[maker-devel] NULs in master_datastore_index.log

[Carson Holt]

Locks for not running contigs are a separate thing from the datastore
index.  They are handled by file locks in the directory for each contig.

For the datastore index, it can be subject to race conditions if two
processes happen to write an output string at the exact same time, but can
be rebuilt in less than 2 min at the end of your job using the -dsindex
flag.

The failure notices you are getting might be because of the configuration
of the NFS mount.  MAKER will check the locks it creates several times
while running a given contig to see if the lock was broken, and if it was
then that contig will fail producing an error.  This will happen before it
does any activity that might conflict with the other active run of that
contig by another process, so the process that broke the lock should
continue without issue somewhere else, but you will get a trail of
messages in the error log and maybe some ugliness in the datastore index.
You can try altering the flags set for the NFS mount, or just set the
retry count really high.

You will have to run maker -dsindex once your MPI job finishes to get the
datastore index log back in order on completion, but that only takes a
couple of minutes to rebuild.

--Carson


On 12-11-13 8:57 AM, "Mikael Brandström Durling" <mikael.durling at slu.se>
wrote:

>Hello,
>
>In order to work around locking problems with SQLite, I tried running
>(latest svn) maker off an NFSv4 export instead of NFSv3. (For some reason
>it seems maker does not detect the automounted volumes as being nfs
>mounted?). Running over NFSv4 makes SQLite happy, but then another
>problem popped up. It seems ds_utility.pm happens to write simultaneously
>to the datastore_index from several MPI ranks, which in my case results
>in stretches of NULs in the datastore file. This seems to result in maker
>trying to analyse the same contig simultaneously on two different MPI
>ranks. (I get rank failure notices on the output twice for the same rank,
>and maker complaining that the same GFF3 ID appears more than once).
>
>Any hints on how this can be handled?
>
>It appears as I should have installed and run maker on some other cluster
>without NFS mounted volumes to save myself a lot of hassles.
>
>Thanks in advance,
>Mikael
>
>
>_______________________________________________
>maker-devel mailing list
>maker-devel at box290.bluehost.com
>http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org