Re: [HTCondor-users] Failed to get address of starter for this job in SchedLog

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

On 6/15/2015 1:43 PM, DiPalma, Michelle wrote:
> Hello,
>
> We’re seeing the following error in SchedLog in about 1 in 10 jobs:
>
> GET_JOB_CONNECT_INFO failed: Failed to get address of starter for this job
>
> When this error occurs, the slot for the job is created successfully but
> the user is never connected via ssh (no address to connect to).
>
> About our pool:
>
> All of the jobs on this particular pool are interactive. The hostnames
> not fully qualified but DEFAULT_DOMAIN_NAME is set properly. It seems to
> happen about 1 in 10 jobs all over the pool. Other interactive jobs
> submitted just before or after this error (from the same user and
> matched against the same host) work perfectly. We turned on full
> debugging but nothing useful was logged.
>
> What are we missing?
>
> Any help is much appreciated.
>
> -Michelle

Hi Michelle,

This sounds suspiciously like a bug that was just fixed a couple weeks 
ago.  Take a peek at
  https://htcondor-wiki.cs.wisc.edu/index.cgi/tktview?tn=5070

The bug fix went into the source code starting with HTCondor v8.3.6, 
which should be released to the web site by the end of this week or 
early next week.  In the meantime, setting JOB_START_DELAY=0 in 
condor_config may help reduce the occurrences of the failure.

regards,
Todd