RunOnFlux/flux v8.12.0

on GitHub

Summary

Release v8.12.0. Bundles four changes:

• Mixed-compose g: handling — Compose apps with a mix of g: (master/slave) and non-g: components no longer leave their non-g: siblings stuck stopped on boot or blocked from auto-restart at runtime.
  Decisions in both stoppedAppsRecovery and peerNotification are now made per-component on the actual containerData, instead of treating the whole app as g: if any sibling is. The 30-minute install grace also
  stops bleeding onto non-syncthing components.
• Larger / longer uploads — formidable body limit raised from 5 GB → 10 GB, and the Node HTTP server requestTimeout is set to 2 hours so app backups and FluxShare uploads survive slow connections.
  headersTimeout is left at the default to keep slowloris protection on the request line.
• Docker image size limit raised to 5 GB — config.fluxapps.maxImageSize bumped from 2 GB to 5 GB for all apps. Allows larger images network-wide without per-owner gating.
• appSpawner cadence and deferral bug fixes — three related bugs that explain why apps could take up to 30 min to install on allowlisted enterprise nodes even though enterprise nodes are supposed to spawn-tick
  every 60s.
• Version bump to 8.12.0.

Changes

g: syncthing — per-component, not per-app

• stoppedAppsRecovery: new getNonGComponentIdentifiers() partitions an app's components and starts only the non-g: ones via appDockerStart(componentName_appName). Apps where every component is g: are still
  skipped entirely (left to masterSlaveApps). Adds an appsPartiallyStarted bucket to results for mixed apps.
• peerNotification: classifies the specific stopped component (not the app) for two decisions — (a) whether to call handleMissingMasterSlaveContainer (only when the stopped component itself is g:), and (b)
  whether the 30-minute install grace applies (only r:/g: components). Gracefully handles legacy v≤3 single-component specs.

Uploads

• ZelBack/src/lib/fluxServer.js: server.requestTimeout = 2h.
• ZelBack/src/services/IOUtils.js: maxFileSize 5 GB → 10 GB.

Image size limit

• ZelBack/config/default.js and tests/unit/globalconfig/default.js: fluxapps.maxImageSize 2 GB → 5 GB. Single uniform limit, no plumbing changes — the existing limit is already enforced through
  verifyRepository and verifyAndPullImage.

appSpawner fixes (ZelBack/src/services/appLifecycle/appSpawner.js)

• Spawn cadence honored on empty-candidate paths. getSpawnDelays(isEnterprise=true, …) returns delayTime: 60_000, but the two early-exit branches in trySpawningGlobalApplication were sleeping on a hardcoded
  30 * 60 * 1000 instead of delayTime. On a quiet enterprise node every spawn cycle was 30 min apart in production logs (00:27:25 → 00:57:25 → 01:27:25 → …), bounding registration→install latency at ~30 min
  instead of the intended ~60s. Both hardcodes replaced with delayTime. Non-enterprise nodes unaffected — getSpawnDelays(false, 0) already returns 30 * 60 * 1000.
• Inverted timeToCheck comparison. appsToBeCheckedLater / appsSyncthingToBeCheckedLater entries are pushed with a future timeToCheck so the spawner can come back to them later. The lookup was
  findIndex((app) => app.timeToCheck >= Date.now()), which matches entries whose timer has not elapsed — opposite of intent. Effect: deferred entries were immediately re-pulled on the next iteration, bypassing every
  "will check in around 27m / 2m" deferral. Both comparisons flipped to <=.
• .includes(predicate) was a silent no-op. Array.prototype.includes does a SameValueZero equality check, not a predicate match, so appsToBeCheckedLater.includes((appAux) => appAux.appName === app.name) was
  always false and the filter never excluded anything. Replaced with .some(predicate).

Tests

• New per-component g: cases in tests/unit/peerNotification.test.js (mixed compose: non-g: auto-starts, g: sibling left alone; r:-only path).
• New cases in tests/unit/stoppedAppsRecovery.test.js covering mixed-compose partial start.
• Small fix in tests/unit/syncthingHealthMonitor.test.js to expect sendMessage=true on the remove-threshold path.

Test plan

• Unit: npm run test:zelback:unit — new peerNotification, stoppedAppsRecovery, and syncthingHealthMonitor cases pass; appSpawner / enterpriseNetwork / globalState suites pass (45/45).
• Boot recovery: install a mixed compose app (one g: component, one plain). Stop all containers, restart FluxOS, confirm only the non-g: component auto-starts and the g: component remains for
  masterSlaveApps to elect a primary.
• Runtime restart: with the same app running, manually stop the non-g: component; confirm peerNotification auto-restarts it without the 30-minute grace.
• All-g: app: confirm it's still skipped end-to-end (no regression in master/slave behavior).
• Legacy v≤3 app with g: containerData: confirm it's still skipped.
• FluxShare upload of a >5 GB file (e.g. 7–8 GB) succeeds.
• Large upload over a slow link runs past the previous 5-minute requestTimeout without being killed.
• Validate a 3–5 GB Docker image deploy succeeds; a >5 GB image is rejected.
• Deploy to an enterprise node and confirm the spawn loop logs trySpawningGlobalApplication - Checking for apps... at ~60s cadence instead of every 30 min.
• Register a fresh enterprise app and confirm the install lands on an allowlisted node within ~2 min of broadcast propagation.