6th DIRAC project Face 2 Face meeting

Europe/Paris
Salle des Séminaires (LUPM)

Salle des Séminaires

LUPM

LUPM – Bâtiment 21 Université de Montpellier – Campus Triolet Place Eugène Bataillon – CC 72 34095 Montpellier Cédex 05 FRANCE
Description
  • Réunion face à face à Montpellier (LUPM)
Inscription
Participants

---------------------------------
Update projet Andrei
+ Développements
 HTTPS Clients/Server framework 
   XML-RPC (au début) -> DISET (15 ans) -> HTTPS (aujourd'hui)
   Service migration in progress but
   -> still an issue with lack of support for file upload (Dirac StorageElement and SandboxStore)
   + Current implemenation 8.0 "Common BaseRequestHandler in the Tornado framework"
     -> complicated but overly universal and required to fork Tornado to be compatible with M2Crypto
   => might require to find a new framework to replace Tornado to enable good streaming
      *open discussion, ideas welcome*
   + HTTPS is good: single component per host, natural load balancing, container installation
     but complicated logging for it requires a central service logging (ELK stack)
 
 Acces to HTCondorCE and ARC with tokens
 CloudComputingElement replaced VMDIRAC:
   simpler but less universal, pragmatic approach
   cloud admin not agains using Application Credentials : good news!
 NB : no recent developments on HPC

 Tokens for storage : work not started, a lot of thinking needed
 
+ EGI/FG services
 running smoothly since beginning of the year : 2 M jobs
 WMS history broken !
 servers running DIRAC 8.0 (but one old REST interface for Complex VO)
   more services to be moved to HTTPS (for token pressure)
 Galaxy : discussions restarted slooowwwly

+ Projet IN2P3: 6 laboratoires, 11 ingénieurs mais 2 FTE au total en 2023
 participation du CC augmente !
 -> Vanessa devient responsable technique du projet IN2P3

+ DIRAC Community Workshop at the EGI conference 2023
 June 2023, scheduled 10 min DIRAC service overview talk (Gino)
 90 minutes training/demo sessions : goal is jupyter nb with tokens

+ CHEP 2023 : DIRAC (Federico) and CTA PS (Alice) - with proceedings
+ ISGC march 2023 : OAuth work - Andrei - proceedings being written


---------------------------------
+ Biomed Update - Sorina
 Vue DIRAC : ~5 utilisateurs
 mais utilisation via VIP ~ 100 utilisateurs actifs pour 1400 enregistrés
 => 80 années CPU
 DMS : utilisation de 3 SE de confiance
 -> difficulté de connaitre l'occupation des disques : DFC size -l
    mais ne donne pas de manière sûre l'espace disponible
 -> "drm -r" ne supprime plus les répertoires ?
 Udocker et CVMFS
   déploiement des applications
   container déployé sur cvmfs, et lancer via udocker (git clone udocker)
 VOMS -> EGI check-in
   envisager la suite de voms = EGI check-in
   mapping des sous-groupe check-in avec les groupes Dirac
   
---------------------------------
Production System for CTADIRAC - Alice - 11h25 -> 11h45
 new YAML based interface
   input/output of each transformation through MD
   set parent/child, MD verifications
 CWL : common workflow language, trough cwltool
    run the same workflow locally and in a distributed environment
 cwltool + code to translate CWL workflow into a transformation
 -> more thought needed to go from TS to PS
 other backends exist and are supported by different frameworks
   https://www.commonwl.org/implementations/
    
---------------------------------
DIRAC Deployment on Kubernetes - Natthan - 11h45 + 20'
 Ideas 
   1. create a Dockerfile with DIRAC+CTADIRAC ++
      Entrypoint : script executed when running the container to launch a specific service
   2. Helm Charts
      Describes the Docker configuration -> one template for all services
 
 Limitations
   "Agents" (or may be "executers") cannot run on Kubernetes due to the tasks queuing framework
      -> not clear if there is a significant issue
   "Logs" of all HTTPS services running are regroup under the Tornado log file
      -> possibility to separated logs?
      
  Outlook for CTA
    CTA has a dev K8s cluster at DESY
    RUCIO is already running on the K8s cluster
    CTADIRAC expected to move to K8s    

---------------------------------
DIRAC-IN2P3 - Vanessa - 12h05
 Ferme de tests montés en recyclant hardware des workers
 Instance DIRAC EGI/FG
   2 fermes : ARC-CE on Slurm pour utilisateurs + LSST
              HTCondor-CE  -> grille + ATLAS
 Configuration CONDOR
   use Tokens : needed debug but successful... but problems with "groups" not available
 Configuration ARC-CE
   possible to use Tokens and define groups... but to be tested
 Configuration of DIRAC@CC-IN2P3
   need to declare 2 CE defined, one for token and one for certificates
 -> pilots running !

---------------------------------
OIDC/OAuth2 Framework - Andrei
 AAI Components : Dirac Client/Service, Token Management, User Management, Connection to ressources
 Workflow
   user login : through "dirac-login" but transmit to multiple identity providers, through an Authorization Server
     1. choose identify provider
     2. Authenticate within Checkin
     3. Consent Screen
     4. Choose Dirac Group
     5. Return to the terminal
 Issue of Token to DIRAC group mapping : claims on user profiles used by IdP
 -> need to add group information to the token : WIP @ Check-in
 Token Manager
   provides token to users : access and refresh tokens
   provides token to other DIRAC components as needed (e.g. pilot user token)
 User Management
   moving towards complete *dynamic* user management by IdP services
     one to tone correspondance of DIRAC groups and IdP claims/scope

---------------------------------
DIRAC Logs - Bertrand
 Logs now in Tornardo, all to the same stream/file
 -> need to push logs to Fluentd/Fluent-bit, then to ElasticSearch to Kibana
 Configuration: input, output, filtres, parser
 Visualisation des logs dans Kibana, e.g. sélection par Composant, recherche de texte possible
 Grafana -> essais pas très concluant
 Andrei mentione le problème de la nécessité de conservation des logs pendant 6 mois
 Autre question de savoir comment on gère les différents niveaux de logs de manière opérationnelle
 
---------------------------------
Projet Européens - Gino
1. INFRA2023-TECH-01-01 - GreenDIFIT
  implement resource optimisation approach in orchestration frameworks
  -> 18PM CDD WP7 tests and validation
  -> 3x5PM pour CPPM,IPHC, CC-IN2P3
2. PEPR Cloud - STEEL
  adaptation dynamique des traitements des flux de données
  -> 1 CDD IR 12PM
 

Il y a un compte-rendu associé à cet événement. Les afficher.
    • 09:30 10:15
      Nouvelles du projet et tour de table 45m
      • DIRAC project update
      Orateurs: Andrei Tsaregorodtsev (Aix Marseille Univ, CNRS/IN2P3, CPPM, Marseille, France), Dr Johan Bregeon (IN2P3 LSPC), Luisa Arrabito (LUPM)
    • 10:15 10:35
      Le retour de la communauté Biomed 20m
      Orateur: Sorina POP (CNRS)
    • 10:35 10:55
      Interface de haut niveau pour le Production System, CWL, etc. 20m
      Orateur: Alice Faure (LUPM)
    • 10:55 11:15
      Conteneurs: retour d'expérience de déploiement DIRAC sur K8 20m
      Orateur: Natthan Pigoux (LUPM)
    • 11:15 11:35
      Installation de test au CC/IN2P3 20m
      Orateur: Vanessa Hamar (CC - IN2P3)
    • 11:35 12:30
      Développement autours des tokens 55m
      Orateur: Dr Andrei Tsaregorodtsev (Aix Marseille Univ, CNRS/IN2P3, CPPM, Marseille, France)
    • 12:30 13:30
      pause déjeuner 1h
    • 13:30 13:50
      Appels avec la participation DIRAC (GreenComputing, PEPR, ) 20m
      Orateur: Gino Marchetti (CC-IN2P3 / CNRS)
    • 13:50 14:10
      Le service de logging centralisé avec FluentD/FluentBit 20m
      Orateur: M. Bertrand Rigaud (CC-IN2P3)
    • 14:10 14:30
      AOB 20m
      • Participation à EGI2023, Poznan
      • Workshop Rucio/DIRAC
      Orateur: Dr Andrei Tsaregorodtsev (Aix Marseille Univ, CNRS/IN2P3, CPPM, Marseille, France)