fifo log problem

Mike Tancsa mike at sentex.net
Mon Nov 3 17:43:49 PST 2008


At 11:48 AM 11/3/2008, Mike Tancsa wrote:
>At 11:34 AM 11/3/2008, Poul-Henning Kamp wrote:
>>In message <200811031547.mA3FlVVs024666 at lava.sentex.ca>, Mike Tancsa writes:
>>
>> >Seems to work fine with cat
>>
>>Ok, and the loss is not from one end, it is random records in
>>the middle ?
>
>
>Yes, they seem to initially get written and then tail off for some 
>reason. I am not sure why.  Actually, if I SIGHUP syslogd, it seems 
>to make a difference, in that I can generally see when newsyslog sig 
>HUPs syslog to do log rotation.  Perhaps this is confusing things ?

I tried changing the config so that there is only the fifo log being 
written to and disabled newsyslog so that syslogd is not getting a 
HUP signal.  The strange thing is that reading from it gives 
different results?!?

Sometimes doing
[ps0278]# fifolog_reader all.fifo | wc
>From    0 Wed Dec 31 19:00:00 1969
To      1225760679 Mon Nov  3 20:04:39 2008
Read from 1d800
       59     413    3068
0[ps0278]#

and a exactly for 1min it will show the correct results

0[ps0278]# fifolog_reader all.fifo | wc
>From    0 Wed Dec 31 19:00:00 1969
To      1225760538 Mon Nov  3 20:02:18 2008
Read from 0
    10765   75995  556816
0[ps0278]#


and then go back to showing just a subset for 4 min.   I am guessing 
this coincides with when the flush runs


This is a nanobsd image, so /var on /dev/md1 and RELENG_7 from a few days ago

I have been running

#!/bin/sh

i=0
while true
do
         i=`expr $i + 1`
         logger $i
         echo $i
         sleep 1
done

and they seem to be there when it shows all the results, but for the 
most part it just shows a subset

         ---Mike 



More information about the freebsd-stable mailing list