stack: every Edge Function request starts a new worker, so the same function never reuses one and each call pays a cold start

Open
  • Is something wrong?
#7,274 0 comments 0 reactions 1 assignee View on GitHub

Maintainers usually reply within 1 day

@7ttp is already working on this.

Since Oct 27, 2026.

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
58/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
deno, typescript
Domain
backend, cli

Research direction

Start in packages/stack/src/functions/serve.main.ts near line 466, where the issue reports forceCreate: true, and compare it with apps/cli/src/shared/functions/serve.main.ts near line 516. Reproduce with the supplied hello function and repeated curl calls on the experimental stack. Done when subsequent requests reuse the worker and its request count increases, matching the legacy stack.

Written by the indexing model from the issue text.

Description

🐛 Bug supabase/cli
Affected area

Edge Functions

Supabase CLI version

2.121.0

Operating system

macOS 27.0.1 (Apple silicon)

Installation method

pnpm

Command
SUPABASE_EXPERIMENTAL_STACK=1 supabase start --runtime native
for i in 1 2 3 4 5; do curl -s -w '  %{time_total}s\n' http://127.0.0.1:54321/functions/v1/hello; done
Actual output
{"requests":1,"workerStartedAt":"2026-10-22T04:47:49.507Z","xForwardedFor":null,"xRealIp":null}  1.111719s
{"requests":1,"workerStartedAt":"2026-10-22T04:47:49.553Z","xForwardedFor":null,"xRealIp":null}  0.023354s
{"requests":1,"workerStartedAt":"2026-10-22T04:47:49.586Z","xForwardedFor":null,"xRealIp":null}  0.018719s
{"requests":1,"workerStartedAt":"2026-10-22T04:47:49.613Z","xForwardedFor":null,"xRealIp":null}  0.017223s
{"requests":1,"workerStartedAt":"2026-10-22T04:47:49.645Z","xForwardedFor":null,"xRealIp":null}  0.021565s

# same project on the legacy stack (SUPABASE_EXPERIMENTAL_STACK=0 supabase start)
{"requests":1,"workerStartedAt":"2026-10-22T04:48:40.137Z","xForwardedFor":"192.168.171.1","xRealIp":"192.168.171.1"}  0.056585s
{"requests":2,"workerStartedAt":"2026-10-22T04:48:40.137Z","xForwardedFor":"192.168.171.1","xRealIp":"192.168.171.1"}  0.004061s
{"requests":3,"workerStartedAt":"2026-10-22T04:48:40.137Z","xForwardedFor":"192.168.171.1","xRealIp":"192.168.171.1"}  0.003451s
{"requests":4,"workerStartedAt":"2026-10-22T04:48:40.137Z","xForwardedFor":"192.168.171.1","xRealIp":"192.168.171.1"}  0.003629s
{"requests":5,"workerStartedAt":"2026-10-22T04:48:40.137Z","xForwardedFor":"192.168.171.1","xRealIp":"192.168.171.1"}  0.003386s

On the experimental stack requests stays at 1 and workerStartedAt changes on every call, so each request gets a brand new worker. The legacy stack keeps one worker and counts up.

Expected behavior

Requests to the same function reuse its running worker, as they do on the legacy stack. Only the first call should pay for booting the worker.

Steps to reproduce
  1. mkdir repro && cd repro && supabase init
  2. Create supabase/functions/hello/index.ts: 
    let requests = 0;
const workerStartedAt = new Date().toISOString();

Deno.serve((req) =>
  Response.json({
    requests: ++requests,
    workerStartedAt,
    xForwardedFor: req.headers.get("x-forwarded-for"),
    xRealIp: req.headers.get("x-real-ip"),
  })
);
  3. Add to supabase/config.toml: 
    [functions.hello]
verify_jwt = false
  4. SUPABASE_EXPERIMENTAL_STACK=1 supabase start --runtime native, then call the function a few times as above. requests is always 1.
  5. Stop it, start the legacy stack with SUPABASE_EXPERIMENTAL_STACK=0 supabase start (I passed -x for everything except the database, Kong and the edge runtime) and make the same calls. requests counts up.

(The two header fields are for #7275 and don't matter here.)

Crash report ID

No response

Docker and service versions
Docker: Client 29.8.2, Server 29.8.2 (linux/arm64)

supabase/postgres       17.11.0.004
supabase/gotrue         v2.197.0
postgrest/postgrest     v16.4
supabase/realtime       v2.143.4
supabase/storage-api    v1.82.1
supabase/edge-runtime   v1.77.4
supabase/studio         2026.10.12-sha-3699278
supabase/postgres-meta  v0.100.0
supabase/logflare       1.53.0
supabase/supavisor      2.9.13
Additional context

In the v2.121.0 tag (97531eb590ccd2c5cf215287ae03e877180810b4), the experimental stack's functions main passes forceCreate: true to EdgeRuntime.userWorkers.create (packages/stack/src/functions/serve.main.ts:466). The legacy main sets const forceCreate = false (apps/cli/src/shared/functions/serve.main.ts:516). develop and next still have forceCreate: true on the same line.

On a hello-world function the cost is small, about 20 ms per call instead of 4 ms. On our real project the functions import a lot more, and every call to the same function took 0.55 to 0.65 s on the native runtime, against 4 ms warm on the legacy stack. That makes the native runtime too slow for our local development and E2E runs, which is the main thing keeping us on Docker.

If forceCreate: true is there so code edits show up on the next request, reusing the worker until a file under the function changes would keep that without a cold start on every call.

I investigated this with help from an AI coding assistant (Claude Code), which also drafted this text. The results above come from real runs.

Dominant language
TypeScript
Stars
2.4k
Forks
540
Avg merge
17h 3m
Merged PRs (30d)
390

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from supabase/cli

All issues in supabase/cli

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.