Am 22.11.2012 um 21:27 schrieb Pierre LINDENBAUM:
>> Am 22.11.2012 um 17:24 schrieb Pierre Lindenbaum:
>>
>>> I'm trying to run a script that fills a sqlite3 database.
>>>
>>> The script creates a lock with `lockfile` , fills the database and
>>> releases the lock.
>>>
>>> $cat test.sh
>>>
>>> #!/bin/sh
>>> lockfile /commun/data/_tmp.idp448.lock
>>> sqlite3 /commun/data/stats.sqlite "create table if not exists
>>> timeBeginDB(target text UNIQUE NOT NULL,inserted int);"
>>> rm -f /commun/data/_tmp.idp448.lock
>>> whoami
>>
>> Is "/commun/" shared between all nodes?
>
> yes.
>
> furthermore, *lockfile* is able to create the file
> /commun/data/_tmp.idp448.lock
> even when the script is called with 'qsub'
>
> the lock file is later deleted with rm -f
>
> only the call to sqlite3 prints an error message.
>
>>> without qsub the script works without any problem :
>>>
>>> [lindenb]$ sh test.sh
>>> lindenb
>>>
>>> when qsub is invoked:
>>>
>>> $ qsub -cwd -S /bin/sh test.sh
>>> Your job 673 ("test.sh") has been submitted
>>>
>>> $ cat test.sh.o673
>>> lindenb
>>>
>>>
>>> $ cat test.sh.e673
>>> Error: database is locked
>>>
>>>
>>> which is a sqlite3 error.
>>>
>>> Do you have any idea, suggestion of what, on the SGE side, could be the
>>> cause of that problem, this difference of behavior ?
>>
>> Is the lockfile you define "_tmp.idp448.lock" in any way related to the
>> database?
>
> no, it's a random name I've created. This file is here to ensure that
> only one process is accessing the sqlite3 database.
>
> I'm the only one here playing with the cluster: so nobody can create the
> file on open the sqlite database at the very same time.
>
>> Maybe the output is not telling the truth.
>>
>> On the command line it runs local, but if it's submitted on any machine in
>> the cluster. Also the environment might miss some settings which are
>> present on the command line.
>
> all the nodes have been installed using 'yum install sqlite3' so I guess
> there is nothing to configure.
Isn't it on the headnode working on a local file system (i.e. from the command
line), but on a shared one from the node by the executing job?
-- Reuti
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
