You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Starting on December 20th, we started getting RateLimiting and 403 (Forbidden) exceptions from GitHub.
The running theory is that the 403s are also because of Rate Limiting, GitHub API documentation says it is a possibility.
The main fact supporting this theory is that looking at a single replica at a point in time, we can clearly see that in some cases the same API call fails with a 403 and sometimes it succeeds, even tho they're made with the same cached token.
Goals
Stop these exceptions from happening. A good start might be to add some logging so we can figure out how many calls we actually make per hour. Currently the only related metric is the number of token cache hits we have per hour (around 4.3k), but that number gives a lower limit, there's a number of places where we reuse the same HttpClient without going to the token cache again.
Also we might have to start respecting the backoff timer we get sent in these Rate Limiting responses
The text was updated successfully, but these errors were encountered:
<!-- Link the GitHub or AzDO issue this pull request is associated with.
Please copy and paste the full URL rather than using the
dotnet/arcade-services# syntax -->
#4410
Context
Starting on December 20th, we started getting RateLimiting and 403 (Forbidden) exceptions from GitHub.
The running theory is that the 403s are also because of Rate Limiting, GitHub API documentation says it is a possibility.
The main fact supporting this theory is that looking at a single replica at a point in time, we can clearly see that in some cases the same API call fails with a
403
and sometimes it succeeds, even tho they're made with the same cached token.Goals
Stop these exceptions from happening. A good start might be to add some logging so we can figure out how many calls we actually make per hour. Currently the only related metric is the number of token cache hits we have per hour (around 4.3k), but that number gives a lower limit, there's a number of places where we reuse the same HttpClient without going to the token cache again.
Also we might have to start respecting the backoff timer we get sent in these Rate Limiting responses
The text was updated successfully, but these errors were encountered: