search

dmwm / WMCore

Core workflow management components for CMS.
Apache License 2.0
46 stars 107 forks source link

how can one know the amount of workflow logically in front of a given request so as to get a feeling on ETA #4906

Closed vlimant closed 6 years ago

vlimant commented 10 years ago

I use to sum-up >=assigned workflow by team in order of priority and with a rough estimate on the processing time to get an estimate of how much is in an "agent pipeline". For running* I am summing up by site. That usually gave an acceptable estimate on "how much had to be done prior to a give workflow" and a rough ETA. If ones cranks the script again and produce the summary files, you'll be able to see the content of the "overhead" column of :

https://cms-pdmv.cern.ch/queue/

Very much like :

http://spinoso.web.cern.ch/spinoso/mc/ http://www.gridpp.rl.ac.uk/cms/reprocessingcampaigns_totals.html

gives the amount of work, and could use to be central monitoring services, what is needed is a breakdown by before/after a given workflow.

ticoann commented 10 years ago

@vlimant sorry for getting back this issue so late. It sounds like that the last product you wanted is ETA. It is still not clear to me what needs to be implemented exactly. Partially due to what alogrim need to be used for ETA and what kind of information is displayed. etc. We need to discuss in detail for the clear requirement, Meanwhile, is there any temp solution which you don't need access for vocms204? I am trying to push following issue for next release (Actually it will be in Ops script) https://github.com/dmwm/WMCore/issues/4916 Could you take look if that would solve the problem temporarily if not could you add what additional information is needed?

sum-up >=assigned workflow by team in order of priority

What is sum-up here number of workflow?

vlimant commented 10 years ago

@ticoann it is not so much of an ETA that is needed, but rather a feeling of how many events are to be processed before another one. The general idea is : given a workflow, how does one check where it is in the huge pile of workflows that are in the system, and get a feeling of how far it is from getting slots through any agent, to any sites. Having this available will become increasingly necessary, to reach mandatory with the global priority pool, point at which we'll have to have such a monitoring service.

@franzoni, @boudoul, @srimanob : this is a topic you might be interested in.