If the incoming internal groupcache request gets canceled, it should not count as a peer error.
With the current code, it's impossible to separate context cancelations (which might be outside the control of the service running groupcache) from "real" peer errors.
Also, with this PR, in case the internal request times-out (via context), the requesting peer can still do its own error accounting.
If the incoming internal groupcache request gets canceled, it should not count as a peer error.
With the current code, it's impossible to separate context cancelations (which might be outside the control of the service running groupcache) from "real" peer errors.
Also, with this PR, in case the internal request times-out (via
context), the requesting peer can still do its own error accounting.WDYT?