: download, `unzip -t` intact,
sha256 == marker's (30f85305…), marker JSON complete.
- **Live enable**: units installed, timer enabled. First live run caught bug
3: `curl --data-binary` buffers the archive in RAM → OOM on the e2-small
with the real 1.02 GiB dump (scratch was 66 KB — scale fidelity gap in the
scratch gate, noted). Fixed with streaming `-T`; re-run green:
`gitea-dump-20260611T044540Z.zip` (1,097,287,374 B) + fresh marker in
`gs://wiggleverse-gitea-backups`, `Result=success`, next timer firing
2026-06-11 08:15 UTC on the 4h grid (00/4:15, Persistent).
- SLICE-2 DoD: snapshot landed + verified, timer enabled + idempotent +
read-only (INV-7); "on schedule" firing observable at 08:15 UTC
(timer verified armed; first run was the manual trigger per plan).
## Deferred decisions
- Tightened the spec's resolved Q2 (8h) to 4h on the operator's one-line
instruction ("SLICE-2 but make it every 4 hours") and amended the spec
(v0.2.1) rather than re-opening Q2 — confident this is what was meant;
flagging only because it edits a resolved decision row.
- Kept the scratch bucket (`…-backups-scratch`, empty but for one 66 KB test
object) instead of deleting it — SLICE-3's T-4 retention test reuses it.
_Autonomous-mode low-confidence calls the driver made and would have
liked operator input on. Appended as the session runs; surfaced at
finalize. Empty if none._