One release, sixteen merged PRs across two repos, and a pooler that learned to reshard without blinking.
pgdog v0.1.31 ships zero-downtime resharding and a sixteen-item changelog
The connection pooler's biggest release in weeks lands online resharding, RDS IAM auth, replica lag banning, and a wall of bug fixes.
pgdog v0.1.31 dropped March 5 with a headline feature: online resharding with zero downtime, orchestrated via admin commands. Beyond that, @levkk and contributors added RDS IAM authentication support for server connections (by @Tolsto), a configurable replica ban when replication lag gets too high, and config linting with jsonschema including language server support (by @meskill). The fix list is long and painful: a binary COPY crash from incorrect partial-packet buffering, a segfault in pg_query caused by empty C strings, transactions incorrectly swallowed when advisory locks were used, and partial requests leaking to backends after abrupt client disconnects. Three new contributors made their debuts: @quentindemetz fixed Aurora TLS verify_ca connections, @mnbbrown unblocked LISTEN/NOTIFY when pub_sub_channel_size was zero, and @Tolsto wired up IAM auth.
pgdog gets weighted load balancing the week after release
Not content with v0.1.31, @levkk merged weighted replica routing four days later.
#816 feat: weighted load balancing landed this week, letting operators assign weights to replicas so traffic distribution stops being a coin flip. The feature sits on top of pgdog's existing load balancer and should pair neatly with the new replica-lag banning from v0.1.31. Expect it in the next tagged release.
Partial requests, advisory locks, and the ghost in ClientRead
Three gnarly connection-state bugs got squashed in a single week.
#805 fixed a nasty issue where abrupt client disconnects left partial requests streaming to backends, stranding server connections in ClientRead. #801 corrected transaction state handling when advisory locks were in play — transactions were being buffered in session mode and swallowed outright in transaction mode. And #811 expanded the parameter ignore list so that SET commands like is_superuser no longer blow up the session. Three fixes, three fewer ways for connections to silently rot.
Resharding cutover gets its logging and replication-delay guardrails tightened
Enterprise resharding hooks and cutover fixes followed the release by hours.
#806 fixed logging during cutover and enforced that replication delay must be zero before the switch completes — a guardrail that was missing in the initial implementation. Separately, #813 added enterprise edition hooks for resharding, signaling that pgdog's sharding story is splitting into open-source and commercial tiers.
pgdog docs finally describe resharding and the enterprise tier
Two documentation PRs close the gap between what pgdog can do and what its docs admit.
@levkk merged #57 finish documenting new resharding and #58 Document some of the enterprise features this week. The resharding docs cover the zero-downtime workflow shipped in v0.1.31; the enterprise docs lay out the commercial feature set for the first time. Five open issues remain on the docs backlog, including generating an LLM-friendly text dump at /llms.txt and documenting schema_path behavior in session mode.
Community PR would teach psql's backslash-dt about shard metadata
An outside contributor wants pgdog to show which shard owns which table, right from the psql prompt.
#812 feat: Extend \dt psql command output with shard metadata by @Adi-Goll is open and under review. If merged, running \dt through pgdog would annotate table listings with shard information — a quality-of-life win for anyone debugging cross-shard queries. The PR references issue #709.