ADC Analytics Fortnightly

Wednesday 2 Nov 2016, 16:00 → 18:00 Europe/Zurich

3196/R-023 (CERN)

Ilija Vukotic (University of Chicago (US)), Mario Lassnig (CERN)

TWiki: https://twiki.cern.ch/twiki/bin/view/AtlasComputing/ATLASAnalytics

Tickets: https://its.cern.ch/jira/secure/RapidBoard.jspa?rapidView=4733

16:00 → 16:05 News

Speakers: Ilija Vukotic (University of Chicago (US)), Mario Lassnig (CERN)

Future ATLAS monitoring.pdf

• ITES backend name has been changed from "public" to dedicated "ATLAS"
  • no problems observed
• Two new user accounts requested on ITES for pilot2 + pilot2wrapper
  • atlas_pilot, atlas_pilotfactory
  • Discussion later for integration workflow(?)
• Pandalogger is last app on aianalytics01 ES, Shaojun now has correct URL/port/etc..
  • decommissioning imminent
• rucio-events being written to both UChicagoES and ITES
  • shutting off writing to UChicago on Monday
  • need to copy historical data over from UChicagoES
• Last two apps that are exclusive on UChicagoES
  • panda_jobsarchive
    • import loss only observed on this index when imported into ITES
  • rucio-logs
    • requires change in logstash hostname
    • can we write to both backends for transition period?
    • move historical data
• Due to Supercomputing conference, potentially shift meeting from Nov 16 to Nov 23 (to be confirmed with Vakho/AES)

 

Notes (News)

• rucio-events will _not_ be decommissioned, but only transfer-* should be sent to UchicagoES
• panda_jobsarchive goes to UMA only
• move from Nov16 to Nov23 confirmed by Vakho

16:05 → 18:00 Dedicated monitoring discussion

Speakers: Alastair Dewhurst (Lancaster University), Alberto Aimar (CERN), Alessandro Di Girolamo (CERN), Andrea Sciaba (CERN), Andrej Filipcic (Jozef Stefan Institute (SI)), Danilo Piparo (CERN), Edward Karavakis (CERN), Ilija Vukotic (University of Chicago (US)), Dr Jaroslav Guenther (Innsbruck University), Joaquin Ignacio Bogado Garcia (Universidad Nacional de La Plata (AR)), Luca Magnoni (CERN), Maksim Gubin (Russian Academy of Sciences (RU)), Mario Lassnig (CERN), Martin Barisits (CERN), Peter Love (Lancaster University (GB)), Sergey Belov (Joint Inst. for Nuclear Research (RU)), Tatiana Korchuganova (Russian Academy of Sciences (RU)), Thomas Beermann (CERN)

Monit_ADC_201611.pdf

Monit_ADC_201611.pptx

• Notes
  • changing history: no; but possibility to fix mistakes foreseen: spark jobs
  • BigPanda: move additional features from BigPanda into Realtime Job View?
  • Link Grafana plots with Kibana? Automatically fill selection-attributes
  • Batch usecases that can impact Interactive ES
• Discussion points
  • ...
  • Matrix: Notebooks with ES as data backend instead of Heatmap?
  • Development for upstream Kibana
  • Inflation of dashboards - how to manage?
  • Workflow to replay historical data from HDFS to ElasticSearch
    • use same kind of "query" to go against both HDFS or ElasticSearch? Spark jobs?
  • Alarming customization (thresholds, detect standard/anomalous behaviour?)
  • Any data still being put double on es-atlas and itmon (e.g., rucio-events? panda_jobsarchive?)
  • Job view vs BigPanda?
  • "conference-quality" plots?
  • IT-DB machine learning Spark jobs - Kafka plugin?
  • Zeppelin vs SWAN ?
  • DDM Accounting - summaries are ready
  • Pilot2 monitoring workflow integration
  • Priorities/deadlines/people