Adding cores counting statistics to Factory monitoring
This is a continuation of [#14559].
Jeff Dost sent a request suggesting the counters that would be more useful to Factory operators:
As an update, I re-ran tests and checked the client ads the frontend publishes to the factory collector, and as Parag said, in fact the core counts are already sent over, and look correct. Please see the ads attached, I did 3 runs again, for partitionable, fixed, and single core. UCSDSleep_gw2_part_fact_ad was a re-run of a partitionable pilot that I sent 1 4-core user job to, and 2 single core user jobs to. Based on this, I think the factory monitoring is mapping the values in the following way: GlideinMonitorGlideinsTotal -> Registered at collector GlideinMonitorGlideinsRunning -> Claimed GlideinMonitorGlideinsIdle -> Unmatched I think that instead the monitoring should show the equivalent Core versions: GlideinMonitorGlideinsTotaCores GlideinMonitorGlideinsRunningCores GlideinMonitorGlideinsIdleCores And that should be enough. If this change is made across the board, single core pilots and fixed slot pilots still will show the correct values, since in those cases the Cores and non-Cores ads happen to be identical. If this is easy enough, do you think it can make it into 3.2.20 along with the rundiff update I suggested in the previous email? Thanks, Jeff