$ the-wire · showcase
Realtime replay ordering reversed, Studio drops extension schema picker
By RepoJournal · Filed · About Supabase · Composed from the cited sources · methodology
11 people shipped this
edgurgel
@edgurgel
2 cited
tirthraj01
@tirthraj01
1 cited
Andrew Valleteau
@avallete
1 cited
Joshen Lim
@joshenlim
1 cited
hsusul
@hsusul
1 cited
supabase-cli-releaser
@supabase-cli-releaser
Danny White
@dnywh
Saxon Fletcher
@SaxonF
github-actions[bot]
@github-actions[bot]
dependabot
@dependabot
seangeo
@seangeo
Realtime fixes a message replay bug that reversed send order inside a single transaction, while Studio removes the schema selector from the extension enable dialog to stop installs into `public` that break RLS.
fix: replay messages sent in one transaction in send order supabase/realtime
Because `realtime.send` and `realtime.send_binary` omit `inserted_at`, the column default `now()` stamps every message in a transaction with the same timestamp; `Realtime.Messages.replay/5` then orders by `inserted_at DESC`, applies the limit, and reverses, so tied rows come back flipped. Messages sent `1..5` in one transaction used to replay in the wrong order and now replay in send order.
Remove schema selector in DB extensions enable dialog (#51440) supabase/supabase
Installing an extension into a schema other than `extensions` left it owned by `supabase_admin`, which blocked enabling RLS, made the extension immovable, and kept the security advisor flagging it. The dialog no longer offers the choice, so users can't pick `public` in the first place.
feat(stack): allow repeatable --stack-id in stack destroy (#6964) supabase/cli
`stack destroy` accepts `--stack-id` more than once and tears down every listed stack after a single confirmation. All ids resolve before the prompt, so a typo destroys nothing; one failure doesn't halt the rest, and the command still exits 1 with an error naming the failed ids and a retry command.
fix: improve authorize function for read checks supabase/realtime
Read authorization checks now look up by the primary key instead of the bare `id`, reaching for `id` plus `inserted_at` which are already known. Avoiding the scan across all partition indices is the stated reason, so authorize calls should get cheaper.
fix(cli): classify new local stack failures in telemetry (#7044) supabase/cli
Failures from the new local stack (`supabase stack *` and stack-backed `db reset`, `db dump`, `db diff`, `test db`, `functions serve`) previously landed in telemetry as category `unknown`, with start and destroy failures sharing a fingerprint. They now carry a cause, and a port conflict counts as user-actionable only when the user pinned the port.
chore: e2e tests on all the PGs supabase/realtime
Elsewhere in realtime: e2e tests now run against all supported Postgres versions, minus pg14 for now, with the PG client swapped because Bun didn't handle `sslnegotiation=direct`. Slim-services published realtime v2.143.3-r0, analytics v1.53.0-r0, and storage v1.80.1-r0 as multi-platform images, and the CLI's OpenAPI sync and a tinypool bump to 2.2.0 were the rest of the day.
