Triying to undestand Throttling Retry Attempts (gRPC retry Design Proposal)

[eem...@gmail.com]

Hi,

Im trying to undestand the throttling retry attempts, as defined on the gRPC retry Design Proposal, here:

https://github.com/grpc/proposal/blob/master/A6-client-retries.md#throttling-retry-attempts-and-hedged-rpcs

The proposal saids:

"If token_count is less than or equal to the threshold, defined to be (maxTokens / 2), then RPCs will not be retried until token_count rises over the threshold."

How token_count will rises over the threshold if "RPCs will not be retried" ?

Is it rising due to other diferent requests, which are not retries, made to the same server?

Thanks!

[Eric Anderson]

On Fri, May 3, 2019 at 9:39 AM <eem...@gmail.com> wrote:

How token_count will rises over the threshold if "RPCs will not be retried" ?

token_count is modified for the initial attempt as well, not only after a failed retry. So as other RPCs succeed on their first attempt tokens will be refilled.

Is it rising due to other diferent requests, which are not retries, made to the same server?

Yes. This is described in the bullets immediately preceding the "If token_count is less than or equal to the threshold" sentence you quoted.