RunOnFlux/flux v8.6.2

on GitHub

PR #1673 revealed that enterprise apps have compose: [] in the local DB (encrypted in the enterprise field), which unearthed multiple critical bugs in the redeployment and recovery logic.

Issues Fixed

1. Repeated Redeployments Every 12 Minutes

• reinstallOldApplications was comparing appSpecifications.hash vs installedApp.hash
• decryptEnterpriseApps called specificationFormatter() which strips metadata (hash, height, instances, expire)
• Result: installedApp.hash was undefined, causing false positive "obsolete" detection
• Fix: Added formatSpecs option (default: true for backward compatibility), getInstalledAppsFromDb now passes formatSpecs: false to preserve metadata

2. Missing Container Recovery

• When containers were missing (not just stopped), no recovery mechanism existed
• Fix: Added recreateMissingContainers() to peerNotification.js that reinstalls components when containers don't exist

3. Syncthing Delay Applied Incorrectly

• Apps with syncthing flags (r:, g:) would wait 30 minutes before recreation even when containers were completely missing
• The delay is meant for stopped containers to allow data syncing, not missing containers
• Fix: Check container existence first; only apply 30-minute delay if container exists but is stopped

4. Component Redeployment Crashes

• softUninstallComponent and hardUninstallComponent called with null appId causing crashes
• Fix: Calculate appId using dockerService.getAppIdentifier() before calling uninstall functions
• Fix: Pass appComponent instead of appSpecifications as third parameter

5. Hard Redeployment Broken for Enterprise Apps

• hardRedeployComponent loaded specs from getStrictApplicationSpecifications (doesn't decrypt)
• Then tried to access empty compose array
• Fix: Added decryption check before accessing compose array

6. Soft Redeployment Component Broken for Enterprise Apps

• softRedeployComponent had the same issue as hardRedeployComponent
• Loaded specs without decryption, then tried to access empty compose array
• Fix: Added decryption check before accessing compose array

7. stopAppMonitoring Parameter Bugs

• softUninstallComponent called with boolean true instead of the actual stopAppMonitoring function (fixed in previous commit)
• hardUninstallComponent also called with boolean true instead of function
• Both functions invoke stopAppMonitoring() as a function, so passing true would crash with "true is not a function"
• Fix: Pass the function reference, not a boolean

Testing

• Verified on nodes: 38.240.227.140, 162.55.241.211, 195.3.222.71, 148.251.229.179, 65.108.231.135, 176.9.31.60, 23.88.73.206
• Confirmed streamrnode apps no longer repeatedly redeploy
• Confirmed wordpress apps with syncthing now recreate immediately when containers are missing
• Confirmed syncthing apps are created but not started, then handleReceiveOnlyTransition starts them after sync
• Confirmed container recreation works for missing enterprise app containers

Related

• Builds on #1673 which identified the root cause (enterprise compose redaction)