litellm-community: same Prisma advisory lock leak as enterprise #34

Context

The enterprise tenant just fixed a production incident where the Prisma schema-engine leaked a session-level pg_advisory_lock(72707369) through PgBouncer in transaction mode, blocking all migration attempts for 53 minutes. See PR #213 for the full fix.

Community has the same exposure

• litellm-community/cnpg-cluster-litellm-community/pooler.yaml: poolMode: transaction
• litellm-community/values.yaml: DATABASE_URL goes through -pooler-rw with ?pgbouncer=true
• USE_PRISMA_MIGRATE: "True" — runs prisma migrate deploy on startup
• No DISABLE_SCHEMA_UPDATE — every replica runs migrations
• No PRISMA_SCHEMA_DISABLE_ADVISORY_LOCK — advisory lock is active
• schema.prisma has no directUrl — schema-engine uses DATABASE_URL (through PgBouncer)
• Same false comments about pool mode being session (actually transaction)

Community runs 10 replicas, so a leaked lock here would block all 10 from migrating.

Fix needed

Apply the same changes as enterprise PR #213:

1. Add PRISMA_SCHEMA_DISABLE_ADVISORY_LOCK: "1" to envVars
2. Add DISABLE_SCHEMA_UPDATE: "True" via a ConfigMap + environmentConfigMaps
3. Correct the false comments about pool mode / DIRECT_URL

Assignee

@barckcode