Edge functions undocumented 4MB limit?

[MT-AI-creator]

Hi! Hitting a deploy issue I’d love some clarity on.

Project: cmrhpmvjmbhzlpfhqzgw
Function: generate-lesson (also generate-practice)
CLI: supabase 2.84.2, --no-verify-jwt
Crash report ID from previous attempt: dd8a0bfc6d614eb1b594c8466f8e3bd2

Symptom: deploy returns HTTP 500 "Function deploy failed due to an internal error" after all assets upload successfully. No 4xx, no parse error — just an opaque 500. ~80+ retries over ~6 hours, identical result.
Other functions in the same project deploy fine.

Empirical investigation:

1. The trigger correlation is with one shared module (exercise-generators.ts) crossing ~4 MB. Below ~3.85 MB → deploys work. Above ~4 MB → 500.
2. Splitting the 4 MB module into multiple files of the same total bytes does NOT help — total bundle bytes seem to be what matters, not per-file.
3. Stripping ~450 KB of comments from the module brings bundle back under threshold and deploys succeed.
4. Smaller sister functions in the same project deploy fine because their transitive import graph is smaller.
5. A minimal entry (216 bytes) that imports the same 4 MB module DOES deploy. So the limit appears to be entry-file complexity × shared graph, not just total bytes.

Questions:

1. Is there a documented (or undocumented) bundle size limit for Edge Functions? The 20 MB figure in docs doesn’t match what we observe.
2. Why a 500 instead of a clear 4xx "bundle too large"? The opaque 500 cost us ~6 hours of bisecting before we figured out the real threshold.
3. Any chance of raising the limit, or guidance on the real-world ceiling so we can plan around it? We’re a Czech edu app generating 200+ exercise types server-side and growing fast.
4. Is dynamic import (runtime-resolved path) supported on Supabase Edge Functions? If yes that would unlock per-task-type lazy loading.

Happy to share more details / repro privately. Thanks!

[kallebysantos]

Hey @MT-AI-creator 💚
Can you share the command you're using to deploy?

[MT-AI-creator]

Sure! Here's the exact command:

npx --yes supabase@2.84.2 functions deploy generate-lesson
--no-verify-jwt
--project-ref cmrhpmvjmbhzlpfhqzgw

We've also tried the same with --use-api and --use-docker — same result.
And on a few different supabase CLI versions (latest stable, 2.84.2, 2.76.15) — same.

Some additional context that might help narrow it down:

1. The deploy uploads all assets successfully (the CLI prints "Uploading assets: ..." for each transitive .ts file). The 500 only happens AFTER the upload is complete, in what I assume is the bundling/compilation step on your side. So it's not a network/upload problem — the bytes get there.

2. Other functions in the SAME project deploy fine with the same command. The difference correlates with one shared module's size (~4 MB local source). Above ~4.2 MB local source totals, the same command returns 500. Below ~4.0 MB, it returns 200. I bisected this 4 times today across different files. We stripped ~450 KB of comments from one file and the same command started succeeding immediately.

3. deno info for the failing function shows total size 12.7 MB (104 unique deps) but most of that is .d.ts type chains pulled in transitively from jsr:@supabase/functions-js/edge-runtime.d.ts (which depends on @types/node and openai types). Those are stripped from the actual upload — the CLI only sends ~4 MB of local .ts source.

4. Crash report ID I got from one of the failing 500 responses:
   dd8a0bfc6d614eb1b594c8466f8e3bd2

5. Project ref: cmrhpmvjmbhzlpfhqzgw (eu-central-1 if it matters).
   Function name where I first saw this: generate-lesson.
   I've since worked around it by moving large data pools out of TypeScript into Postgres rows, but I'd love to understand the actual limit so we don't keep hitting it.

The full minimal repro shape if it helps:
– entry index.ts ~14–25 KB
– imports one big shared module exercise-generators.ts (3.5–4.0 MB)
– that module transitively pulls in ~20 small shared files
– total local source bytes uploaded by CLI: 4.0–4.3 MB
– below 4.14 MB → deploy 200, above 4.31 MB → consistent 500

Happy to share the full repo URL or set up a minimal reproducer if you want one — just let me know what's most useful.

[MT-AI-creator]

Hey @MT-AI-creator 💚 Can you share the command you're using to deploy?
Hi team, have you please had a chance to review this behavior? I can provide more info if needed, I'm just really stuck...