A quiet week: three merged fixes, two open PRs, and a documentation backlog that keeps growing.
pgdog stops sending clients two error messages when one will do
A bug that fired duplicate errors on PostgreSQL connection drops has been squashed.
#841 fix: sending two errors to client when Pg shuts the connection down addresses a case where pgdog would relay two error packets to the client whenever a backend PostgreSQL instance dropped its connection. The fix, merged this week by @levkk, ensures only a single, accurate error reaches the client. One less thing for application-side retry logic to untangle.
Replication state sync gets its wires uncrossed
A fix to replica tracking lands alongside the error deduplication work.
@levkk merged #833 fix: fix replication state sync, correcting how pgdog tracks the replication state of backend servers. Getting this wrong can mean stale reads or misrouted queries in a read-replica topology. The fix was tight — three commits, no fanfare.
Passthrough users may soon change their own passwords without locking themselves out
An open PR tackles a sharp edge in passthrough authentication.
#842 feat: allow passthrough user to change password, opened by @levkk, addresses a reported issue where changing a user's password with passthrough auth in plain mode would prevent new connections entirely. The fix lets pgdog accept the new credentials gracefully. Still open, not yet merged.
Community contributor tunes pool wait-time metrics
An outside fix recalculates how long queries actually spend waiting for a connection.
#836 fix(metrics): wait_time for pool calculation, opened by contributor meskill, proposes a correction to how pgdog computes pool wait time. Accurate wait-time numbers matter for capacity planning — bad math here means bad dashboards. The PR is still under review.
Five new documentation issues pile up with no PRs in sight
The docs backlog grows: session behavior, timeouts, LLM-friendly output, and cross-links to upstream PostgreSQL.
The docs repo saw two commits this week but no merged or open PRs. Meanwhile, five issues accumulated: documenting schema_path behavior in sessions, idle_session_timeout, generating an LLM-friendly text endpoint at /llms.txt, and adding links from the Query Protocol and Cross Shard pages to upstream PostgreSQL documentation. The to-do list is writing itself faster than the docs are.