fix(destregistry): handle publisher creation errors as delivery failures

[alexluong]

Summary

• When CreatePublisher fails (e.g. invalid GCP service account credentials missing the type field), the error was wrapped as a PreDeliveryError which caused a nack → retry loop until messages exhausted Pub/Sub max delivery attempts and landed in the DLQ — completely invisible to the customer
• Root cause: a customer configured a GCP Pub/Sub destination with invalid service_account_json (valid JSON but missing required GCP fields). The delivery worker couldn't create a Pub/Sub client, nacked every attempt, and the events silently went to DLQ after 6 retries
• Events were successfully delivered to the customer's other 3 destinations — only the misconfigured GCP Pub/Sub destination was affected

Changes

destgcppubsub.go — Destination-level fixes:

• CreatePublisher now returns ErrDestinationPublishAttempt (instead of plain errors) for both resolveMetadata failures (validation_failed) and pubsub.NewClient failures (client_creation_failed), signaling to the registry that these are delivery errors
• resolveMetadata now validates that service_account_json contains the required type field, catching invalid credentials at destination creation time

registry.go — Registry-level safety net:

• PublishEvent now checks for ErrDestinationPublishAttempt and ErrDestinationValidation from ResolvePublisher and creates a failed attempt record instead of returning nil — this ensures the error flows through the AttemptError path (logged to ClickHouse, visible in dashboard, retry-scheduled then acked) instead of the PreDeliveryError path (nack → DLQ)

Error flow before

CreatePublisher fails → plain error
  → ResolvePublisher returns (nil, err)
  → PublishEvent returns (nil, err)
  → doHandle sees nil attempt → PreDeliveryError
  → shouldNackError → nack → Pub/Sub retries 6x → DLQ

Error flow after

CreatePublisher fails → ErrDestinationPublishAttempt
  → ResolvePublisher returns (nil, err)
  → PublishEvent detects ErrDestinationPublishAttempt → creates failed Attempt
  → doHandle sees non-nil attempt → AttemptError
  → logged to ClickHouse → retry scheduling → ack
  → customer sees failed attempt in dashboard

Test plan

• New validation test: missing type field in service_account_json — rejects at Validate() time
• New validation test: missing type field - but passes with emulator endpoint — emulator skips credential validation
• New registry test: CreatePublisher returning ErrDestinationPublishAttempt → failed attempt returned
• New registry test: CreatePublisher returning ErrDestinationValidation → failed attempt returned
• New registry test: CreatePublisher returning unknown error → nil attempt (existing behavior preserved)
• All existing destregistry and deliverymq tests pass

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
outpost-docs	Ready	Preview, Comment	Apr 13, 2026 10:38am
outpost-website	Ready	Preview, Comment	Apr 13, 2026 10:38am