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
conn pool: fix bugs leading to incorrect conns created (#39446)
Backport #39310 and #39349
If a connection starts draining while it has negative unused capacity
(which happens if a SETTINGS frame reduces allowed concurrency to below
the current number of requests), that connections unused capacity will
be included in total pool capacity even though it is unusable because it
is draining. This can result in not enough connections being established
for current pending requests.
This is most problematic for long-lived requests (such as streaming gRPC
requests or long-poll requests) because a connection could be in the
draining state for a long time.
Maybe fixes: #39238
Fixed an issue that could lead to too many connections when using
:ref:`AutoHttpConfig
<envoy_v3_api_msg_extensions.upstreams.http.v3.HttpProtocolOptions.AutoHttpConfig>`
if the
established connection is ``http/2`` and Envoy predicted it would have
lower concurrent capacity.
---------
Signed-off-by: Greg Greenway <[email protected]>
Signed-off-by: Dario Cillerai <[email protected]>
0 commit comments