broken cache problem for arla

[Takashi Ichihara]

>Takashi Ichihara <ichihara at rarfaxp.riken.go.jp> writes:
>
>>   Following problems occurs frequently around our environment: When 
>> a linux system "crash" with panic, some parts of cache files of the 
>> arla seems to be destroyed.  After this crash, arla does not work 
>> correctly even rebooting the system. Access to the specific files on 
>> /afs tree, which are related to the broken cache files, has problem.  
>
>
>Starting arla with the -z flag should solve the problem, as of 0.21
>the linux startarla does that by default. That is a better way to fix
>the problem.
>
>You have the -z flag in your startarla ?
>
>Love

  Yes.  We are using arla 0.21 and I have confirmed that arlad is 
started with -z flags in startarla in arla 0.21. But we have this 
problem.  We have small 8 Linux CPU farms (Kernel 2.2.0, arla 0.21) 
and found that arla of 3 nodes are out of order due to this broken 
cache problem. This problem was fixed by manually cleaning the 
cache directory.

    To confirm -z flags works correctly or not, I have rebooted the 
system and check the arla cache directory. The system was booted at
20:23. (arlad was started with -z flag just after 20:23) In this 
condition there are many files whcih were created before arlad 
(system) was started.

[system booting started at 20:03:15]

  taken from /var/log/message
  Feb 19 20:23:15 p6spin kernel: Linux version 2.2.0 (root at xx.riken.go.jp)
 (gcc version 2.7.2.3) #1 Wed Jan 27 10:36:09 JST 1999

[arlad is started with -z flags]
  # ps ax | grep ar
    330  ?  S    0:00 /usr/arla/bin/arlad -z

[There are some old cache files in arla cache directory]

  # ls -l /usr/arla/cache

  -rw-------   1 root     root         2048 Feb 19 20:23 0001
  -rw-------   1 root     root           11 Feb 19 20:23 0002
  -rw-------   1 root     root         2048 Feb 19 20:23 0003
  -rw-------   1 root     root           15 Feb 19 20:23 0004
  -rw-------   1 root     root         2048 Feb 19 20:23 0005
     :
  -rw-------   1 root     root            0 Feb 19 20:24 0F9F
  -rw-------   1 root     root            0 Feb 19 20:24 0FA0
  -rw-------   1 root     root         1024 Feb 19 19:49 @001    
  -rw-------   1 root     root         1024 Feb 19 19:49 @003
  -rw-------   1 root     root         1024 Feb 19 19:49 @00D
  -rw-------   1 root     root         1024 Feb 19 19:49 @010
  -rw-------   1 root     root         1024 Feb 19 19:49 @027
  -rw-------   1 root     root        15308 Feb 19 19:50 fcache
  -rw-------   1 root     root            0 Feb 19 16:54 log
  -rw-------   1 root     root         4256 Feb 19 19:50 volcache


   Therefore starting arlad with -z flag (as default option in arla 
0.21 startarla) does not clean the cache files completely, at leaset 
in our environment.
					Takashi Ichihara