Doctrine

ORA-2026-0030: Direct SQL unreachable — use pooler

fleet-opssupabasesqlnetwork-boundary

ORA-2026-0030: Direct SQL unreachable — use pooler

Rule

Never construct psql connection strings to db.*.supabase.co. These hosts publish only AAAA (IPv6) DNS records. This network has no IPv6 route. Every direct psql connection fails with "No route to host."

Use supabase-psql <project> instead (orbit|camber|heartwood). It routes through the Supabase connection pooler which has IPv4 addresses.

Pooler mapping

ProjectRegionPooler host
orbit (epxbhfxvfjuycoxjcpag)us-east-2aws-1-us-east-2.pooler.supabase.com
camber (rjhdwidddtfetbwqolof)us-west-2aws-0-us-west-2.pooler.supabase.com
heartwood (wdebxkpingkraxvkzobn)us-west-2aws-0-us-west-2.pooler.supabase.com

Evidence of fleet cost

Fleet feed analysis found 13 distinct signal events over 27 days where agents hit this failure:

  • First signal: 2026-03-22 (CODEX: "Without direct SQL... I cannot produce
  • a valid write identity")

  • Explicit IPv6 diagnosis: 2026-03-25 (EHOSTUNREACH) and 2026-04-17
  • ("No route to host on IPv6")

  • One multi-hour outage investigation (2026-03-24, 15+ feed posts) where the
  • root cause was never identified

  • Multiple silent REST fallbacks where agents switched transport without
  • escalating the pattern

  • Auth failures misattributed to credentials when the actual cause was
  • routing (ECIRCUITBREAKER on unreachable host)

Every instance: agent hit the wall, burned 2-5 minutes of context, worked around it, moved on. Pattern repeated across sessions for 27 days.

Why ORA missed it

ORA's 0/18 self-initiation rate on doctrine (all 18 prior doctrines were Chad-initiated) is the systemic gap. The fleet feed contained the signal; no mechanism existed to correlate repeated workarounds into a systemic root-cause investigation. This doctrine is evidence for the need for proactive ORA pattern detection (proposed but not yet built).