Closed n-hebert closed 2 years ago
n-hebert
commented
2 years ago I'm also thinking that the jitter of 500ms is too small since the rate limit "bucket" probably refills at 20 calls per second and 500ms leaves all of the envs in a matrix fairly close together.
A 2000ms or even higher would help ungroup the calls. (e.g. https://github.com/rewindio/beanstalk-deploy/tree/hotfix/increase-jitter)
n-hebert
commented
2 years ago Some AWS blog reading on this shows that we may want to look into "Decorrelated" or "Full" jitter alternatives to the current impl. https://aws.amazon.com/blogs/architecture/exponential-backoff-and-jitter/
einaregilsson
commented
2 years ago Sorry, haven't had time to look at this, will do so when I get back to Iceland in a few weeks.
n-hebert
commented
2 years ago No worries! We've taken it the changes submitted here for a test drive and they work.
Implementing the jitter properly to the spec linked above should simply speed it up :smile:
n-hebert
commented
2 years ago Hi @n-hebert - I'm sorry to report that Einar passed away earlier this year.
My name is Gissur and I have assumed stewardship over Einar's GitHub account on behalf of his widow.
On to the PR
This looks good to me, mostly.
Just some minor cleanup, which you can see in the comments.
Hi @gzur ,
Thanks for letting me know. I'm very sorry to hear. I send my condolences to his family and friends.
Regarding the PR, I'll patch those things up shortly.
n-hebert
commented
2 years ago Replaced by #96 since I don't have write-access to Rewind's fork
Hello! :wave:
The new version works great! We do have a few environments that still fail due to heavy load, so I came up with a small change to fix that.
This will allow users to back off to arbitrary lengths, if they have heavy load environments.