[HTCondor-users] Negotiation Cycles

["Kandes, Martin" <mkandes@xxxxxxxx>]

Hi all, 

I'm trying to better understand how negotiation cycles occur. In particular, I am trying to sort out how all the timing configuration options work together.

For example, the NEGOTIATOR_INTERVAL sets how often the NEGOTIATOR starts a negotiation cycle. It is defined in seconds and defaults to 60 seconds (or every 1 minute). However, the NEGOTIATOR_UPDATE_INTERVAL determines how often the NEGOTIATOR sends a ClassAd update to the COLLECTOR. This is defined in seconds and defaults to 300 seconds (or every 5 minutes). 

So my assumption here is that by default there will be 5 negotiation cycles started every 5 minute period (one every minute). But the COLLECTOR won't know the results of each cycle until 5 minutes after the cycle has begun. I assume this is because we obviously need to wait for each cycle to complete (or the COLLECTOR doesn't really need to know). 

But what if some cycles do not complete within the 5 minute window here? Yes, I could increase this value. But what if a negotiation cycle hangs for some reason? Could we not run into a situation where cycles start, but never end? I assume this is what the NEGOTIATOR_MAX_TIME_* options are for. Does anyone make sure these are not set to last one year? 

Any advice and/or experience you can offer would be much appreciated.

Thanks,

Marty Kandes

UCSD 

P.S. Also, if anyone has or knows of a nice Gantt-Like chart breaking down time in how HTCondor works, that would be awesome.