Update comment about retries.
This comment previously assumed we would have at most one pooled connection per host. That's no longer true, so update the comment.
This commit is contained in:
committed by
Martin Algesten
parent
155edeef19
commit
be9e3ca936
@@ -176,9 +176,10 @@ pub(crate) fn connect(
|
||||
// sequence of requests if all of those requests have idempotent
|
||||
// methods.
|
||||
//
|
||||
// We choose to retry only once. To do that, we rely on is_recycled,
|
||||
// the "one connection per hostname" police of the ConnectionPool,
|
||||
// and the fact that connections with errors are dropped.
|
||||
// We choose to retry only requests that used a recycled connection
|
||||
// from the ConnectionPool, since those are most likely to have
|
||||
// reached a server-side timeout. Note that this means we may do
|
||||
// up to N+1 total tries, where N is max_idle_connections_per_host.
|
||||
//
|
||||
// TODO: is_bad_status_read is too narrow since it covers only the
|
||||
// first line. It's also allowable to retry requests that hit a
|
||||
|
||||
Reference in New Issue
Block a user