HEPiX Batch Monitoring Working Group

Thursday Jan 24, 2019, 9:00 AM → 9:45 AM America/New_York

Fifemon used at

• BNL: inject custom classads in hierarchy

• CERN: working on packaging in RPM (will share); added metrics to monitor cluster utilization per experiment (and react on under-utilization w.r.t. quota)

  • Mix: fifemon + collectd based layer

• FNAL: Added multiple schedds support; quota collection (from accounting classads in case one cannot contact negotiator); config options to process properly; need to push upstream

  • Classads with parameters with number in the name, causing issues (WillSK)

• Deployment with Docker? Dependency hell either way.

• Python 3 migration? Noone yet.

• Elasticsearch / Filebeat configuration for fifemon repo as well

• Nicolas (CC-IN2P3): mod to mon jobs submitted to local schedd (as opposed to CE) in jobs.py

  • (BNL) We also had to remove some FNAL-specific assumptions in the code

• Include cgroups mon into fifemon?

  • (BNL) Separate tool for us, direct monitoring on node (https://github.com/HEPiX-batchmonitoring/condor_graphite)

• Mixed solutions for monitoring with some overlap between what each tool can do

• https://indico.cern.ch/event/778660/contributions/3245464/attachments/1770416/2876523/CERNHTCondorMonitoring.pdf

• Common repository?

9:00 AM → 9:10 AM HTCondor Monitoring (Fifemon) Overview

I'll start discussion with an overview of how we at BNL use the Fifemon (https://fifemon.github.io/) monitoring tool including changes we made to it

Speaker: William Edward Strecker-Kellogg (Brookhaven National Laboratory (US))

Fifemon Monitoring

9:20 AM → 9:45 AM Discussion on Fifemon

Other case studies are welcome, as well as feedback and input from the author (Kevin, who should be in attendance).

CERN HTCondor Monitoring