You are viewing a plain text version of this content. The canonical link for it is here.
Posted to httpclient-users@hc.apache.org by Jo...@wellsfargo.com.INVALID on 2022/11/29 16:22:10 UTC
Increased new connection rate
All,
I'm using 4.5.12.
I'm trying to reproduce a problem where it appears we're creating a lot more new backend connections than we expect even though incoming request volume stays about the same.
I have max per route set to 5 and have confirmed with enabling debug on PoolingHttpClientConnectionManager that the max is in fact 5. There is only one route.
The TTL is set to 2 minutes and that also seems to work as advertised. AFAIK the backend does not close connections by sending "connection: close" or any other means, though that's not 100% confirmed.
Taking a guess that connect or read timeouts could cause a connection to get closed (and a new one opened,) I added artificial latency to the backend responses to trigger timeouts. That did cause the connection creation rate to rise above the steady state. However it also appears to have caused more than 5 connections to be in an "established" state at once according to netstat. It wasn't unusual to see 2-3x as many established connections. Because of the primitive way I'm counting open connections, the actual connection creation rate could be even higher.
I'm looking for suggestions on what could cause:
* Increased connection creation rate
* Established connection counts higher than the pool size
Thanks
Re: Increased new connection rate
Posted by Gary Gregory <ga...@gmail.com>.
You would also want to use latest version of the 4.x line of artifacts I
would think.
Gary
On Tue, Nov 29, 2022, 14:59 Oleg Kalnichevski <ol...@apache.org> wrote:
> On Tue, 2022-11-29 at 16:22 +0000, John.E.Gregg@wellsfargo.com.INVALID
> wrote:
> > All,
> >
> > I'm using 4.5.12.
> >
> > I'm trying to reproduce a problem where it appears we're creating a
> > lot more new backend connections than we expect even though incoming
> > request volume stays about the same.
> >
> > I have max per route set to 5 and have confirmed with enabling debug
> > on PoolingHttpClientConnectionManager that the max is in fact 5.
> > There is only one route.
> >
> > The TTL is set to 2 minutes and that also seems to work as
> > advertised. AFAIK the backend does not close connections by sending
> > "connection: close" or any other means, though that's not 100%
> > confirmed.
> >
> > Taking a guess that connect or read timeouts could cause a connection
> > to get closed (and a new one opened,) I added artificial latency to
> > the backend responses to trigger timeouts. That did cause the
> > connection creation rate to rise above the steady state. However it
> > also appears to have caused more than 5 connections to be in an
> > "established" state at once according to netstat. It wasn't unusual
> > to see 2-3x as many established connections. Because of the
> > primitive way I'm counting open connections, the actual connection
> > creation rate could be even higher.
> >
> > I'm looking for suggestions on what could cause:
> >
> >
> > * Increased connection creation rate
> > * Established connection counts higher than the pool size
> >
> > Thanks
> >
>
> I do not think there is not much we can unless you can provide us with
> a test application that can be used to reproduce the issue
> independently from your specific application context.
>
> Oleg
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: httpclient-users-unsubscribe@hc.apache.org
> For additional commands, e-mail: httpclient-users-help@hc.apache.org
>
>
Re: Increased new connection rate
Posted by Oleg Kalnichevski <ol...@apache.org>.
On Tue, 2022-11-29 at 16:22 +0000, John.E.Gregg@wellsfargo.com.INVALID
wrote:
> All,
>
> I'm using 4.5.12.
>
> I'm trying to reproduce a problem where it appears we're creating a
> lot more new backend connections than we expect even though incoming
> request volume stays about the same.
>
> I have max per route set to 5 and have confirmed with enabling debug
> on PoolingHttpClientConnectionManager that the max is in fact 5.
> There is only one route.
>
> The TTL is set to 2 minutes and that also seems to work as
> advertised. AFAIK the backend does not close connections by sending
> "connection: close" or any other means, though that's not 100%
> confirmed.
>
> Taking a guess that connect or read timeouts could cause a connection
> to get closed (and a new one opened,) I added artificial latency to
> the backend responses to trigger timeouts. That did cause the
> connection creation rate to rise above the steady state. However it
> also appears to have caused more than 5 connections to be in an
> "established" state at once according to netstat. It wasn't unusual
> to see 2-3x as many established connections. Because of the
> primitive way I'm counting open connections, the actual connection
> creation rate could be even higher.
>
> I'm looking for suggestions on what could cause:
>
>
> * Increased connection creation rate
> * Established connection counts higher than the pool size
>
> Thanks
>
I do not think there is not much we can unless you can provide us with
a test application that can be used to reproduce the issue
independently from your specific application context.
Oleg
---------------------------------------------------------------------
To unsubscribe, e-mail: httpclient-users-unsubscribe@hc.apache.org
For additional commands, e-mail: httpclient-users-help@hc.apache.org