innfeed - "blocked sleeping 120" errors"
Petr Novopashenniy
pety at rusnet.ru
Tue Apr 14 13:21:37 UTC 2009
Good day, Julien!
On Sat, 11 Apr 2009, Julien ?LIE wrote:
JL> Hi Petr,
JL>
JL> > This is news.err log (INN-CURRENT):
JL> >
JL> > Mar 27 12:26:31 goblin innd: innfeed-2!:25:proc:78875 blocked sleeping 120
JL> > Mar 27 12:26:59 goblin innd: innfeed!:142:proc:86571 blocked sleeping 120
JL>
JL> Have you tried to change blockbackoff: since the time you reported the
JL> issue?
JL> Any news about it?
Yes. Now "blockbackoff: 5". No problem with this, but I do not have a
large load since Mart, not more 15-20 articles per second..
I saw this error only one time without receiving backlog from my other
server in this time... I try to monitor this issue.
I am also concerned that:
PID USERNAME THR PRI NICE SIZE RES STATE C TIME WCPU
COMMAND
865 news 1 97 0 380M 313M select 1 22.3H 7.81%
innd
74712 news 1 97 0 259M 257M select 1 216:57 4.30%
innfeed
74713 news 1 96 0 491M 489M select 1 177:02 2.54%
innfeed
74714 news 1 96 0 17796K 15760K select 0 64:43 0.34%
innfeed
###
pid 74713 started Fri Apr 10 03:32:23 2009
Updated: Tue Apr 14 16:52:46 2009
(peers: 88 active-cxns: 148 sleeping-cxns: 16 idle-cxns: 2)
###
pid 74712 started Fri Apr 10 03:32:23 2009
Updated: Tue Apr 14 16:53:28 2009
(peers: 87 active-cxns: 170 sleeping-cxns: 4 idle-cxns: 3)
###
pid 74714 started Fri Apr 10 03:32:23 2009
Updated: Tue Apr 14 16:52:18 2009
(peers: 17 active-cxns: 32 sleeping-cxns: 2 idle-cxns: 1)
###
innfeed-current uses a very large amount of memory.
For example, INN-2.4.3 (and 2.4.6 without problem), with similar
environment (~75 peers per innfeed):
PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU
COMMAND
15596 news 2 0 621M 492M RUN 20.6H 6.88% 6.88% innd
89271 news 28 0 6200K 4836K RUN 104:14 1.90% 1.90%
innfeed
89272 news 2 0 7624K 6140K RUN 102:24 1.81% 1.81%
innfeed
After restart, innfeed-current uses about the same amount of memory..
PS:
I have conducted several tests in our test lab, but I could not get a
memory leak..
I use simple test article and feed it to test server (~1500 articles
per/second).
Simple scheme:
FeedGenerator -> TestServer -> SecondTestServer
I tried parallel feeding between servers, with (and without) backlog and
drop-deferred. I tried 10 nonexistent peers (bad DNS/connection
refused/timeout). No problem. And no "blocked" errors. Maybe I tried
too short articles (not more than 1 KBytes).
JL> --
JL> Julien ?LIE
--pety
More information about the inn-workers
mailing list