Hi
Resin license #1016608
Recently we've migrated to a ceph-based storage setup where we use NFS to allow
our multiple Resin instances to write their logs directly to Ceph over NFS.
However, on 2 of our client-machines we see NFS stalls and retransmissions a
few minutes past midnight for some files.
So far I've managed to write off posibility after posibility and right now I'm
stuck a few last possibilities, and the next one in line is the way Resin does
logrotation.
A poll of open files each second (via lsof output) shows this where out stdout
logfile gets rotated:
COMMAND PID USER FD TYPE DEVICE SIZE/OFF
NODE NAME
Sun Feb 7 23:59:58 CET 2016
java 30865 results 238w REG 0,25 1215099201
1073774597
/nfsdata-web11nfs/results/logs/web9_results_backend1/resin-web-stdout.log
Sun Feb 7 23:59:59 CET 2016
java 30865 results 238w REG 0,25 1215100254
1073774597
/nfsdata-web11nfs/results/logs/web9_results_backend1/resin-web-stdout.log
Mon Feb 8 00:00:00 CET 2016
java 30865 results 107w REG 0,25 233381888
1073774632
/nfsdata-web11nfs/results/logs/web9_results_backend1/resin-web-stdout.log.20160207
java 30865 results 118r REG 0,25 1215103872
1073774597
/nfsdata-web11nfs/results/logs/web9_results_backend1/resin-web-stdout.log
Mon Feb 8 00:00:04 CET 2016
java 30865 results 107w REG 0,25 929890304
1073774632
/nfsdata-web11nfs/results/logs/web9_results_backend1/resin-web-stdout.log.20160207
java 30865 results 118r REG 0,25 1215103872
1073774597
/nfsdata-web11nfs/results/logs/web9_results_backend1/resin-web-stdout.log
Mon Feb 8 00:00:55 CET 2016
java 30865 results 161r REG 0,25 1439432
1073774597
/nfsdata-web11nfs/results/logs/web9_results_backend1/resin-web-stdout.log
All day on the 7th the logfile is opened on filehandle 238 with write access
(238w in the FD column)
Then around midnight a new file is created named resin-web-stdout.log.20160207
on filehandle 107 with write permission (107w in the FD column).
At the same time it seems resin-web-stdout.log is opened with a new filehandle
and only with read permission (118r).
Then over time data is copied from the file with handle 118r to 107w - the
size/offset numbers increase between 00:00:00 and 00:00:04.
The stall I experience happens between 00:00:04 and 00:00:55 where lsof output
stalls (the man-page mentions that lstat(2), readlink(2), and stat(2) calls are
blocked if the NFS server is unresponsive).
Other open files in other Resin instances running on the same server does not
see these stalls - nor does other client-machines, so I am confident that our
NFS-server is alive and answering requests.
Now for the real questions:
Since we are running an ancient version of Resin (3.1.13 and 3.1.14), have the
way of rotating logfiles changed in Resin 4.0 or later?
Are the stalls I'm seeing (I assume it could just as well happen locally if a
disksubsystem was slow enough) a known issue that has been fixed in Resin 4.0
or later?
Right now the best way of solving this is to change to syslog logging, but
since we already have a roadmap for Resin 4 planned for later this year (yeay -
finally!) this might just be enough for us to move it up to the top of our list.
Regards,
Jens Dueholm Christensen
Survey IT
P +45 5161 7879
j...@ramboll.com<mailto:j...@ramboll.com>
________________________________________
Rambøll
Olof Palmes Allé 20
DK-8200 Aarhus N
Denmark
www.ramboll.dk<http://www.ramboll-management.dk/>
_______________________________________________
resin-interest mailing list
resin-interest@caucho.com
http://maillist.caucho.com/mailman/listinfo/resin-interest
