fix(tra-974): pin preview backend LOG_LEVEL=info to stop debug log firehose

[mikestankavich]

Problem

Preview detects as EnvDev, so the Go backend logger defaults to debug (logger/config.go) and emits a per-message ingest message processed line ~3/sec, 24/7. That's a GCP Cloud Logging ingest cost with no steady-state value, and it drowns other sessions troubleshooting on preview. Decision: preview should be quiet by default, escalate to debug only during an active debugging window.

Root cause (found while implementing — would have been a no-op otherwise)

The chart's config.logLevel only ever emitted BACKEND_LOG_LEVEL, which the current backend does not read. Proven against the live cluster: preview's ConfigMap already had BACKEND_LOG_LEVEL=info, yet the pod logged at debug — and an explicit LOG_LEVEL=info env (the stopgap kubectl set env) is what actually silenced it. So pinning config.logLevel would have shipped a silent no-op.

Fix — preview-only, prod untouched

• chart (helm/trakrf-backend): new optional config.runtimeLogLevel that emits a real LOG_LEVEL ConfigMap key only when set. Empty default → envs that omit it (prod) render no LOG_LEVEL key and keep their APP_ENV→warn default — ConfigMap byte-identical to today, no prod pod bounce.
• root app (argocd/root): per-env logLevel: info for preview, appended to the config block via the template's existing no-env-conditional values idiom (same pattern as jwtExpirationSeconds).

The legacy/dead BACKEND_LOG_LEVEL key is left in place and flagged for a separate cleanup. The backend-side detection fix (preview gets its own env defaulting to info) is the platform-repo option (b), out of scope here.

Verification (helm template)

# chart default (runtimeLogLevel empty)        → BACKEND_LOG_LEVEL only, no LOG_LEVEL
# chart --set config.runtimeLogLevel=info       → LOG_LEVEL: "info"
# root render (gke): trakrf-backend-preview      → inlineValues config.runtimeLogLevel: "info"
#                    trakrf-backend-prod          → (absent)

Deploy note

Root-template change → does not auto-sync. Requires scripts/apply-root-app.sh gke post-merge to materialize (re-renders all env child apps; only preview's ConfigMap actually changes, so only the preview backend pod bounces). The interim kubectl set env LOG_LEVEL=info override on preview keeps it quiet until then.

Tracking: TRA-974.

🤖 Generated with Claude Code