Greetings,
I am running GATK on a cluster. I am using '--num_threads 4' (although I don't even know if this works for indel realignment). It looks like I running into an I/O bottleneck. Has anyone ever experienced this problem?
The obvious answer is to run using less threads and launching only a couple jobs at a time. Is there another solution? I would rather batch everything at once.
error:
INFO 12:59:56,053 HttpMethodDirector - I/O exception (java.net.ConnectException) caught when processing request: Connection timed out
INFO 12:59:56,054 HttpMethodDirector - Retrying request
The IndelRealigner doesn't support -nt for multi-threading IIRC (I don't have that option in my command-line in my exome alignment pipeline) so I am guessing that is your problem. There may be others. I would try running your intended command not through a submission queue so you can see if it throws any other potential errors.
i am pretty sure it is ignoring --num_threads. I will remove it anyway. I don't want to try it outside the submission queue as this is the world I am confined to. Interactive mode won't give me a more robust error message.
GATK calls home to report to the borg collective. My jobs are finishing, but my data isn't being sent off to GATK. This is generating the errors.
I though I was allowed to accept my own answer?
Interesting, I didn't know about this "feature." Just curious, was it a firewall on your side blocking this?
For reference, you can read more here:
http://gatkforums.broadinstitute.org/discussion/1250/what-is-phone-home-and-how-does-it-affect-me
In later versions, you have to request a key to disable this (!!).
You have to request the key using an academic email address as well. It wasn't a firewall issue. On the cluster when a job is submitted it does not have access to the internet.
Log in to answer this question.
