[Moosefs-users] mfschunkserver crash

[Laurent W. <lw...@hy...>]

Hi,

a chunkserver in our mfs volume is being retired so we can put in newer
and bigger disks. (*/path in mfshdd.cfg)
So, chunks are being replicated on others chunkservers.
One of the chunkservers is really close to being full, but chunks
continue to get written to it. Others CS are not so full (disks are
being changed box per box, so fill % is not the same on every CS).
mfschunkserver process disappeared at least 10 times without any error
message, nothing in dmesg, in /var/log/messages.
Some corner case badly handled by the algorithm ? The only thing I get
on master is chunkserver disconnected, and on chunkserver (write) write
error: ECONNRESET (Connection reset by peer).
Network is (apparently) fine, as it's running fine for months.
reserved disks for mfs on the chunkserver:
/dev/sdh1               670487    669993       494 100% /dataj
/dev/sdi1              1408053   1405512      2542 100% /datak
/dev/sdb1               632095    631663       432 100% /datab
/dev/sdc1               670487    669804       684 100% /datad
/dev/sdf1              1408053   1405889      2165 100% /datag
/dev/sdk1              1408053   1405532      2522 100% /dataf
/dev/sdl1               670487    669907       580 100% /datal
/dev/sdd1              1408053   1405454      2600 100% /datac
/dev/sde1               670487    669918       570 100% /datae
/dev/sdg1               670487    670047       440 100% /datai

Any idea ?
Thanks,
-- 
Laurent Wandrebeck
HYGEOS, Earth Observation Department / Observation de la Terre
Euratechnologies
165 Avenue de Bretagne
59000 Lille, France
tel: +33 3 20 08 24 98
http://www.hygeos.com
GPG fingerprint/Empreinte GPG: F5CA 37A4 6D03 A90C 7A1D  2A62 54E6 EF2C
D17C F64C