Wrong throttleSlotInterval (retry) functionality

[xdzurman]

Ethers Version

5.6.4

Search Terms

No response

Describe the Problem

Big percentage of retries are fired right away due to

stall = throttleSlotInterval * parseInt(String(Math.random() * Math.pow(2, attempt)));

This means half of first tries is going to be 0, which does not help the 429 problem, but further worsens it. For increasing tries, there's still a high chance of hitting 0. Since API users explicitly provided throttleSlotInterval, it is incorrect that anything lower than that (0) is ever tried. Having the multiplier lower capped to 1 would avoid hitting 0 retry interval, which does not help anything, and make this more useful:

stall = throttleSlotInterval * (parseInt(String(Math.random() * Math.pow(2, attempt))) + 1);

[ricmoo]

Which service are you using? Most explicitly set a retry-after value, which auperceeds the exponential back-off algorithm.

Orherwise, the next one is supposed to wait zero seconds half the time. The rationale for exponential backoff can be found here. But if your backend isn’t using an explicit retry-after, let me know and I can possibly reach out to them.

[xdzurman]

Alchemy, but the retry-after is missing:

[xdzurman]

Curl response header output:

* We are completely uploaded and fine
* TLSv1.3 (IN), TLS handshake, Newsession Ticket (4):
* TLSv1.3 (IN), TLS handshake, Newsession Ticket (4):
* old SSL session ID is stale, removing
* Connection state changed (MAX_CONCURRENT_STREAMS == 256)!
< HTTP/2 200 
< date: Wed, 30 Nov 2022 12:56:08 GMT
< content-type: application/json
< content-length: 489
< x-alchemy-trace-id: ae6225465c4d60bde741ef04a4e9d806
< access-control-allow-origin: https://localhost:8090
< vary: Origin
< access-control-allow-credentials: true
< content-encoding: gzip
< cf-cache-status: DYNAMIC
< server: cloudflare
< cf-ray: 7723bdcefab3b34b-PRG