Supabase Edge Function timeout: three different limits, not one
A function that "times out" on Supabase can be hitting any of three separate, documented ceilings — wall-clock duration, CPU time, or memory — and the fix is different for each. Supabase's own limits page states all three; here is how to tell which one is actually firing.
Last checked November 18, 2026.
The three limits
| Limit | Value |
|---|---|
| Wall-clock duration (Free) | 150 seconds |
| Wall-clock duration (paid) | 400 seconds |
| Memory | 256 MB |
| CPU time per request | 2 seconds (actual CPU time, excluding async I/O wait) |
| Request idle timeout | 150 seconds before a 504 if no response is sent |
Sources: Supabase: Edge Function limits
Which limit is actually firing
- A function waiting on a slow upstream API or database query is spending wall-clock time, not CPU time — the 2-second CPU limit explicitly excludes async I/O wait, so a function stuck waiting (not computing) runs into the 150s/400s wall-clock or the 150s idle timeout first, not the CPU cap.
- A function doing heavy in-process computation (parsing a large payload, running a tight loop, image or data transformation in pure JS/TS) burns CPU time directly and can hit the 2-second CPU cap long before wall-clock duration becomes the issue.
- A 504 specifically means the request idle timeout fired: no response was sent within 150 seconds, regardless of plan — this is the idle timeout, separate from the longer paid wall-clock ceiling for functions that are still actively running (streaming, background work) rather than stuck idle.
- Running out of the 256MB memory limit crashes the function outright rather than producing a timeout; the symptom there is a function error, not a slow response.
Sources: Supabase: Edge Function limits
Same Deno runtime family on Scribase
Scribase Functions run on the same Deno runtime family as Supabase Edge Functions, so the same category of limit (wall-clock duration, CPU time, memory) applies in the same shape — splitting CPU-bound work from I/O-bound waiting is the same fix on either platform: move heavy computation out of the request path (a queue, a background job) rather than trying to out-wait a fixed ceiling.
Sources: Scribase pricing
Questions people ask
Does upgrading from Free to a paid plan fix every timeout?
It raises the wall-clock ceiling (150s to 400s), per Supabase's own limits page, but it does not raise the 2-second CPU time limit or the 256MB memory limit — a function hitting either of those keeps hitting them on a paid plan too.
Why does my function time out even though it finishes fast locally?
Local runs usually skip the network latency to a managed database or an external API that production traffic hits every time; a function that is fast locally but slow in production is more likely hitting the wall-clock/idle-timeout limit from real-world I/O latency than the CPU limit.
Does the 2-second CPU limit include time spent waiting on a database query?
No. Supabase's limits page specifies CPU time as actual CPU time, excluding async I/O wait — a function awaiting a query is not spending CPU time while it waits.
What gets a 504 specifically, versus some other error?
A 504 Gateway Timeout is what Supabase's docs attach to the 150-second request idle timeout: no response sent in that window. A CPU-limit or memory-limit failure surfaces as a function execution error, not a 504.