Authentication has been failing for more than 24 hours.

[pedrohenriquescav-blip]

Project ref: ysoocfjamvyfumcavknb
Region: sa-east-1
Project: Central Principal

Authentication has been failing for more than 24 hours.

Last 24h Auth logs:

• /token refresh_token: 107 HTTP 500 unexpected_failure
• /token refresh_token: 29 HTTP 504 request_timeout
• /token password: 18 HTTP 500
• /token password: 5 HTTP 504
• PostgreSQL: 6 errors for supabase_auth_admin, SQLSTATE 57014

Storage also intermittently returned HTTP 544 DatabaseTimeout.

Please investigate project-specific Auth/Postgres connection degradation and restore reliable login/refresh. No schema or production changes were made.

[potomyznaesh]

SQLSTATE 57014 is query_canceled — here almost certainly canceling statement due to statement timeout. Together with the 504s on /token and Storage's 544 DatabaseTimeout, it points at the database being too slow to respond; Auth is just the first thing to notice. I'd check, in this order:

1. Database resources. Dashboard → Reports → Database: CPU, memory and especially Disk IO. On the smaller compute sizes, once the burst IO budget is exhausted every query slows down sharply and Auth starts hitting its timeouts.

2. Long-running or blocking queries. In the SQL editor:

select pid, usename, state, wait_event_type, wait_event,
       now() - query_start as duration, left(query, 120) as query
from pg_stat_activity
where state <> 'idle'
order by duration desc nulls last;

Anything running for minutes or holding locks on auth.* tables can be stopped with select pg_terminate_backend(<pid>);.

3. Your own code on the Auth path — the most common cause of unexpected_failure on /token:

• a Custom Access Token Hook (Auth → Hooks) runs on every refresh_token and password grant, so one slow query inside it breaks every login;
• triggers on auth.users: a password login updates last_sign_in_at, so AFTER UPDATE triggers (syncing to profiles, pg_net/http calls, etc.) run on every sign-in.

Temporarily disabling the hook/trigger and watching whether /token recovers is a quick way to confirm.

4. Restart. Settings → General → Restart project clears stuck connections. If it only helps for a while, the root cause is in 1–3.

If all of the above looks clean, attach the CPU/IO graph and the pg_stat_activity output to your support ticket together with the project ref — it speeds up the investigation a lot.