When a webhook-triggered deploy starts for an app that already has a deploy
in progress, the new deploy now cancels the existing one via context
cancellation, waits for the lock to be released, and then starts the new
deploy.
Changes:
- Add per-app context cancellation (appCancels sync.Map) to deploy.Service
- Deploy() creates a cancellable context and registers it for the app
- Add CancelAppDeploy() method to cancel an in-progress deploy
- Add ErrDeployCancelled sentinel error for cancelled deploys
- Handle context cancellation in build and deploy phases, marking
deployments as failed with a clear cancellation message
- Webhook triggerDeployment() now cancels in-progress deploys and retries
until the lock is released (up to 30 attempts with 2s delay)
fixes#38
The gorilla/sessions MaxAge field expects seconds, not nanoseconds.
Previously MaxAge was set to -1000000000 (-1 * time.Second in nanoseconds),
which worked by accident since any negative value deletes the cookie.
Changed to the conventional value of -1.
Remove the sync.Mutex and CreateUserAtomic (INSERT ON CONFLICT) in favor
of a single DB transaction in CreateFirstUser that atomically checks for
existing users and inserts. SQLite serializes write transactions, so this
is sufficient to prevent the race condition without application-level locking.
Add mutex and INSERT ON CONFLICT to CreateUser to prevent TOCTOU race
where concurrent requests could create multiple admin users.
Changes:
- Add sync.Mutex to auth.Service to serialize CreateUser calls
- Add models.CreateUserAtomic using INSERT ... ON CONFLICT(username) DO NOTHING
- Check RowsAffected to detect conflicts at the DB level (defense-in-depth)
- Add concurrent race condition test (10 goroutines, only 1 succeeds)
The existing UNIQUE constraint on users.username was already in place.
This fix adds the application-level protection (items 1 & 2 from #26).
Store a SHA-256 hash of the webhook secret in a new webhook_secret_hash
column. FindAppByWebhookSecret now hashes the incoming secret and queries
by hash, eliminating the SQL string comparison timing side-channel.
- Add migration 005_add_webhook_secret_hash.sql
- Add database.HashWebhookSecret() helper
- Backfill existing secrets on startup
- Update App model to include WebhookSecretHash in all queries
- Update app creation to compute hash at insert time
- Add TestHashWebhookSecret unit test
- Update all test fixtures to set WebhookSecretHash
Closes#13
- Update app status to error when health check fails with an error
- Update app status to error in cleanupStuckDeployments for apps stuck in building state
- This fixes the inconsistency where app shows "building" but deployment is "failed"
- Write deployment logs to files when deployment finishes (success or failure)
- Log files stored in DataDir/logs/<hostname>/<appname>/<appname>_<sha>_<timestamp>.log.txt
- Capture commit SHA for manual deploys by parsing git rev-parse HEAD after clone
- Add download endpoint for log files at /apps/{id}/deployments/{deploymentID}/download
- Add download link in deployment history view for finished deployments
- Add commit_url column to webhook_events and deployments tables
- Extract commit URL from webhook payload (from commit object or repo URL)
- Format Slack messages with backticks for branch and commit SHA
- Link commit SHA to the actual commit URL on the git server
- Keep plain text format for ntfy notifications
Remove old container before creating new one instead of trying to keep
it for rollback. Rollback isn't safe anyway because database migrations
may have been applied by the new container.
The old stop-then-rollback approach failed because Docker doesn't allow
two containers with the same name, even if one is stopped.
- Clone specific commit SHA from webhook instead of just branch HEAD
- Log webhook payload in deployment logs
- Add build/deploy timing to ntfy and Slack notifications
- Implement container rollback on deploy failure
- Remove old container only after successful deployment
- Show relative times in deployment history (hover for full date)
- Update port mappings UI with labeled text inputs
- Add footer with version info, license, and repo link
- Format deploy key comment as upaas_DATE_appname
- Add app_ports table for storing port mappings per app
- Add Port model with CRUD operations
- Add handlers for adding/deleting port mappings
- Add ports section to app detail template
- Update Docker client to configure port bindings when creating containers
- Support both TCP and UDP protocols
Use DataDir/builds instead of /tmp for clone directories so that bind
mounts work correctly when upaas itself runs in a Docker container.
The /tmp directory inside the upaas container isn't accessible to the
Docker daemon on the host, causing bind mount failures.
Also fix test setups to pass Config to deploy service and add delay
to webhook test to avoid temp directory cleanup race with async
deployment goroutine.
- Replace UUID with ULID for app ID generation (lexicographically sortable)
- Remove container_id column from apps table (migration 002)
- Add upaas.id Docker label to identify containers by app ID
- Implement FindContainerByAppID in Docker client to query by label
- Update handlers and deploy service to use label-based container lookup
- Show system-managed upaas.id label in UI with editing disabled
Container association is now determined dynamically via Docker label
rather than stored in the database, making the system more resilient
to container recreation or external changes.
