The nice thing about condor_wait is it does not query the schedd - but rather just tails the user logfile.
The 8.1.3 release introduced a new API call, read_events, (see http://research.cs.wisc.edu/htcondor/manual/v8.1/6_7Python_Bindings.html) which can parse the user logfile in a manner similar to condor_wait. You could parse that file, determine the state of the jobs, and then go back to sleep.
Unfortunately, I don't know if read_events will work well with resuming in the middle of the file (or if you'll need to re-parse the whole thing).
[At one point, I had a clever C++ class that combined polling with inotify, but this never made it into HTCondor proper. Maybe I'll have some free time in a few weeks...]
So - the basic tools are there, but maybe not "put together"?
On Jan 29, 2014, at 6:39 AM, Alessandro <a.sabellico@xxxxxxxxx> wrote: