zram/overlayfs: reliable swap init and image-owned init-script refresh

[pdscomp]

zram/overlayfs: reliable swap init and image-owned init-script refresh

Split from #252.

Why this PR

This revision greatly simplifies zram initialization compared with the original version of PR #252. It more closely follows the original script from @jamesturton while retaining added logging, comprehensive state and device checks, idempotent behavior, and bounded retry logic (as well as a test case to validate script execution on the host after any updates).

A CC1 boot that fails to bring zram online is the main OOM path during normal load or calibration. The old one-shot zram init could silently continue without swap. Separately, the persistent /etc overlay could keep stale package-owned init scripts after an image update, silently defeating runtime fixes.

Justification / measurements

Reliability change, not a performance change — no MiB claim.

• New zram init: idempotent, waits for device/sysfs, picks first available compressor, verifies /dev/zram0 in /proc/swaps, retries once via module reload, logs failures.
• Overlay refresh: on first boot after A/B update, removes only upper-layer copies/whiteouts shadowing scripts present in the new image. User-created scripts untouched.

Risk / rollback

Low. Reverts cleanly; worst case is reverting to previous one-shot zram init behavior.

[pdscomp]

Feedback from @suchmememanyskill

• overlayfs-etc-preinit.sh.in: doesn't know what it's supposed to do.
• init.d cleanup "should just be a single line rm -f /data/overlay-etc/upper/init.d/*" — user services should be re-symlinked manually per update, since the update could have broken their service anyway.

[pdscomp]

@suchmememanyskill

1. The overlayfs preinit block removes overlay copies of image-owned init scripts before /etc is mounted, so an OTA update's new init script isn't shadowed by a stale overlay copy. It only deletes files that also exist in the image's /etc/init.d.

2. A single rm -f /data/overlay-etc/upper/init.d/* would delete user-added init scripts too. The current loop only removes overlays of image-owned scripts, matching the existing behavior we already had for other services.

[github-actions]

✅ Build Artifacts

Branch: doom/zram-init
Build: 9b0eb8b (merge into main)

Artifact	Size
CC1 Firmware	91.31 MB

View workflow run

[pdscomp]

Validation

Ran meta-opencentauri/recipes-core/zram/tests/init-test.sh against the new zram init script.

All cases passed:

• Normal startup: PASS — zram swap activates successfully.
• Startup logging: PASS — successful activation is logged.
• Idempotent startup: PASS — a second start detects active swap and is a no-op.
• Transient module failure: PASS — recovers after one modprobe failure and succeeds on retry.
• Retry logging: PASS — retry attempt is recorded.
• Permanent module failure: PASS — setup fails cleanly after the bounded five-attempt limit.
• Failure cleanup: PASS — no swap remains active after permanent failure.

Result: PASS

On-machine testing will be next once a formal PR build is complete.

[pdscomp]

On-machine CC1 validation

Validated the formal PR-254 build on carbon2u after reboot.

zram/swap

• /proc/swaps: PASS — /dev/zram0 active, 229,308 KiB total, 9,216 KiB used, priority 100.
• swapon --show: PASS — approximately 223.9 MiB zram swap online.
• free -m: PASS — approximately 28.4 MiB RAM available with zram enabled.
• Boot log: PASS — kernel added zram0, configured capacity, and enabled swap successfully.
• Deployed init script: PASS — contains retry, verification, idempotence, and failure-bounding logic.
• Live idempotence check: PASS — running /etc/init.d/zram start while active returned zram: already active.
• VM tuning: PASS — swappiness 150, watermark boost 0, watermark scale 125, page-cluster 0, and vfs cache pressure 500.
• No zram or out-of-memory failures were found in the inspected logs.

Service health

• Klipper: PASS — running and API state ready.
• Moonraker: PASS — connected to Klipper, no failed components reported.
• uStreamer: PASS — running.
• Printer API: PASS — reports Release - PR-254-9b0, state ready.
• System uptime/load: PASS — approximately 7 minutes uptime, load average below 0.1.

The PR is now ready for review and possible merge.

[github-actions]

✅ Build Artifacts

Branch: doom/zram-init
Build: 3fceb5f (merge into main)

Artifact	Size
CC1 Firmware	91.31 MB

View workflow run