Saltar al contenido principal

Secrets and Connection Pool Fix - What You Need to Know

⚠️ Critical: Secrets Cannot Fix This Issue​

The connection pool exhaustion is NOT caused by missing or incorrect secrets.

The root cause is that your Cloud SQL instance (db-f1-micro) is too small - it only supports ~25 connections, but you need ~64 connections.

Secrets in deploy-metabase.yml Workflow​

Secrets That Exist (But Don't Control Connection Pools)​

Looking at the Metabase deployment workflow, these secrets are used:

  1. METABASE_POOL_MAX (Optional, Line 469)

    • Status: Referenced but DOES NOT ACTUALLY CONTROL Metabase's connection pool
    • Purpose: Only used for logging/reference in the workflow
    • Reality: Metabase uses HikariCP defaults (~10-15 connections per instance) which cannot be easily configured via environment variables when using MB_DB_* connection variables
    • Action: You can ignore this secret - setting it won't change anything

  2. Required Metabase Secrets:

    • METABASE_DB_CONNECTION_URI - Database connection (required)
    • METABASE_SITE_URL - Public URL (optional but recommended)
    • METABASE_EMBED_SECRET_KEY - JWT secret (required)
    • MB_ANON_TRACKING_ENABLED - Optional (defaults to false)
    • MB_ENABLE_PUBLIC_SHARING - Optional (defaults to false)

What Actually Controls Metabase Connections​

  • Instance scaling: --max-instances=2 in the workflow (Line 512)
  • Already optimized: Metabase is already set to 2 max instances

The Real Fix: Upgrade Cloud SQL Instance​

The issue is your Cloud SQL instance tier, not secrets:

# Upgrade the instance (this is the real fix)
gcloud sql instances patch metabase-db-staging \
  --tier=db-g1-small \
  --project=barto-dev

Current: db-f1-micro → ~25 max connections
Needed: db-g1-small → ~100 max connections
Your usage: ~64 connections

Temporary Mitigation (Backend Only)​

If you cannot upgrade immediately, you can reduce backend connection usage (NOT in the Metabase workflow):

In GitHub Secrets → staging environment:

DB_POOL_MAX=2

This reduces backend connections from 30 (10 instances × 3) to 20 (10 instances × 2), saving 10 connections.

Note: This secret is used in the backend deployment workflows (deploy-staging-from-main.yml and deploy-production.yml), NOT in the Metabase workflow.

Summary​

SecretWorkflowDoes It Help?Action
METABASE_POOL_MAXdeploy-metabase.yml❌ NoIgnore - doesn't actually control pool
DB_POOL_MAXBackend workflows⚠️ Temporary onlyCan reduce backend usage (saves 10 connections)
Cloud SQL Instance TierN/A✅ YES - This is the fixUpgrade from db-f1-micro to db-g1-small

  1. Immediate (Required): Upgrade Cloud SQL instance

    ./scripts/upgrade-cloudsql-instance.sh

  2. Optional (Temporary): Reduce backend pool if upgrade is delayed

    • Set DB_POOL_MAX=2 in GitHub Secrets → staging environment
    • This is a temporary workaround, not a permanent solution

  3. Don't bother with: METABASE_POOL_MAX secret - it doesn't do anything