Uh oh!
There was an error while loading. Please reload this page.
docs(release): notes for v2.116.0 - #6346
Conversation
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit:7ade228ec0
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| ### Highlights | ||
| - **`supabase start` and `stop` are 10 seconds faster** — Docker shutdown no longer waits; saves time on every restart. (#6203) |
There was a problem hiding this comment.
supabase stop
PR #6203 explicitly reports that start time is unchanged and that stop --no-backup improved from an average of 12.64s to 5.14s. Claiming both start and stop are 10 seconds faster therefore gives users an incorrect performance expectation; describe the measured roughly 7.5-second stop-only improvement instead.
Useful? React with 👍 / 👎.
| ### Highlights | ||
| - **`supabase start` and `stop` are 10 seconds faster** — Docker shutdown no longer waits; saves time on every restart. (#6203) | ||
| - **Migration testing is faster with shadow database baseline caching** — Your baseline is cached as a PGDATA snapshot, speeding up repeated diffs and migrations. (#6184) |
There was a problem hiding this comment.
The cache added by PR #6184 is disabled unless SUPABASE_SHADOW_CACHE=1 or true is set, so users following these notes receive no caching or repeated-run speedup by default. The highlight should either document the opt-in and the affected commands or be moved to the internal tail until the cache becomes default-on.
Useful? React with 👍 / 👎.
| - **`supabase start` and `stop` are 10 seconds faster** — Docker shutdown no longer waits; saves time on every restart. (#6203) | ||
| - **Migration testing is faster with shadow database baseline caching** — Your baseline is cached as a PGDATA snapshot, speeding up repeated diffs and migrations. (#6184) | ||
| - **Edge runtime bootstrap streams output in real time** — See progress instead of waiting for full completion. (#6273) |
There was a problem hiding this comment.
PR #6273 does not stream progress output to the terminal: it streams the bundled index.ts template into the created container with docker cp. The user-visible fix is that supabase start and functions serve now work with remote Docker or Podman daemons that cannot access client-side bind mounts, so the current promise of real-time output is unsupported.
Useful? React with 👍 / 👎.
| - **`supabase start` and `stop` are 10 seconds faster** — Docker shutdown no longer waits; saves time on every restart. (#6203) | ||
| - **Migration testing is faster with shadow database baseline caching** — Your baseline is cached as a PGDATA snapshot, speeding up repeated diffs and migrations. (#6184) | ||
| - **Edge runtime bootstrap streams output in real time** — See progress instead of waiting for full completion. (#6273) | ||
| - **New `[workers]` configuration section** — Configure workers in `supabase/config.toml`. (#6260) |
There was a problem hiding this comment.
PR #6260 states that no CLI surface consumes [workers] yet and that the worker commands arrive in later changes. Although this release can parse the section, adding it to supabase/config.toml cannot configure or deploy a worker, so advertising it as an available workflow will lead users to expect behavior that is not present in v2.116.0.
Useful? React with 👍 / 👎.
| - **Piped stdin is no longer buffered endlessly** — Fixed memory overflow when piping data to the CLI. (#6290) | ||
| - **Stale database connections are detected and dropped** — Operations no longer hang on dead connections. (#6277) | ||
| - **Edge runtime respects system file descriptor limits** — Prevents "too many open files" crashes. (#6284) |
There was a problem hiding this comment.
PR #6284 addresses Linux hosts whose hard nofile limit is below 65536, where Docker rejects the requested container ulimit and Edge Runtime fails to start; it clamps the request to the host limit. It does not prevent runtime exhaustion or a subsequent "too many open files" crash, so this wording describes a different failure mode from the one fixed.
Useful? React with 👍 / 👎.
| - **Postgres role management is more stable** — Roles are restored correctly after changes. (#6246) | ||
| - **Loopback network proxying is bypassed** — Local requests route directly, improving reliability. (#6283) | ||
| - **Kong local DNS cache is optimized** — DNS resolution is more responsive and consistent. (#6324) | ||
| - **Service startup is more reliable** — Fixed edge cases during start and stop. (#6245) |
There was a problem hiding this comment.
PR #6245 changes only the private @supabase/stack managed-runtime implementation, while the production legacy supabase start and stop paths explicitly do not use that orchestration model. CLI users therefore do not receive the advertised service-startup improvement from this PR; it should be counted as an internal change rather than presented as a CLI bug fix.
Useful? React with 👍 / 👎.
| ### Improvements | ||
| - **`auto_expose_new_tables` defaults to `true`** — Matches your Supabase project behavior by default. (#6337) |
There was a problem hiding this comment.
PR #6337 only changes database setup for fresh volumes; a volume initialized while an unset auto_expose_new_tables meant false keeps its revoked grants after upgrading. Users with those existing projects will not observe the advertised platform-matching behavior until they run db reset or recreate the volume, so the release note needs that migration caveat.
Useful? React with 👍 / 👎.
| --- | ||
| Plus 14 internal improvements and dependency updates. |
There was a problem hiding this comment.
PR #6309 is a user-facing fix for monorepos: functions serve, start, and local functions deploy previously omitted import-map scopes targets outside the Supabase project, causing functions to fail with unresolved relative imports. Treating this as part of the internal tail leaves affected users with no indication that upgrading fixes their broken function workflow.
Useful? React with 👍 / 👎.
| --- | ||
| Plus 14 internal improvements and dependency updates. |
There was a problem hiding this comment.
PR #6300 modifies the production legacy commands db diff, db pull, and db schema declarative generate, including new default SQL formatting and fixes for ownership, extension, ACL, vault, and stalled-load handling. Despite the PR title containing “next,” these paths are user-facing in this release, so burying the change in the internal tail omits relevant output changes and database-workflow fixes.
Useful? React with 👍 / 👎.
| ### Bug fixes | ||
| - **Piped stdin is no longer buffered endlessly** — Fixed memory overflow when piping data to the CLI. (#6290) |
There was a problem hiding this comment.
When a producer continuously writes a single line without a newline, Stream.splitLines still accumulates that unterminated line before data reaches the new 1,024-line bounded queue, so the CLI can still grow without bound and be OOM-killed. PR #6290 explicitly identifies this residual case, making the unconditional claim that piped stdin is no longer buffered endlessly inaccurate.
Useful? React with 👍 / 👎.
Proposed user-facing release notes for
v2.116.0, generated byapps/cli/scripts/propose-release-notes.tsagainsttools/release/release-notes-prompt.md.How to update the notes
Edit
release-notes/v2.116.0.mddirectly on this branch — use the GitHub web editor or push commits torelease-notes/v2.116.0— before approving. The applied notes will reflect the file at the approved commit.How to publish
Approve this PR as a
supabase/cliteam member. The.github/workflows/apply-release-notes.ymlworkflow will then:v2.116.0with the contents ofrelease-notes/v2.116.0.md.release-notes/v2.116.0branch.This PR is not merged — the
do not mergelabel is a reminder. It targetsdevelopso that even an accidental merge never rewritesmain. Nothing is meant to land on any branch.Approvals from anyone outside the
supabase/cliteam are ignored; the workflow will post a comment explaining that and leave the release untouched.How to abandon
Close the PR without approving. The auto-generated semantic-release body for
v2.116.0stays in place.Re-generation
After this PR is closed, rerun the Propose release notes workflow from the Actions tab against
v2.116.0to get a fresh proposal.