innfeed - "blocked sleeping 120" errors"

[Petr Novopashenniy]

Hi, Julien!

On Sat, 28 Mar 2009, Julien ?LIE wrote:

JL> Hi Petr,
JL> 
JL> > Mar 27 12:26:59 goblin innd: innfeed!:142:proc:86571 blocked sleeping 120
JL> > Mar 27 12:28:59 goblin innd: innfeed!:142:proc:86571 blocked sleeping 120
JL> > Mar 27 12:30:59 goblin innd: innfeed!:142:proc:86571 blocked sleeping 120
JL> > Mar 27 12:32:59 goblin innd: innfeed!:142:proc:86571 blocked sleeping 120
JL> > Mar 27 12:35:04 goblin innd: innfeed!:142:proc:86571 blocked sleeping 120
JL> > Mar 27 12:37:09 goblin innd: innfeed!:142:proc:86571 blocked sleeping 120
JL> 
JL> and
JL> 
JL> > Mar 27 12:26:31 goblin innd: innfeed-2!:25:proc:78875 blocked sleeping 120
JL> > Mar 27 12:28:33 goblin innd: innfeed-2!:25:proc:78875 blocked sleeping 120
JL> > Mar 27 12:30:33 goblin innd: innfeed-2!:25:proc:78875 blocked sleeping 120
JL> > Mar 27 12:32:33 goblin innd: innfeed-2!:25:proc:78875 blocked sleeping 120
JL> > Mar 27 12:34:33 goblin innd: innfeed-2!:25:proc:78875 blocked sleeping 120
JL> > Mar 27 12:36:36 goblin innd: innfeed-2!:25:proc:78875 blocked sleeping 120
JL> 
JL> This is related to the blockbackoff: parameter in inn.conf:
JL> 
JL>    Each time an attempted write returns EAGAIN or EWOULDBLOCK, innd(8) will
JL>    wait for an increasing number of seconds before trying it again.  This is
JL>    the multiplier for the sleep time.  If you're having trouble with channel
JL>    feeds not keeping up, it may be good to change this value to 2 or 3,
JL> since
JL>    then when the channel fills INN will try again in a couple of seconds
JL> rather
JL>    than waiting two minutes.  The default value is 120.
JL> 
JL> 
JL> Do you also have "blocked closing" logs?
No. Only this (in news.err and news.notice):

Mar 28 16:46:37 goblin innd: innfeed!:21:proc:51361 blocked sleeping 120
Mar 28 16:48:37 goblin innd: innfeed!:21:proc:51361 blocked sleeping 120
Mar 28 16:50:53 goblin innd: innfeed!:21:proc:51361 blocked sleeping 120
Mar 28 16:52:53 goblin innd: innfeed!:21:proc:51361 blocked sleeping 120
Mar 28 22:25:15 goblin innd: innfeed!:24:proc:84764 blocked sleeping 120


JL> 
JL>    badiocount:
JL>    How many read or write failures until a channel is put to sleep or
JL> closed.
JL>    The default value is 5.
JL> 
JL> 
JL> Maybe you only have *1* blocked write each time.  I do not know why.
JL> Is innfeed sending well the articles you receive?

Only after "unblocking" (120 sec after without these errors).
These innfeed processes supports ~85 peers. Yes, I have broken peers with 
"rejected 502" errors, or with broken DNS..

JL> Any backlog in <pathspool>/innfeed?
Now I can not answer this question. I try again later...

This server is mostly feeder, "no-backlog" is turned on for many peers, 
excluding my other feed servers...

JL> 
JL> 
JL> > What does this mean?
JL> 
JL> It means that innfeed does not manage to keep up with the input innd gives
JL> him.
JL> The problem with 120 seconds is that innfeed will have accumulated many
JL> articles and will probably still not keep up (as shown in your logs).
JL> Does it work better with 3 seconds?

I will try this later.

JL> 
JL> Is it happening all the day?

Mostly after crash, when server accepts backlog from my other INN 
(2.4.3) servers.

JL> -- 
JL> Julien ?LIE
JL> 

--pety