[ensembl-dev] Variant effect predictor and local database

Will McLaren wm2 at ebi.ac.uk
Fri Jun 17 11:22:21 BST 2011

Hi Henrikki,

As I mentioned to you in an earlier thread, there will be some
improvements coming in the next release of the script. They will
improve the number of queries being made, as well as give you the
option to run the script without using the database.

I have tried using the script in whole-genome mode in the parallel
against a local server as you have done, but I don't see any of these
errors. I suspect there are some parameters on the MySQL server you
can change which should allow a larger number of connections. How many
parallel processes are you running? I have tried running one per
chromosome in human (~25 parallel processes) and I didn't get any
connection problems.



On 17 June 2011 10:46, Henrikki Almusa <henrikki.almusa at helsinki.fi> wrote:
> Hi,
> I was trying to put the variant effect predictor to run in cluster and
> parallelise each chromsome to their own job. I got it running but soon
> encountered a problem with database. I got connection error to database on
> some of the runs. If I run without whole_genome I got some results before
> the execution failed. The log of the error can be seen in attachment.
> After a quick peek on the database server, it was noticed that the amount of
> inactive connections went up dramatically. At around 32k inactive
> connections, the errors started and executions failed. One execution alone
> generated thousands of inactive connections, but managed to finish before
> they grew too far.
> Why does this generate that huge number of connections? Is there a way to
> not do that as this means that I cannot use this tool in automatic fashion
> (as more than one or two cannot be running at the same time).
> Thanks,
> --
> Henrikki Almusa
> _______________________________________________
> Dev mailing list    Dev at ensembl.org
> List admin (including subscribe/unsubscribe):
> http://lists.ensembl.org/mailman/listinfo/dev
> Ensembl Blog: http://www.ensembl.info/

More information about the Dev mailing list