Skip to content

chore: production deploy - #7653

Open
supabase-cli-releaser[bot] wants to merge 55 commits into
mainfrom
develop
Open

supabase-cli-releaser[bot] wants to merge 55 commits into
mainfrom
develop

Conversation

@supabase-cli-releaser

Copy link
Copy Markdown
Contributor

grdsdev and others added 2 commits November 16, 2026 18:08
…en 0.4.0 (#7608)

## Summary

`supabase gen types --lang swift` now builds the `supabase-typegen`
generator from supabase-swift's source, through `@supabase/typegen`
0.4.0 (SDK-3045, supabase/sdk#361). This PR does two things:

1. **Passes `HOME` and `XDG_CACHE_HOME` to the Swift driver.** The
driver reads them from the filtered environment the generator layer
gives it, to find its cache of built binaries. Before this PR, the layer
passed only `PATH`, `PATHEXT` and `ComSpec`. The cache path became
relative and resolved against the working directory, so binaries landed
inside the user's project. Other languages do not change: the spawned
tool already inherits the full environment, and the Dart driver reads
nothing from this object.
2. **Bumps `@supabase/typegen` from 0.3.4 to 0.4.0.** With it, the Swift
driver:
- builds `supabase-typegen` from source at a supabase-swift commit, and
caches the binary by commit. It no longer downloads a prebuilt binary;
- picks the commit from the new `--swift-typegen-source
<version|branch|sha|path>` flag, else from the supabase-swift pin in the
nearest `Package.resolved`, else from the latest 3.x release (with a
notice on stderr);
- drops `SUPABASE_TYPEGEN` and the `PATH` fallback. The flag's
local-path form covers what `SUPABASE_TYPEGEN` did.

The CLI needs no code for the new flag. It renders language flags from
the registry's option specs, so `--swift-typegen-source` appears with
the bump, like `--swift-access-control`. Like every language flag, it is
in the mutex group with `--remote`, `--linked` and `--project-ref`.

The driver change is not marked as breaking. The Swift generator is
still alpha in supabase-swift, so 0.4.0 is a minor bump.

## Linked issue

Linear: CLI-3138 (replaces CLI-3134). Driver: SDK-3045.

- [x] The linked issue is **open** and carries the
`open-for-contribution` label (or I'm a Supabase maintainer).

## How it was tested

- End to end, with the CLI from this branch (`bun apps/cli/src/main.ts
gen types --lang swift --db-url …`), a Postgres 17 container with a
table, an enum, a view and a function, and Swift 6.4 on macOS. The cache
was a fresh `XDG_CACHE_HOME`.

  | Case | Result |
  | -- | -- |
| `--swift-typegen-source main`, no `Package.resolved` | Built
supabase-swift `main` (`f4f132a`) from source in 212 s, cached it, and
printed the `@Table` structs, the enum and the function. |
| Same, again, with `--swift-access-control public` | Used the cached
binary in 2 s. The output is `public`. |
| `Package.resolved` two levels up pins `main` at `f4f132a` | `Using
branch main at f4f132a…, pinned in ../../Package.resolved.` Cached, 2 s,
same output. |
| No pin, no flag | Fails: supabase-swift has no 3.x release yet. The
message names `--swift-typegen-source`. |
| `--swift-typegen-source f4f132a` | Fails before git runs: an
abbreviated SHA is refused. |
| `--swift-typegen-source 3.0.0-alpha.1` | Fails: no tag
`v3.0.0-alpha.1` yet. |
| `--swift-typegen-source ./supabase-swift` (a clone) | Built in the
checkout in 203 s. Same output as the other builds. The checkout has no
git changes after the build. |

No run wrote files into the project folders, and no scratch folders were
left in `$TMPDIR`.

- Unit and integration tests:
- `types.generator.integration.test.ts` runs a fake `supabase-typegen`
from the cache under `XDG_CACHE_HOME`, and under `HOME`. The cache is
now keyed by the pinned commit SHA.
- `types.integration.test.ts` checks that `--swift-typegen-source`
reaches the generator, and that it is rejected with `--project-ref`.
- `types.generator.unit.test.ts` checks that only the Swift generator
gets the flag.
- All unit tests pass (7655). The integration tests for
`src/commands/gen` pass (271).
- In the full integration run, 3 TLS tests in
`db-connection.sql-pg.integration.test.ts` fail on my machine. They
passed in CI on the previous head, and this PR does not touch that code.
A compute logs test failed once and passes when run alone.
- `oxfmt`, `oxlint`, `tsc` and `knip` are clean. A frozen `pnpm install`
passes.

## Checklist

- [x] The PR title follows [Conventional
Commits](https://www.conventionalcommits.org/) (e.g. `fix(cli): …`).
- [x] Tests added or updated for the change.
- [x] Relevant package tests pass for every touched workspace, and the
touched TypeScript workspace type-checks. Locally, `lint:check`,
`fmt:check` and `knip:check` pass. `lint:actions:check` did not run,
because `jactionlint` is not installed, and this PR changes no workflow.

## Why now

supabase-swift v3 launches as Public Alpha on 2026-12-11. Its demo flow
is `supabase gen types --lang swift`. This PR must ship in a CLI release
before that date.

<!-- ci retrigger -->

---------

Co-authored-by: Andrew Valleteau <avallete@users.noreply.github.com>
Co-authored-by: avallete <8771783+avallete@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
This bumps storage from v1.96.0-r0 to v1.96.2-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.96.2-r0

Planned after supabase/slim-services published v1.96.2-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
@supabase-cli-releaser supabase-cli-releaser Bot added the do not merge Approve to apply; do not merge. label Nov 17, 2026
@supabase-cli-releaser
supabase-cli-releaser Bot requested a review from a team as a code owner November 17, 2026 02:16
supabase-cli-releaser Bot and others added 6 commits November 17, 2026 02:36
This bumps realtime from v2.167.0-r0 to v2.168.0-r0.


https://github.com/supabase/slim-services/releases/tag/realtime-v2.168.0-r0

Planned after supabase/slim-services published v2.168.0-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
`functions deploy`, `projects create` and `bootstrap` call the
Management API with `executeRaw` and build their own status errors, so a
plan-gate denial (`entitlement_required`) reached the user without the
upgrade hint, the JSON `error.entitlement` or the
`cli_upgrade_suggested` event that `mapHttpError` callers already get. I
moved the envelope read into one shared function,
`planGateFromResponse`, and both paths now use it.

**Changed:**
- **Plan-gated project create**: I made `projects create` and
`bootstrap` (which shares the create step) handle a 4xx that carries the
envelope: each prints the upgrade hint once, adds `entitlement` to the
JSON error, fires `cli_upgrade_suggested` and classifies the error as
`plan_limit`.
- **Plan-gated function deploy**: I applied the same handling to
function create, update, API deploy and bulk update.
- **`mapHttpError`**: I switched it to read the envelope from the body
text it already holds, through the same shared function. Its behavior is
unchanged.

**Note:** a `--use-api` deploy of several functions still reports its
failures as one aggregate error with no entitlement. I left that path
for a follow-up.

## Linear
- fixes GROWTH-1484
The gateway CORS live test intermittently failed on the live job in two
ways, both from a brief 403 on the freshly created staging project:

- Staging answered the two \`OPTIONS /rest/v1/\` probes with a bare 403
and no CORS headers, so the stack-versus-staging comparison failed.
- \`functions deploy\` got a 403 from the Management API's
list-functions call.

The staging readiness gate now also waits for both REST preflight probes
to stop answering 403 before staging is recorded, and \`functions
deploy\` is retried (5s apart, up to 12 times) when it fails with a 403.
Other deploy failures still fail immediately, and the
stack-versus-staging assertion is unchanged.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
This bumps postgrest from v16.5-r0 to v16.6-r0.


https://github.com/supabase/slim-services/releases/tag/postgrest-v16.6-r0

Planned after supabase/slim-services published v16.6-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## TL;DR
The routed-peer plain probe now ignores SIGPIPE, so it reports the
proxy's close instead of dying.

- `HttpProxy.integration.test.ts › gives a routed peer inside the bridge
network no plain byte, but TLS`: the namespace `bash` probe failed with
`Process interrupted due to receipt of signal: 'SIGPIPE'`. The proxy
closes a plain connection from a non-loopback peer. When that close
arrives before `printf >&3` writes the request, bash dies of SIGPIPE,
and the `&&` chain never reaches the read. The probe now uses the
api.tls probe pattern from #7590: `trap '' PIPE`, a write that can fail
without stopping the chain, and the read that reports the close.

## ref
spotted in:
https://github.com/supabase/cli/actions/runs/43972802044/job/133669050994

---
_Generated by [Claude
Code](https://claude.ai/code/session_01MBEk1LoYxzbfzYDtuz4z4F)_

Co-authored-by: Claude <noreply@anthropic.com>
This PR was automatically created to sync the generated `@supabase/api`
package with the latest Management API OpenAPI document.

Changes were detected in the upstream OpenAPI documents exposed by
`https://api.supabase.com/api/v1-json` and
`https://api.supabase.com/api/v2-json`.

Co-authored-by: jgoux <1443499+jgoux@users.noreply.github.com>
supabase-cli-releaser Bot and others added 4 commits November 17, 2026 10:33
This bumps storage from v1.96.2-r0 to v1.96.3-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.96.3-r0

Planned after supabase/slim-services published v1.96.3-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This bumps realtime from v2.168.0-r0 to v2.170.0-r0.


https://github.com/supabase/slim-services/releases/tag/realtime-v2.170.0-r0

Planned after supabase/slim-services published v2.169.0-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This bumps storage from v1.96.3-r0 to v1.96.4-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.96.4-r0

Planned after supabase/slim-services published v1.96.4-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Summary

The published config schema
(`https://supabase.com/schemas/config/v1.json`) sets
`additionalProperties: false` at the root but never declared `$schema`.
VS Code's JSON validator doesn't exempt `$schema`, so a `config.json` or
`config.jsonc` that points at our schema shows `Property $schema is not
allowed.` on that line. The same applies to a top-level `$schema` key in
TOML for editors that read the schema. The CLI itself already accepts
the key; only the editor schema rejected it.

This declares `$schema` as the first root property: an optional string
with the hosted URL in `examples`. `examples` still gives completions in
VS Code and a hint to agents reading the schema, without rejecting other
values such as a local copy or a `node_modules` path. `CliConfigSchema`
is unchanged, so `$schema` doesn't become a decoded config field.

- `toCliConfigJsonSchema()` adds the property
- `apps/cli/schemas/config/v1.json` is regenerated with `pnpm build`
(config) and `pnpm schemas:generate` (cli). The diff is only the new
property; project-config and lock outputs are unchanged
- A new unit test checks the property shape, that it isn't required, and
validates a document with `$schema` against the published schema via the
existing Ajv helper

The lock schema's `$schema` handling is discussed separately in
CLI-3145.

## Linked issue

No issue; found while discussing CLI-3145.

- [x] The linked issue is **open** and carries the
`open-for-contribution` label (or I'm a Supabase maintainer).

## Checklist

- [x] The PR title follows [Conventional
Commits](https://www.conventionalcommits.org/) (e.g. `fix(cli): …`).
- [x] Tests added or updated for the change.
- [x] From the repository root, `pnpm check:all` passes; relevant
package tests pass for every touched workspace, and `pnpm types:check`
passes for each touched TypeScript workspace (or workspace declaring
it).
supabase-cli-releaser Bot and others added 12 commits November 17, 2026 12:37
This bumps pgmeta from v0.100.0-r0 to v0.100.1-r0.


https://github.com/supabase/slim-services/releases/tag/pgmeta-v0.100.1-r0

Planned after supabase/slim-services published v0.100.1-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
… release (#7666)

## What

Moves the `supabase/setup-cli` smoke test out of the release pipeline
and into a daily scheduled workflow.

## Why

The post-release smoke test installed through setup-cli seconds after
the npm publish, while the registry could still serve an older package
list. npm then skipped the new `@supabase/cli-<platform>` optional
dependency without failing, the npm shim could not find a binary, and
setup-cli fell back to the GitHub release. Slack reported this as a
setup-cli npm fallback on most releases, although it was only
propagation delay and the release itself was fine.

## How

- `setup-cli-smoke-test.yml` runs daily. It resolves the newest version
on each npm dist-tag that ships the `@supabase/cli-*` platform packages
(`latest`, `beta`, `next`, `v<N>.stable`) and skips versions published
less than an hour earlier. Each version runs the existing setup-cli
v1/v2/v3 matrix on Linux, macOS, Windows and Alpine, plus the forced
GitHub release fallback job.
- Scheduled runs post to Slack when a job fails or when setup-cli v3
falls back from npm. `workflow_dispatch` with a `version` tests one
release without posting.
- The release pipeline no longer calls the smoke test: the
`setup-cli-smoke` job, its fallback output, and the separate fallback
Slack notice are removed, along with the fallback notice on success
messages.
- `wait-for-npm-version.sh` is removed, since tested versions are
already an hour old.
- The release process doc describes the daily check.

`v2.x` still runs the post-release smoke test from its own copy of the
release pipeline; a backport that removes the `setup-cli-smoke` job
there stops the same alerts on maintenance releases.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
## TL;DR
Owner integration tests that start real services ran on the virtual
`TestClock`, so any retry in their path could hang until the test
timeout.

- `Owner.integration.test.ts › credential lookup leaves a fresh stack
untouched for custom database credentials`: hung until its 30s timeout
once, green on rerun of the same SHA. The test ran under `it.effect`,
whose `TestClock` never advances, while the database readiness probe
(`Database.ts`) and the lock and port-reservation acquisitions retry on
`Schedule.spaced`. When the first probe or lock attempt failed, the
retry waited forever. The tests in this file that do real owner, SQLite,
file and process I/O now run under `it.live`, like their siblings. No
test in the file relies on virtual time.

## ref
spotted in:
-
https://github.com/supabase/cli/actions/runs/42807806171/job/129801375563
(failing)
-
https://github.com/supabase/cli/actions/runs/42807806171/job/129803236031
(green rerun)

---
_Generated by [Claude
Code](https://claude.ai/code/session_01MBEk1Pf3phzjWscZfmvHZY)_

Co-authored-by: Claude <noreply@anthropic.com>
This bumps storage from v1.96.4-r0 to v1.97.0-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.97.0-r0

Planned after supabase/slim-services published v1.97.0-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Summary

Extend the existing AI review with advisory feedback on changed CLI
reference prose, using the Supabase docs team’s current style guide
rather than the retired prose linter.

A trusted preparation job resolves one upstream revision and supplies
the same readable guidance snapshot to both reviewers and adjudication.
Findings must quote changed prose and cite a rule at that revision.
Retrieval and delivery failures are reported without blocking ordinary
code review or claiming a style pass.

A local CLI reference skill shares the review procedure without copying
upstream style rules. Existing review authorization and trigger behavior
remain unchanged.

Closes
[CLI-2681](https://linear.app/supabase/issue/CLI-2681/apply-upstream-docs-guidance-in-cli-reference-ai-review).
## Summary

Validate the generated CLI reference before writing a local preview or
starting Git operations for the docs-site PR. The publisher now checks
nested document fields and compares command IDs and subcommand links
against the documented command tree, replacing the fixed command-count
threshold.

Missing, duplicate, unexpected, or malformed entries stop publishing
with an error. Unlisted and explicitly excluded commands remain outside
the reference.

Closes
[CLI-2680](https://linear.app/supabase/issue/CLI-2680/validate-generated-files-before-pushing).
<!--
Before opening this PR, confirm the linked issue is open and carries the
`open-for-contribution` label. PRs from external contributors that don't
follow
the workflow in CONTRIBUTING.md are closed automatically.
@supabase members working from Linear tickets are exempt.
-->

## Summary

addresses a few `mise` housekeeping tasks that i noticed when running
`mise install`:
- [x] updates to the recommended backend for `jactionlint` installations
- [x] upgrades our `mise.lock` lockfile version
- [x] updates our `mise` `min_version` and our `mise` version in CI

## Checklist

- [x] The PR title follows [Conventional
Commits](https://www.conventionalcommits.org/) (e.g. `fix(cli): …`).
- [x] Tests added or updated for the change.
- [x] From the repository root, `pnpm check:all` passes; relevant
package tests pass for every touched workspace, and `pnpm types:check`
passes for each touched TypeScript workspace (or workspace declaring
it).
This bumps realtime from v2.170.0-r0 to v2.171.0-r0.


https://github.com/supabase/slim-services/releases/tag/realtime-v2.171.0-r0

Planned after supabase/slim-services published v2.171.0-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This PR was automatically created to sync the generated `@supabase/api`
package with the latest Management API OpenAPI document.

Changes were detected in the upstream OpenAPI documents exposed by
`https://api.supabase.com/api/v1-json` and
`https://api.supabase.com/api/v2-json`.

Co-authored-by: jgoux <1443499+jgoux@users.noreply.github.com>
## Summary

Switches the Codex side of the AI review from `gpt-6.1-sol` to
`gpt-6.2-sol`.

- `.github/ai-models.json`: `codex.sol` is now `gpt-6.2-sol`
- `ai-review.yml`: Codex CLI pin goes from 0.159.2 to 0.169.1, the first
stable release whose model catalog carries `gpt-6.2-sol` (0.169.0
doesn't have it)

The codex-action v1.11 pin and its Dependabot ignore are untouched.
`codex-rs/exec/src/cli.rs` only gained flags between 0.159.2 and
0.169.1, so everything the action passes still parses.

`effort: high` stays explicit - 6.2-sol's catalog default is `low`. Its
default context window is 500k (max 872k), so very large diffs have less
headroom than before.

This hasn't run against a real PR yet. I'd like a `workflow_dispatch`
run on one before we trust it on the automatic path, which also confirms
our `OPENAI_API_KEY` has access to the model.

## Linked issue

N/A

- [x] The linked issue is **open** and carries the
`open-for-contribution` label (or I'm a Supabase maintainer).

## Checklist

- [x] The PR title follows [Conventional
Commits](https://www.conventionalcommits.org/) (e.g. `fix(cli): …`).
- [ ] Tests added or updated for the change.
- [x] From the repository root, `pnpm check:all` passes; relevant
package tests pass for every touched workspace, and `pnpm types:check`
passes for each touched TypeScript workspace (or workspace declaring
it).
This bumps realtime from v2.171.0-r0 to v2.172.0-r0.


https://github.com/supabase/slim-services/releases/tag/realtime-v2.172.0-r0

Planned after supabase/slim-services published v2.171.1-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Summary

Mirrors supabase/supabase#53589 and refreshes the rest of the
mise-managed toolchain:

- Node 24 → 26 LTS (26.13.1), including the npm-only publish job in
`release-config.yml`
- pnpm 12.15.0 → 12.19.0
- jactionlint 2.3.0 → 2.5.0
- `jdx/mise-action` v5.2.2 → v5.2.7
- `mise.lock` regenerated

Each version is the newest release past the repository's 7-day cooldown
(Dependabot `cooldown` / pnpm `minimumReleaseAge`). mise 2026.11.6, Bun
1.4.4 and shellcheck 0.11.0 are already current under that rule.

This touches `.github/actions/setup`, which is release infrastructure,
but only the `mise-action` pin changes; `v2.x` does not need a backport.
supabase-cli-releaser Bot and others added 9 commits November 18, 2026 19:47
This adds postgres 17.11.0.020-orioledb-r0 as a new release line.


https://github.com/supabase/slim-services/releases/tag/postgres-17.11.0.020-orioledb-r0

Planned after supabase/slim-services published 15.20.0.001-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This bumps postgres from 15.20.0.000-r0 to 15.20.0.001-r0.


https://github.com/supabase/slim-services/releases/tag/postgres-15.20.0.001-r0

Planned after supabase/slim-services published 15.20.0.001-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This bumps postgres from 17.12.0.000-r0 to 17.12.0.001-r0.


https://github.com/supabase/slim-services/releases/tag/postgres-17.12.0.001-r0

Planned after supabase/slim-services published 15.20.0.001-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Remove the contribution gate’s `pull_request_target` trigger and its
single-PR job so untrusted PR events no longer start this privileged
workflow.

Keep the manual sweep and dry-run option. Update contributor and
maintainer documentation to explain that contribution-policy enforcement
now requires a maintainer to run the sweep.
Restore automated contribution-gate enforcement after #7693 removed the
per-PR trigger. Sweep every 15 minutes, with manual sweeps and dry-run
support retained.

Run only trusted default-branch code and treat PR metadata as API data.
Keep scoped token permissions, disable the Bun executable cache, and
bound each run to 15 minutes. Serialize sweeps and queue pending runs so
a scheduled run does not replace a pending manual sweep.

Continue processing later PRs after an individual failure, then fail the
run with the affected PR numbers. Reuse existing bot-authored comments
when the explanation is unchanged, skip linked-issue queries for exempt
authors, and cache permission results within each sweep.

Update contribution documentation to describe scheduled enforcement,
possible scheduling delays, and re-evaluation after issue-label changes
or PR reopening.
This bumps storage from v1.97.1-r0 to v1.97.2-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.97.2-r0

Planned after supabase/slim-services published v1.97.2-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This bumps edge-runtime from v1.79.1-r0 to v1.79.2-r0.


https://github.com/supabase/slim-services/releases/tag/edge-runtime-v1.79.2-r0

Planned after supabase/slim-services published v1.79.2-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This bumps realtime from v2.172.0-r0 to v2.172.1-r0.


https://github.com/supabase/slim-services/releases/tag/realtime-v2.172.1-r0

Planned after supabase/slim-services published v2.172.1-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…ilures (#7681)

A `functions deploy --use-api` run that deploys several functions
reports its per-function failures as one aggregate error. That aggregate
dropped the `entitlement` each failure carries, so a deploy denied on
plan showed no upgrade hint, no `error.entitlement` in JSON and no
`cli_upgrade_suggested`, and it classified as `api_status`. I left this
path out of #7632. I also reworded the text-mode hint, which told every
user to upgrade their plan.

**Changed:**
- **A plan-gated multi-function deploy now behaves like a
single-function one**: `FunctionsDeployFailedError` exposes the
entitlement of its first plan-gated cause, so the shared output and
telemetry paths print the hint, add the JSON field, fire the event and
report `plan_limit`. The command has no hint or event code of its own.
- **The text-mode hint no longer says "Upgrade your plan"**: it now
reads "Your organization's plan does not include this feature. Plans and
add-ons: <url>". Some gates clear without a plan upgrade (a paid add-on
for custom domains, a custom SMTP provider for email templates), and the
server message printed under the hint already names the right fix, so
the two lines disagreed. JSON output is unchanged.

**Note:** one gated function is enough. A run where one upload is denied
on plan and another fails for any other reason, a 5xx or a CLI error
included, reports `plan_limit`, where the aggregate otherwise reports
its most severe cause. A run with no plan-gated failure keeps its
current message and classification.

## Linear
- fixes GROWTH-1504
7ttp and others added 6 commits November 19, 2026 07:26
This bumps storage from v1.97.2-r0 to v1.97.3-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.97.3-r0

Planned after supabase/slim-services published v1.97.3-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…7711)

## Summary

Recent `release` runs on `next` failed in stack integration and
whole-stack E2E tests. The release workflow forces the full test matrix,
so intermittent failures there block the release. This PR fixes one
production gap in the Docker storage helpers. It also hardens the flaky
tests and gives them better diagnostics.

### Storage: retry a helper exec the engine lost
(`DockerDatabaseStorage.ts`)

- **Change:** Both helper paths now retry the script once with a fresh
helper when the helper is lost:
- The container is missing or not running, so the engine refused the
exec and the script never ran.
- An exec exits non-zero with empty stderr, and inspect then shows the
helper is missing or stopped. If the helper is still running, the
original error is reported unchanged.

- **Shared-helper path:** this replaces the old `no such container|is
not running` match.
- **Per-storage path:** this path is used when no helper registry is
passed, and it had no retry before.
- **Snapshot programs:** a helper lost during the exec is dropped, but
the program is not replayed. It fails with a `Helper … stopped during
exec` error.
- **Why:** A CI run failed when a helper exec exited with code 128 and
empty stderr. Every helper script writes its own failures to stderr, so
a silent non-zero exit from a helper that has stopped means the engine
lost the exec together with its helper. Setup scripts can run twice
safely. Snapshot programs cannot: a replay would find the partial output
of the lost run (for example, an existing copy destination) and fail
with a misleading error. Checking the helper's state first keeps a
silent failure inside a healthy helper from being retried and hidden.
Before this change, losing the helper on the per-storage path was fatal.
- **Test:**
- The "helper removed externally" integration test now runs for both
shared and per-storage helpers. The per-storage case fails without the
retry.
- The fake container engine tracks each helper's state, so an exec only
succeeds against a helper that was created and started.
  - New tests on both paths cover a silent exec failure:
    - When the helper is gone, the script is retried on a new helper.
- When the replacement helper is lost too, the operation fails after
exactly two execs.
- When the helper still runs, the original error is reported without a
retry.
    - A lost snapshot program runs only once.

### Whole-stack Analytics probe: retry transient failures
(`tests/whole-stack/analytics.ts`)

- **Change:**
- 5xx responses are now classified as `post-unavailable` and
`query-unavailable`, and the response body is kept for diagnostics. If
the body cannot be read, the error still carries the status, so the
retry decision is unchanged.
- These errors and dropped connections are retried at 1-second
intervals, up to 30 times.
- The retry applies to both marker posts and marker queries. That
includes the cross-stack isolation check, which reuses
`queryAnalyticsMarker`.
- **Why:** Analytics can drop connections or return 5xx while it
restarts. Before this change, a single failed request failed the whole
scenario, even though the next request would have succeeded. Posting a
marker twice is harmless because queries only check that a matching row
exists.

### Whole-stack Realtime probe: retry the WebSocket open
(`tests/whole-stack/websocket.ts`)

- **Change:**
  - The open timeout per attempt drops from 60s to 20s.
- A failed open (`SocketOpenError`) is retried up to 3 times. Each
attempt opens a fresh connection in its own scope, and a failed
attempt's scope is closed before the next one, so an abandoned socket
does not stay open until the scenario ends.
- **Why:** A macOS run failed with `timeout waiting for "open"`. When a
handshake is stuck behind a busy proxy, a fresh connection usually
succeeds sooner than one that keeps waiting.

### Pooler integration test: allow slower starts
(`Pooler.integration.test.ts`)

- **Change:**
  - The `pooler.start` deadline rises from 60s to 120s.
  - The test timeout rises from 120s to 240s.
- **Why:** The test starts the pooler six times, across native and
container runtimes and both pool modes, and includes a native restart.
On loaded CI runners one start can take longer than 60s, and the six
starts together can exceed the old test timeout.

### Mail integration test: longer timeout and stall diagnostics
(`Mail.integration.test.ts`)

- **Change:**
- The two-instance native test now has a 60s timeout instead of the 30s
default.
  - It records which phase it is in: start, readiness, stop or teardown.
- It keeps a bounded log tail for each instance, labelled with the
instance id.
  - It reports the phase and logs if it stalls.
- **Why:** On slow runners the test timed out with no indication of
where it stopped. The extra time covers two sequential native starts. If
it still stalls, the diagnostics show which instance and phase hung.

### Whole-stack diagnostics: keep longer log tails
(`tests/whole-stack/fixture.ts`)

- **Change:**
  - Service log tails rise from 8 KiB to 64 KiB.
  - Owner log tails rise from 4 KiB / 40 lines to 64 KiB / 400 lines.
- **Why:** One failure showed Analytics shutting down, but the truncated
tail had already dropped what led up to the shutdown. Longer tails keep
enough context to find the trigger the next time it happens.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This bumps storage from v1.97.3-r0 to v1.98.1-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.98.1-r0

Planned after supabase/slim-services published v1.98.1-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…7698)

Generate the CLI reference from command annotations instead of
maintaining separate lists of experimental and deprecated commands.
Parent help and shell completion derive the `(deprecated)` label from
the same annotation that excludes a command from the website. Deprecated
commands remain callable.

For experimental access, the command annotates the policy and its
handler explicitly checks that policy. This preserves which error users
see first: `network-bans get` validates the output format before
checking access, while `db schema declarative sync` reads project config
before checking access. CI checks that the annotation and handler
reference the same policy. The annotation does not itself enforce
access.
## TL;DR

the first Data API request after a local `db reset` now sees the reset
schema instead of a 503, a 400 or the old schema

## whats broken?

`db reset` recreates the database while PostgREST keeps running, so
PostgREST drops its connection and reconnects after a short backoff,
missing the schema reload that migrations send

reset never waited for it, so it could print `Finished supabase db
reset.` while PostgREST was still offline or serving the old schema

## now fixed by:

restarting PostgREST once migrations and seeds are done so it loads the
reset schema, reloading the API gateway to route to it, then waiting
until PostgREST reports ready through the gateway

a failure in this step only prints a warning and the reset still
succeeds. the `[experimental] stack` path is unchanged

<details><summary>visual rep</summary>

```mermaid
%%{init: {"theme": "base", "themeVariables": {"background": "transparent", "mainBkg": "transparent", "primaryColor": "transparent", "secondaryColor": "transparent", "tertiaryColor": "transparent", "actorBkg": "transparent", "noteBkgColor": "transparent", "labelBoxBkgColor": "transparent", "activationBkgColor": "transparent", "primaryBorderColor": "#8b949e", "actorBorder": "#8b949e", "actorLineColor": "#8b949e", "noteBorderColor": "#8b949e", "labelBoxBorderColor": "#8b949e", "lineColor": "#8b949e", "signalColor": "#8b949e", "textColor": "#8b949e", "actorTextColor": "#8b949e", "signalTextColor": "#8b949e", "noteTextColor": "#8b949e", "labelTextColor": "#8b949e", "loopTextColor": "#8b949e"}, "sequence": {"mirrorActors": false}}}%%
sequenceDiagram
    participant cli as CLI
    participant db as database
    participant rest as PostgREST
    participant gw as API gateway
    participant app as your app

    alt before
        Note right of cli: supabase db reset
        cli->>db: recreate
        db--xrest: connection dropped
        Note right of rest: reconnects in 1s, 2s, …
        cli->>db: migrations + seed
        db--xrest: reload notice missed while reconnecting
        Note right of cli: Finished supabase db reset.
        app->>gw: GET /rest/v1/
        gw->>rest: forward
        rest-->>gw: 503, 400 or old schema
        gw-->>app: request fails
    else now
        Note right of cli: supabase db reset
        cli->>db: recreate, migrations + seed
        Note right of rest: restarted
        rest->>db: load reset schema
        cli->>gw: reload
        loop every 1s until ready
            cli->>gw: ready?
            gw-->>cli: 200
        end
        Note right of cli: Finished supabase db reset.
        app->>gw: GET /rest/v1/
        gw->>rest: forward
        rest-->>gw: 200, reset schema
        gw-->>app: new column is there
    end
```

</details>

## ref:

- closes: #7712
yield* waitForHealthyServices(spawner, [restId], {
postgrest: {
containerId: restId,
apiExternalUrl: values.apiUrl,

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Severity: MEDIUM

Repository-controlled api.external_url becomes values.apiUrl, while this readiness probe attaches values.secretKey as apikey. Running db reset in an untrusted checkout can therefore send a configured SUPABASE_AUTH_SECRET_KEY to an attacker-controlled HTTP endpoint, even when no storage seeding occurs.
Helpful? Add 👍 / 👎

💡 Fix Suggestion

Suggestion: Don't send the secret key to a URL taken from repository-controlled config. For the local PostgREST readiness probe, build the URL from the known local gateway (the local Kong host and port the CLI started) instead of config.api.external_url. If the configured external URL must be used, check before the probe that it is a loopback or local-stack origin (localhost, 127.0.0.1, [::1], or the CLI's own Kong host). If it isn't, skip the apikey and Authorization headers or fail with an error. Put this check in checkHttpReady/kongAuthHeaders or in resolveApiExternalUrl, so every caller is covered.

mdonnalley and others added 11 commits November 19, 2026 16:06
Text help and the website reference now read literal defaults from flag
definitions without evaluating runtime defaults. Flags declare
documentation exceptions locally instead of relying on separate lookup
tables. Parsing and JSON help are unchanged.

Text help omits a fixed default for the shared `--output` flag because
commands such as `db query` choose the format at runtime. The only
change to the published reference is `logs --limit`: its documented
default is now `100`, matching the existing CLI behavior.
This bumps postgres from 17.12.0.001-r0 to 17.12.0.002-r0.


https://github.com/supabase/slim-services/releases/tag/postgres-17.12.0.002-r0

Planned after supabase/slim-services published 15.20.0.002-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This bumps postgres from 15.20.0.001-r0 to 15.20.0.002-r0.


https://github.com/supabase/slim-services/releases/tag/postgres-15.20.0.002-r0

Planned after supabase/slim-services published 15.20.0.002-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This adds postgres 17.11.0.021-orioledb-r0 as a new release line.


https://github.com/supabase/slim-services/releases/tag/postgres-17.11.0.021-orioledb-r0

Planned after supabase/slim-services published 15.20.0.002-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Summary

Reduces the Actions secrets these workflows need.

- Deletes `publish-migra.yml` and `publish-pg-prove.yml`, leaving the
`DOCKER_USERNAME` and `DOCKER_PASSWORD` secrets unused on this branch.
This is a trade-off. The workflows last pushed
`supabase/migra:3.0.1663481299` and `supabase/pg_prove:3.36` to Docker
Hub on 2024-01-02, and their run history has since expired. After this
change the repo can no longer rebuild either image. The mirror workflow
still copies the existing tags to `public.ecr.aws`, which `supabase test
db` pulls. Neither image has changed in about two years, so we accept
that. The Dependabot config comment now says those two images stay at
their current pins.
- Reads `POSTHOG_ENDPOINT` from a repository variable instead of a
secret. The build bakes the endpoint into the published CLI, so it isn't
sensitive. `POSTHOG_API_KEY` stays a secret.
- Stops passing `SENTRY_DSN` to the build. Nothing has read it since the
Go binary was removed.

## Backport

#7701 brings the same cleanup to
`v2.x`, with one difference: `SENTRY_DSN` stays a secret there. The v2
Go binary uses it to send an error report to Sentry when a user passes
`--create-ticket`.

## Linked issue

Part of CLI-3161

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude <noreply@anthropic.com>
## Summary

Split out from #7699, which stops
passing `SENTRY_DSN` to the build. Review there noted that this doc
still describes Sentry, with a `SENTRY_DSN`, as the current telemetry
backend.

- Removes `docs/telemetry.md`, an early design doc that described Sentry
and Grafana as the telemetry backends, along with collection and consent
designs the CLI never shipped. `apps/cli/docs/analytics.md` and
`apps/cli/docs/tracing-monitoring.md` document how analytics and tracing
work today.
- Points the "See Also" link in ADR 0002 at those two docs. The ADR's
decision text is unchanged.
- Drops a line in `analytics.md` that described Sentry as a current
telemetry system.

## Linked issue

Part of CLI-3161

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude <noreply@anthropic.com>
This bumps storage from v1.98.1-r0 to v1.98.2-r0.


https://github.com/supabase/slim-services/releases/tag/storage-v1.98.2-r0

Planned after supabase/slim-services published v1.98.2-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
… live test (#7710)

## Summary

The `ssl-enforcement update` live test has failed on every live e2e run
since 2026-11-18, on `develop`, `main` and `next` alike, with:

```
database reboot after ssl-enforcement update: the postmaster started at … is still running
```

#7016 made the test wait for a new postmaster after each toggle, because
an SSL enforcement change used to restart the database after
`ssl-enforcement get` had already reported it applied, and that late
restart killed the next live file's `migration up` with `57P01`. The
platform now applies the change without a restart: the same `main`
commit passed and then failed a day later, and the failing runs show no
restart afterwards either. So the wait never ends.

This drops the postmaster-start check. The test still proves the toggle
through `ssl-enforcement get`, restores the captured posture, and
reports toggle and restore failures together. `queryLiveDb`'s handling
of a dropped idle client from #7016 stays, since other live tests use
it.

If the platform goes back to restarting on this change, the original
leak would show up again as `57P01` in the next file's `migration up`.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This PR was automatically created to sync the generated `@supabase/api`
package with the latest Management API OpenAPI document.

Changes were detected in the upstream OpenAPI documents exposed by
`https://api.supabase.com/api/v1-json` and
`https://api.supabase.com/api/v2-json`.

---------

Co-authored-by: jgoux <1443499+jgoux@users.noreply.github.com>
Co-authored-by: Julien Goux <hi@jgoux.dev>
Removes the per-command `SIDE_EFFECTS.md` files and
`apps/cli/src/SIDE_EFFECTS_TEMPLATE.md`. They duplicated what the
handlers, tests and user docs already state, and they drifted out of
date.

- Rationale that existed only in those files now lives next to the code
as short comments, or in the existing guides
(`apps/cli/docs/compute-commands.md`, `stack-commands.md`,
`command-target.md`).
- `apps/cli/AGENTS.md` names tests, the error-tag fixture, the telemetry
event catalog and shipped behavior as the source of truth.
- The automated review prompts judge breaking changes against what the
stable release ships and what the tests at the merge base assert,
instead of the old `SIDE_EFFECTS.md`.
- ADRs are left unchanged as historical records.
This PR was automatically created to sync the generated `@supabase/api`
package with the latest Management API OpenAPI document.

Changes were detected in the upstream OpenAPI documents exposed by
`https://api.supabase.com/api/v1-json` and
`https://api.supabase.com/api/v2-json`.

Co-authored-by: jgoux <1443499+jgoux@users.noreply.github.com>
@mdonnalley
mdonnalley deployed to auto-approve November 19, 2026 21:31 — with GitHub Actions Active
This bumps realtime from v2.172.1-r0 to v2.174.0-r0.


https://github.com/supabase/slim-services/releases/tag/realtime-v2.174.0-r0

Planned after supabase/slim-services published v2.174.0-r0. A newer
relevant release rewrites this branch from develop and replays any
commits others pushed to it.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

This branch was successfully deployed

2 active (outdated) deployments
auto-approve — a54075e3 Deployed Nov 19, 2026 by mdonnalley via Approve the reviewed sync #12441
config-release — 0f27780a Deployed Nov 17, 2026 by github-merge-queue[bot] via Publish #61
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

do not merge Approve to apply; do not merge.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

9 participants