BLE commissioning, 24 October 2026¶
I used the focused BLE playbook to add the missing production source and
appender on JASMIN. The service runs dashboard commit
ac672ae6d835d05461b2fcb8c209151aa8ace2de from its separate release checkout,
with /opt/aurora-presentations/runtime-20261024. Other scientific writers
keep their existing checkout and runtime.
Infrastructure preparation is in PR 63.
The first dry run found that the source role would change the shared state
directory's mode. PR 66
fixed that before activation; its 28 BLE tests and all four CI checks passed.
The applied infrastructure revision was 500036f7da505a387f2baaae1615391905e76c3a.
Archive boundary and preserved settings¶
Both product archive copies were already running. I paused only their timers and stopped their copies through the existing systemd stop behaviour. No raw archive jobs, source files, partial transfers or checkpoints were removed.
| UTC window | Result |
|---|---|
| 07:00:37–07:01:49 | Initial idle preflight; both product timers and unfinished copies restored while the shared-directory fix passed CI. |
| 07:03:14–07:04:44 | Final dry run and focused installation; both product timers restored active/enabled and both unfinished copies restarted. |
The saved pre-change state is under
/var/backups/aurora-blesensors/commission-20261024T070033Z on JASMIN. Archive
registration kept its own backup under
/var/backups/aurora-blesensors/20261024T070402Z-9d2b400e. These protected
backups contain configuration and stay off Git.
The non-BLE environment entries were byte-identical before and after the
release. /var/lib/aurora-cloud remained owned by aurora:aurora, mode 0755.
The focused registration added only BLE stream bindings and product exclusions;
it retained the installed archive code and other settings. Both preflight and
registration enforced the product-worker idle gate.
Production evidence¶
The first source cycle ran from 07:04:45 to 07:05:43 UTC, received 25 files and enqueued them through the existing raw archive dispatcher. The initial history append finished at 07:06:50: 132,096 input records, 66,545 accepted rows and no rejected records. The resulting store had no duplicate times and observations through 07:04:54 UTC. Raw files occupied about 8.8 MiB and the product 1.4 MiB.
The next scheduled source cycle finished at 07:06:52, with two changed files and 23 unchanged files. The scheduled append ran at 07:08:53–07:08:55 and added 114 rows with no rejection. At 07:09:13 the store contained 66,659 unique times; both APS and ASS temperature observations had advanced to 07:06:50 UTC. A third source cycle completed successfully at 07:08:57. Both source and append timers are enabled. The primary inventory now retains the dedicated BLE paths so a later full deployment uses the same reviewed code and runtime.
Development mirror evidence¶
The mirror-only dry run passed, then the focused standby tag installed just the raw and product BLE stages. Both first transfers succeeded at 07:09:50 UTC; scheduled checks succeeded again at 07:10:09 and 07:10:24. Both timers are active/enabled. The existing direct-field continuity timer remains active and enabled, with its separate data and environment unchanged.
At 07:10:07 on JASMIN and 07:10:17 on data-ocean, the copies agreed:
| Data | Files | SHA-256 over sorted relative paths and file bytes |
|---|---|---|
| BLE raw | 25 | 42dea448785a71d8429c82c61b0955a6588a3fb8b0447ea7a28bfba2c099fcc7 |
| BLE Zarr | 212 | 8686cb6b7eec158a7adb1aa1710b956578b03fef23caa24eb21e46e731547507 |
Both products had 66,659 rows and the same 07:06:50 sensor observations. Ingestion cursors, writer locks and staging stores were excluded from the product mirror. These hashes describe this check; later successful cycles will change them as observations arrive.
Scope and next checks¶
This is BLE acquisition/product commissioning, not evidence that all archive queues are drained or that the operational mobile mirror has passed its full acceptance gate. The development direct-field continuity service remains a separate path. Current procedures are in BLE enclosure data.
Archive verifier refresh¶
The normal verifier completed a new full pass at 07:20:36 (its summary records 07:10:56 as the run start). The BLE section now contains 25 source files, 25 local files and 23 GWS files, with zero missing or mismatched files in the settled local and GWS windows. The product gate passed and the appender had a successful exit at 07:19:19. The two newest files were outside the settled GWS window; the total GWS coverage was 92%, not complete all-file coverage.
The archive health snapshot at 07:21:28 no longer reported invalid BLE source
evidence. All four BLE settled-source missing/mismatch counts were zero, the
source timer was active and enabled, and the source service was healthy. The
BLE prune-ready state was still zero. Verification used size and modification
time (checksum_mode=none); this does not establish independent object-store
parity. Other archive failures remained, and retention stayed disabled. No
archive policy change or separate verifier run was needed.