Commits on Dec 6, 2022

1. Use the default retry predicate in transport …

   We have two levels of retries here.
   
   The first is in the retryTransport, which currently just tests
   IsTemporary. This has a short backoff and was intended to handle blips
   in the network.
   
   The second level was intended for higher level issues, e.g. if an upload
   PUT failed, we have to retry more than just a single HTTP request,
   because we might have to restart the upload with a new POST and PATCH.
   
   This change propagates the default retryable predicate (from the scond
   level) down to the retryTransport. This also adds a list of HTTP codes
   that are retryable, so that the retryTransport can also handle
   server-side errors that should be retried.
   
   Note that this doesn't respect remote.WithRetry{Backoff,Predicate}
   because callers can just override the entire transport via
   remote.WithTransport. This change tries to do the right thing by default
   without changing behavior for folks who seem to already know what
   they're doing.
   
   Originally, I started down the path of wrapping various remote methods
   in a retry.Retry function (similar to writes) and decided against it due
   to the size of the change and how difficult it would be to maintain.
   
   In order to reduce retry ampliciation, this adds a little context
   wrapper to signal to retryTransport not to retry (because the retry is
   happening at a higher level).
   
   This also reduces the number of steps in the default backoff for
   retryTransport because more kinds of things are being retried, and I
   want to reduce the blast radius if something is erroneously retried.

   

   jonjohnsonjr committed Dec 6, 2022
   Configuration menu

   • View commit details
   • Copy full SHA for 71df22f
   • Browse repository at this point

   Copy the full SHA

   71df22f View commit details

   Browse the repository at this point in the history