Agenda-15-05-2017

From EGIWiki
Jump to: navigation, search
Main EGI.eu operations services Support Documentation Tools Activities Performance Technology Catch-all Services Resource Allocation Security


Documentation menu: Home Manuals Procedures Training Other Contact For: VO managers Administrators


General information

Middleware

CMD

  • still working on CMD-OS updates
  • CMD-ONE first major to be released for OpenNebula 5
    • CESGA will update to OpenNebula 5 and test in particular the new cloudkeeper (former vmcatcher)

UMD

  • UMD 4.5 (June)
    • WN/UI for C7
    • CREAM for C7

Preview repository

released on:

  • 2017-04-26
    • Preview 1.11.0 AppDB info (sl6): ARC 15.03 update 12, CernVM-FS 2.3.5, gfal2 2.13.3, gfal2-utils 1.5.0, srm-ifce 1.24.2, XRootD 4.6.0
    • Preview 2.11.0 AppDB info (CentOS 7): ARC 15.03 update 12, CernVM-FS 2.3.5, dCache 3.0.12, emi-UI 4.0.2, gfal2 2.13.3, gfal2-utils 1.5.0, glite-ce-cream-client 1.15, glite-yaim-clients 5.2.1-1, srm-ifce 1.24.2, XRootD 4.6.0

Operations

Testing FedCloud sites

Resource Centre
STATUS
IFCA-LCG2 OK
IN2P3-IRES OK
100IT OK
RECAS-BARI OK
CESNET-MetaCloud OK
FZJ OK
BEgrid-BELNET OK
INFN-CATANIA-STACK OK
TR-FC1-ULAKBIM OK
INFN-PADOVA-STACK OK
IISAS-FedCloud OK
UPV-GRyCAP OK
IISAS-Nebula

OK (but not supporting fedcloud VO)

CLOUDIFIN <style type="text/css"><!--td {border: 1px solid #ccc;}br {mso-data-placement:same-cell;}--></style>https://ggus.eu/index.php?mode=ticket_info&ticket_id=128104
SCAI <style type="text/css"><!--td {border: 1px solid #ccc;}br {mso-data-placement:same-cell;}--></style>https://ggus.eu/?mode=ticket_info&ticket_id=127821
CYFRONET-CLOUD <style type="text/css"><!--td {border: 1px solid #ccc;}br {mso-data-placement:same-cell;}--></style>https://ggus.eu/index.php?mode=ticket_info&ticket_id=128100
CESGA <style type="text/css"><!--td {border: 1px solid #ccc;}br {mso-data-placement:same-cell;}--></style>https://ggus.eu/?mode=ticket_info&ticket_id=127815
BIFI <style type="text/css"><!--td {border: 1px solid #ccc;}br {mso-data-placement:same-cell;}--></style>https://ggus.eu/index.php?mode=ticket_info&ticket_id=128096
HG-09-Okeanos-Cloud replacing VMCaster with Cloudkeeper
CETA-GRID https://ggus.eu/?mode=ticket_info&ticket_id=124224
GoeGrid <style type="text/css"><!--td {border: 1px solid #ccc;}br {mso-data-placement:same-cell;}--></style>https://ggus.eu/?mode=ticket_info&ticket_id=128101
NCG-INGRID-PT keystone v3 with OpenID Connect (experimental)


Feedback from Helpdesk

yearly review of the information registered into GOC-DB

2017-04-07

On a yearly basis, the information registered into GOC-DB need to be verified. NGIs and RCs have been asked to check them. In particular:

  1. NGI managers should review the people registered and the roles assigned to them, and in particular check the following information:
    • E-Mail
    • ROD E-Mail
    • Security E-Mail
NGI Managers should also review the status of the "not certified" RCs, in according to the RC Status Workflow;
  1. RCs administrators should review the people registered and the roles assigned to them, and in particular check the following information:
    • E-Mail
    • telephone numbers
    • CSIRT E-Mail
RC administrators should also review the information related to the registered service endpoints.

The process should be completed by Apr 28th.

To track the process, a series of tickets have been opened.

2017-05-15 UPDATE:

  • no feedback yet by: AfricaArabia, NGI_DE, NGI_FI, NGI_IL, NGI_NL, NGI_UA;
  • still reviewing: NGI_GRNET, NGI_HR, NGI_IBERGRID, NGI_IT, NGI_PL, NGI_RO, ROC_LA.

Failures with the updated CREAM probes

After the release of the updated CREAM probes on May 4th, several sites are failing the JobCancel and/or JobPurge ones (GGUS 128151):

  • the error message is: "Received timeout while fetching results".

The main reason is that in those CEs there isn't a job slot reserved for the ops tests.

As explained in the CREAM probes wiki:

  • JobCancel: cancel an active job
    • This metric submits a job directly to the selected CREAM CE, waits until the job gain the IDLE, RUNNING or REALLY-RUNNING state and then tries to cancel it. Finally it checks if the job has been correctly cancelled.
  • JobPurge: purge a terminted job
    • This metric is analogous of cream_jobCancel.py. It submits a short job (e.g. hostname.jdl), waits its termination (e.g DONE-OK) and then it tries to purge it. Finally, in order to verify the purging operation was successfully executed, the probe checks the job status by executing the glite-ce-job-status command which just in this scenario, must fail because the job doesn't exist anymore.

They both have a timeout of 15 minutes, so if the test job is not executed by that time, the probes return a failure. Please assign the ops jobs an higher priority and reserve them 1 job slot, they only require few seconds for being executed.

These failures didn't occur before May 4th because in the first version of the probes the returned status was "UNKNOWN" instead of the most proper one "CRITICAL".

List of failing CREAM-CEs from nagios (not all of them are affected by this problem):

  • 45 CREAM-CEs affected (13% of the total ones)
  • the sites can ask a recomputation of the May statistics

Monthly Availability/Reliability

Proposal to modify the declaration of scheduled interventions

Currently (see MAN02 Service intervention management) scheduled interventions (of any duration) MUST be declared at least 24 hours in advance, specifying reason and duration; any intervention declared less than 24 hours in advance will be considered unscheduled.

WLCG proposed the following modification:

  • a scheduled intervention shorter than 5 days must be declared at least 24 hours in advance
  • a scheduled intervention longer than 5 days must be declared at least 1 month in advance
  • any other intervention that don't fulfill the rules above will be considered unscheduled

2017-05-15 UPDATE

At the last OMB, the WLCG proposal was rejected:

  • the declaration rule for downtime longer than 5 days is too strict
  • difficult planning a downtime one month in advance (at least for non-WLCG sites)

Then it was proposed to extend the advance notice time from 24 hours to 5 days, but neither in this case the NGIs was in favour of it.

New proposal: Would you be in favour of extending the advance notice time to 3 days for scheduled downtimes of any duration?

Decommissioning EMI WMS

As discussed at the February and April/May OMBs, we are making plans for decommissioning the WMS and moving to DIRAC.

NGIs provided WMS usage statistics, and in general the usage is relatively low, mainly for local testing

Moderate usage by few VOs:

  • NGI_CZ: eli-beams.eu
  • NGI_GRNET: see
  • NGI_IT: calet.org, compchem, theophys, virgo
  • NGI_PL: gaussian, vo.plgrid.pl, vo.nedm.cyfronet
  • NGI_UK: mice, t2k.org

EGI contacted these VOs to agree a smooth migration of their activities to DIRAC, only some of them replied till now:

  • compchem is already testing DIRAC
  • calet.org: discussing with the users the migration to DIRAC. Interested in a webinar on DIRAC.
  • mice: enabled on the GridPP DIRAC server

We need the VO feedback for better defining technical details and timeline:

  • NGIs with VOs using WMS (not necessarily limited to the VOs above), please contact them to ensure that these VOs have a back-up plan.

WMS servers can be decommissioned as soon as the supported VOs do not need them any more. The proposal is:

  • WMS will be removed from production starting from 1st January 2018.
    • VOs have 8 months to find alternatives or migrate to DIRAC
  • Considering that this is not an update, the decommission can be performed in few weeks.

IPv6 readiness plans

    • Resource Centres: assess the IPv6 readiness of the site infrastructure (real machines, cloud managers)
      • NGIs/ROCs please start discussing with sites and provide suggestions for the overall plan

Decommissioning of dCache 2.10 and 2.13 (to modify)

  • support for the dCache 2.10 ended at December 2016
  • according to EGI policies, dCache 2.10 must be decommissioned https://wiki.egi.eu/wiki/PROC16_Decommissioning_of_unsupported_software
  • broadcast sent on Feb2 https://operations-portal.egi.eu/broadcast/archive/1631 + email sent on Feb7 to noc-managers mailing list
  • sites to upgrade their 2.10 endpoints to a newer "golden release" of dCache
    • 2.13, whose support ends on July 2017, which means in about 7 months from now, or
    • 2.16, whose support ends on May 2018: here take care that the dCache team does not support the upgrade from 2.10 directly to 2.16; only 2.10->2.13 and 2.13->2.16 transitions are supported.
  • decommissioning campaign will be started by EGI Operations to monitor the upgrade of the dCache 2.10 instances and follow up with the NGIs/sites;
  • deadline is end of April
  • probe will be WARNING for two months until April 17th, when it will switch to CRITICAL
  • in May EGI Operations will open tickets against sites still publishing dCache 2.10 and follow up on the upgrade plans
  • reference: https://www.dcache.org/downloads/1.9/index.shtml



  • support for the dCache 2.13 will end on July 2017
  • date of starting the campaign: May 1st (-3m)
  • date of ending the campaign: Aug 31st (+1m)
  • to be announced at OMB and in the April EGI Monthly Broadcast

Testing the new webdav probes

Site Host GGUSID note
CYFRONET-LCG2 se01.grid.cyfronet.pl https://ggus.eu/index.php?mode=ticket_info&ticket_id=128325 SOLVED
GRIF node12.datagrid.cea.fr https://ggus.eu/index.php?mode=ticket_info&ticket_id=128329
IGI-BOLOGNA darkstorm.cnaf.infn.it https://ggus.eu/index.php?mode=ticket_info&ticket_id=127930 SOLVED
INFN-T1 storm-fe-lhcb.cr.cnaf.infn.it, storm-fe.cr.cnaf.infn.it, storm-fe-archive.cr.cnaf.infn.it https://ggus.eu/index.php?mode=ticket_info&ticket_id=128326
NCG-INGRID-PT gftp01.ncg.ingrid.pt https://ggus.eu/index.php?mode=ticket_info&ticket_id=128327 SOLVED
UKI-NORTHGRID-LIV-HEP hepgrid11.ph.liv.ac.uk https://ggus.eu/index.php?mode=ticket_info&ticket_id=128328 SOLVED
egee.irb.hr lorienmaster.irb.hr

Missing steps:

Testing of the storage accounting

As discussed during the January OMB, the APEL team would need one site per NGI for testing the storage accounting. The eligible sites are the ones providing either dCache or DPM storage elements.

More information can be found in the following wiki: https://wiki.egi.eu/wiki/APEL/Storage

List of sites available for test.

2017-05-15 UPDATE:

  • 26 sites are sending storage accounting data (only from dCache and DPM SEs). The data has to be verified before deploying the script in production.
  • After the discussion at the March OMB, we are evaluating the creation of a new service type for monitoring the publication of storage accounting data.
Currently the accounting service types are:
  1. glite-APEL: for authorizing the sending of the messages
  2. APEL: to monitor the accounting data publication


Monitoring of the UNCERTIFIED sites

Information about the proposal for using GOCDB as the only source of topology information for ARGO:

  • Timescale:
    • New GOC-DB release on Dec 7th including a boolean ‘monitored’ flag for the service endpoints: DONE
    • Then creation of a web UI view for uncertified sites in ARGO: DONE
    • Uncertified sites will be asked to fill in the service endpoints information. Follow the How to add URL service endpoint information into GOC-DB DONE
    • As information is added in the GOCDB, uncertified sites/services will be picked up by the ARGO Monitoring Engine and they will start to be monitored DONE
    • By Q2 2017: support for multiple service endpoints


Nagios server for the uncertified sites: https://argo-mon-uncert.cro-ngi.hr/nagios/

  • Configuration is regenerated every hour
  • uncertified sites report on the ARGO development instance
  • IMPORTANT: for being correctly monitored, the uncertified sites have to fill in the proper services information into GOC-DB: please follow the HOWTO21

PROC09 modified accordingly.

VAPOR

  • VAPOR 2.2 released on March 16th
  • important for presenting the amount of computing and storage resources of the infrastructure
  • There are several improvements and new features: the computation of values of CPU and storages have been deeply reviewed, nevertheless some values are still not in line with the reality.

Next version will be focused on these computations to be able to provide better figures.

  • Please have a look at the information displayed and report us any inconsistency you should spot.

AOB

Next meeting