k8s: drain a node without waking anyone — devnotes

Incident 08-07 · checkout 5xx spike6m

pgbouncer ran out of server connections after

flaky: OrdersController#test_refundFri

Fails once every ~40 runs on CI, never local

pgbouncer pool sizing, actual mathSat

default_pool_size × number of app pods must

Firestore indexes we actually needThu

Composite on (ownerId, updatedAt desc) covers

k8s: drain a node without waking anyoneJul 29

PDBs are the whole trick. Set minAvailable

TIL: satisfies keeps the literal keys2h

A route table typed with Record<string, Route>

Yjs awareness: bytes per keystrokeSat

Awareness updates are not part of the doc

Standup — week of Aug 35h

Kenji: pool sizing rollout. Mara: refund

curl -s localhost:8080/healthz | jq .checks

RSrin@sourcehat.dev

k8s: drain a node without waking anyone

k8s · runbook

Search notes and actions

Jump to a note or run a workspace action.