OK, I found it. It looks like the default was changed from 2000ms to null (disabled) in the move to 5.x. This is a promising line of investigation for me to pursue; I'll report back with what I find.
On Fri, Mar 12, 2021 at 12:34 AM Ryan Schmitt <[email protected]> wrote: > I found changes related to HTTPCLIENT-2094, but it seems to me like the > default value for `validateAfterInactivity` is `null` and has been for > years. Is there something in the classic client (not the fluent client) on > 4.5.13 that enabled connection validation by default? > > On Fri, Mar 12, 2021 at 12:18 AM Ryan Schmitt <[email protected]> wrote: > >> What you're saying about the idle connection validation interval is >> interesting. Is this the `validateAfterInactivity` setting defined in >> `PoolingHttpClientConnectionManager`? Are the exceptions caused by the >> client trying to reuse a connection that's been closed, or half-closed? I >> had a hunch that something like this could be happening, but I couldn't see >> how it was possible, particularly for the classic client. (I've only >> actually seen this sort of issue with async clients.) >> >> On Thu, Mar 11, 2021 at 6:58 PM Carter Kozak <[email protected]> wrote: >> >>> Is there any way you could reproduce the latency change and failures in >>> a synthetic environment? If you could share a reproducer it would be >>> incredibly helpful. >>> >>> I recall some changes to the idle connection validation interval default >>> value, if the value isn’t set or the feature is disabled, connections >>> closed by remote servers will result in NoHttpResponseExceptions. Bear in >>> mind the check it triggers can cost a full millisecond on every request. >>> >>> -Carter >>> >>> > On Mar 11, 2021, at 9:45 PM, Gary Gregory <[email protected]> >>> wrote: >>> > >>> > Also note that 5.1 just hit Maven Central. >>> > >>> > Gary >>> > >>> >> On Thu, Mar 11, 2021 at 9:39 PM Ryan Schmitt <[email protected]> >>> wrote: >>> >> >>> >> We override (and test) most or all of the timeouts. The calls in >>> question >>> >> don't seem to be taking long enough to be getting timed out. >>> >> >>> >>> On Thu, Mar 11, 2021 at 6:09 PM Gary Gregory <[email protected]> >>> wrote: >>> >>> >>> >>> Maybe a change in default timeouts from infinite to 1 or 2 minutes? >>> >>> >>> >>> Gary >>> >>> >>> >>> On Thu, Mar 11, 2021 at 9:00 PM Ryan Schmitt <[email protected]> >>> wrote: >>> >>>> >>> >>>> On Saturday we rolled out a company-wide upgrade from Apache client >>> >>> 4.5.13 >>> >>>> to 5.0.3, and yesterday we ended up rolling it back due to several >>> >>> services >>> >>>> reporting significant increases in client-side latency and request >>> >>> failures >>> >>>> due to NoHttpResponseException. Can someone suggest a good place to >>> start >>> >>>> looking for the root cause? >>> >>> >>> >>> --------------------------------------------------------------------- >>> >>> To unsubscribe, e-mail: [email protected] >>> >>> For additional commands, e-mail: [email protected] >>> >>> >>> >>> >>> > >>> > --------------------------------------------------------------------- >>> > To unsubscribe, e-mail: [email protected] >>> > For additional commands, e-mail: [email protected] >>> > >>> >>> >>> --------------------------------------------------------------------- >>> To unsubscribe, e-mail: [email protected] >>> For additional commands, e-mail: [email protected] >>> >>>
