$ the-wire · showcase
pg_net moves worker state out of shared memory, pgmq untangles drop_queue
By RepoJournal · Filed · About Supabase · Composed from the cited sources · methodology
A pg_net background worker could run its exit cleanup against another worker's stale fds and shared state, and pgmq's two drop_queue overloads are finally both extension-owned, so an unqualified call resolves again.
fix: undefined behaviour in worker supabase/pg_net
The `epfd` and `curl_mhandle` globals are moved out of shared memory (epfd becomes a static in event.c), because a worker that terminated before initializing them left net_on_exit cleaning up the previous worker's values. The stale path also closed a stale fd, which could take down an unrelated descriptor.
fix(pgmq): fix drop_queue overload ambiguity supabase/postgres
pgmq has shipped both drop_queue(text) and drop_queue(text, boolean) since 1.5.0, which breaks any unqualified reference to the function. after-create.sql now always detaches and recreates both signatures explicitly instead of branching on extversion, and keeps both attached to the extension so DROP EXTENSION removes them cleanly.
fix(stack): recreate containers whose published host port was taken before start (#7081) supabase/cli
Containers publish with an engine-chosen host port (127.0.0.1::<port>); Podman picks that port at create by opening and closing a socket, and pasta only binds it at start. Anything grabbing the port in between, such as an outbound connection from another stack, made the service fail to start. Supabase CLI now recreates those containers.
fix(cli): name unclassified defects in the error fingerprint (#7067) supabase/cli
Unhandled crash defects that match no classification used to land under one bare error:Defect fingerprint. They now fingerprint as error:Defect:<identity>, so unrelated crashes stop collapsing into a single bucket and the most frequent one can be picked out of telemetry.
test(stack): wait on podman container removal instead of its rotating event log (#7091) supabase/cli
Also in the CLI: the Podman sweep integration test waited on a specific `remove` event from `podman events`, and Podman's shared event log rotates. A reader that hadn't polled yet skipped the event and the wait never completed.
feat(storage): archived objects data layer supabase/supabase
Studio's storage versioning work adds the archived-objects data layer: queries for archived objects and versions, plus restore, permanent purge, and single-version delete mutations. The companion UI PR adds a Show archived toggle to the file explorer, with the preference stored per project.
docs: add introduction blog post for .NET SDK supabase/supabase
The rest is groundwork and reach: a getting-started blog post for the .NET SDK v8, a .NET option in the Dashboard Connect flow, an account-settings apps page for seeing and disconnecting your own OAuth grants (mock data, behind a flag on staging), and CI moving from ap-southeast-1 to us-east-1.
