[Web10g-user] problem when using estats_list_conns()
rapier at psc.edu
Tue Mar 11 12:16:29 EDT 2014
I apologize about the delay in replying - being that you didn't
subscribe to the list this message ended up in holding. Anyway, the
first thing I would suggest is upgrading to the 2.0.7 userland and
kernel patch set. This may not resolve your problem but we'd at least be
able to eliminate that as a possible issue.
That being said, there is no reason why we should be seeing this error
after only 320 connections. I'll take a closer look at the listconns
utility to see what might be happening. As a note, I believe in 2.0.7 we
changed the naming convention and refaced web10g- to all of the userland
utilities. This is to avoid namespace collisions on systems that also
have web100 installed.
You are correct in that listconns continues to show connections that are
no longer active. This is, believe it or not, a design feature and not a
bug. If the connection ID was destroyed on socket close it also removes
access to all of the metrics for that connection. There are a number of
scenarios in which we want to get those metrics after the socket's
application exists and/or closes a socket. By maintaining that
connection ID for a set period of time (60 seconds in the version you
are using) it provides us a window in which to collect those final
metrics. In more recent versions (currently only in the git repo) this
time period is reduced significantly in order to decrease memory use.
We hope to have a new version of the userland and kernel patches
available on Web10G in the next week or so. This version will likely be
the version we submit to the Linux kernel team for consideration and
incorporates a number of advances.
On 3/7/14, 3:52 AM, Marcel Vogt wrote:
> I’m experiencing problems with web10g/web10g-userland:
> Until around 320 connections listconns works fine, but then it shows the
> following error message:
> error: Error thrown from genetlink library (The error is thrown in the
> userland API call estats_list_conns.)
> Furthermore I noticed, that listconns (estat_list_conns) shows old
> connections for a certain time, meaning that listconns shows connections
> which I don’t see with netstat –inet
> Is there a limitation on how many connections web1g can maintain?
> What could be the reason for this behavior and how can it be solved?
> My setup:
> Linux kernel: 3.8.11
> Web10g Userland: 126.96.36.199
> Libmnl library: 1.0.3-3
> Web10g-user mailing list
> Web10g-user at web10g.org
> To UNSUBSCRIBE visit https://lists.psc.edu/mailman/unsubscribe/web10g-user
More information about the Web10g-user