Discord Feedback Notification Bug Fix + Resolution Notification

Date2026-04-09

ImpactMedium

Decisions

Context: In Sprint 63, the identity-discord-secret K8s Secret and source repo manifest were written, but the env block was not reflected in the GitOps repo (aether-gitops), causing Discord notifications to be silently skipped.
Choice: Add DISCORD_FEEDBACK_WEBHOOK_URL secretKeyRef env block to identity-service.yaml in aether-gitops
Alternatives: None (obvious omission fix)
Code Paths: aether-gitops/algosu/base/identity-service.yaml

Context: Need to notify feedback authors when their feedback is resolved. Existing NotificationService + SSE infrastructure is complete, so no new infrastructure needed.
Choice: Add FEEDBACK_RESOLVED NotificationType + call NotificationService.create() on RESOLVED transition in FeedbackService
Alternatives: Discord DM (not possible via webhook), email (email service not built)
Code Paths: services/identity/src/notification/notification.entity.ts, services/identity/src/feedback/feedback.service.ts

Where: services/identity/src/feedback/feedback.service.ts (updateStatus method)
When to Reuse: When attaching supplemental notifications to core business logic (status change). Absorb failures with Promise.catch() to avoid affecting the main flow.

Where: services/identity/src/database/migrations/1709000019000-AddFeedbackResolvedNotificationType.ts
When to Reuse: When adding a new value to a PostgreSQL ENUM. Order is mandatory: COMMIT → ALTER TYPE ADD VALUE IF NOT EXISTS → BEGIN

Lessons Learned

Symptom: Secret creation complete, source repo manifest written, but env variable not injected into actual Pod
Root Cause: Env block not reflected in aether-gitops repo, so ArgoCD-synchronized actual Deployment is missing it
Fix: Always update GitOps repo and source repo manifests together. After deploy, verify actual injection with kubectl exec -- printenv

Symptom: GitOps tag updated + ArgoCD Synced status, but Pod running old image
Root Cause: imagePullPolicy: IfNotPresent set, so cached image on same node is reused
Fix: Force new image pull with kubectl rollout restart to recreate Pod

Symptom: In-app notification creation silently fails because FEEDBACK_RESOLVED ENUM doesn't exist in DB
Root Cause: Attempting feature test before image containing migration file is actually deployed to Pod
Fix: Must confirm that image containing migration has been deployed to actual Pod and init container has run