Difference between revisions of "Tier1 Operations Report 2016-02-10"

From GridPP Wiki
Jump to: navigation, search
()
()
 
(2 intermediate revisions by one user not shown)
Line 11: Line 11:
 
* On Wednesday afternoon (27th Jan) one of the Windows hypervisors crashed. This was a local storage node - with no failover to other hypervisors. A number of services were unavailable or degraded during the afternoon and an outage for those affected services in the GOC DB was declared.
 
* On Wednesday afternoon (27th Jan) one of the Windows hypervisors crashed. This was a local storage node - with no failover to other hypervisors. A number of services were unavailable or degraded during the afternoon and an outage for those affected services in the GOC DB was declared.
 
* The OPN link to CERN has been very busy during this last fortnight.
 
* The OPN link to CERN has been very busy during this last fortnight.
* We have had a problem with the Atlas HammerCloud Tests and running Atlas analysis jobs. Initially this was thought to be due to a missing disk server (GDSS677). However, it subsequently became clear that the cause was the CVMFS inode count on teh Worker Nodes exceeding 2**32. This is a problem that we had seen some months ago.  
+
* We have had a lot of failures of the Atlas HammerCloud Tests and running Atlas analysis jobs. Initially this was thought to be due to a missing disk server (GDSS677). However, it subsequently became clear that the cause was the CVMFS inode count on the Worker Nodes exceeding 2**32. This is a problem that we had seen some months ago.  
 
* During last week we had a problem with a particular Atlas Tape. We were unable to read most of the files off the tape. We have declared 5898 files lost from this tape. This was a T10KC tape and it is being sent off for analysis to see if we can understand what happened.
 
* During last week we had a problem with a particular Atlas Tape. We were unable to read most of the files off the tape. We have declared 5898 files lost from this tape. This was a T10KC tape and it is being sent off for analysis to see if we can understand what happened.
 
<!-- ***********End Review of Issues during last week*********** ----->
 
<!-- ***********End Review of Issues during last week*********** ----->
Line 98: Line 98:
 
** Migration of data from T10KC to T10KD tapes (affects Atlas & LHCb data).
 
** Migration of data from T10KC to T10KD tapes (affects Atlas & LHCb data).
 
* Networking:
 
* Networking:
** Replace the UKLight Router.
+
** Replace the UKLight Router. Then upgrade the 'bypass' link to the RAL border routers to 2*10Gbit.
 
* Fabric
 
* Fabric
 
** Firmware updates on remaining EMC disk arrays (Castor, LFC)
 
** Firmware updates on remaining EMC disk arrays (Castor, LFC)

Latest revision as of 09:48, 10 February 2016

RAL Tier1 Operations Report for 10th February 2016

Review of Issues during the week 27th January to 3rd February 2016.
  • On Wednesday afternoon (27th Jan) one of the Windows hypervisors crashed. This was a local storage node - with no failover to other hypervisors. A number of services were unavailable or degraded during the afternoon and an outage for those affected services in the GOC DB was declared.
  • The OPN link to CERN has been very busy during this last fortnight.
  • We have had a lot of failures of the Atlas HammerCloud Tests and running Atlas analysis jobs. Initially this was thought to be due to a missing disk server (GDSS677). However, it subsequently became clear that the cause was the CVMFS inode count on the Worker Nodes exceeding 2**32. This is a problem that we had seen some months ago.
  • During last week we had a problem with a particular Atlas Tape. We were unable to read most of the files off the tape. We have declared 5898 files lost from this tape. This was a T10KC tape and it is being sent off for analysis to see if we can understand what happened.
Resolved Disk Server Issues
  • GDSS620 (GenTape - D0T1) failed on the first January. The server was returned to service on 1st February. This server had failed on repeat occasions before. Two disks were replaced in the server.
  • GDSS682 (AtlasDataDisk - D1T0) failed on Sunday evening (31st Jan). It was returned to service at the end of the afternoon the following day although investigations had failed to find a cause. There was one corrupt file found on gdss682 when it returned to production. This file was declared lost to Atlas.
  • GDSS667 (AtlasScratchDisk - D1T0) Failed on Monday 18th Jan with a read-only file system. On investigation three disks in the RAID set had problems. Following a lot of work a small number of files were recovered from the server. However, the large majority of the files were declared lost to Atlas. The server re-ran the acceptance tests before being put back into service on Friday 5th Feb. As reported a fortnight ago most of the files on this server were reported lost to Atlas.
  • GDSS744 (AtlasDataDisk - D1T0) failed on Sunday evening (7th Feb). A RAID consistency check threw out a bad drive. Following the replacement of that drive the server was returned to service yesterday afternoon (9th Feb). It is initially in read-only mode - and will be put back in read/write once the RAID consistency check has been re-run.
  • GDSS706 (AtlasDataDisk - D1T0) was taken out of service for a couple of yours yesterday (9th Feb) for the battery on the RAID card to be replaced.
Current operational status and issues
  • There is a problem seen by LHCb of a low but persistent rate of failure when copying the results of batch jobs to Castor. There is also a further problem that sometimes occurs when these (failed) writes are attempted to storage at other sites. A recent modification has improved, but not completed fixed this.
  • The intermittent, low-level, load-related packet loss seen over external connections is still being tracked. Likewise we have been working to understand some remaining low level of packet loss seen within a part of our Tier1 network.
Ongoing Disk Server Issues
  • None
Notable Changes made since the last meeting.
  • ARC-CE02,3 & 4 have been upgraded to version 5.0.5. (ARC-CE01 was done last week).
  • Batches of worker nodes are being drained to reset the CVMFS inodes count.
  • Condor is being updated (to version 8.4.4) on a couple of batches of worker nodes. (These nodes are being drained and rebooted - which also resolves the CVMFS inodes problem on them).
Declared in the GOC DB
  • None
Advanced warning for other interventions
The following items are being discussed and are still to be formally scheduled and announced.
  • Update the repack Castor repack instance from version 2.1.14.13 to 2.1.14.15. (Proposed for 10/2/16)

Listing by category:

  • Databases:
    • Switch LFC/3D to new Database Infrastructure.
  • Castor:
    • Update SRMs to new version (includes updating to SL6).
    • Update to Castor version 2.1.15.
    • Migration of data from T10KC to T10KD tapes (affects Atlas & LHCb data).
  • Networking:
    • Replace the UKLight Router. Then upgrade the 'bypass' link to the RAL border routers to 2*10Gbit.
  • Fabric
    • Firmware updates on remaining EMC disk arrays (Castor, LFC)
  • Grid Services
    • A Load Balancer (HAProxy) will be used in front of the FTS service.
Entries in GOC DB starting since the last report.
Service Scheduled? Outage/At Risk Start End Duration Reason
arc-ce04 SCHEDULED WARNING 01/02/2016 11:00 01/02/2016 12:00 1 hour Upgrade ARC-CE04 to version 5.0.5
arc-ce02, arc-ce03 SCHEDULED WARNING 28/01/2016 12:00 28/01/2016 13:00 1 hour Upgrading ARC-CE02 & ARC-CE03 to version 5.0.5
lcgvo07, lcgvo08, lcgwms04 UNSCHEDULED OUTAGE 27/01/2016 13:00 27/01/2016 17:30 4 hours and 30 minutes Some services unavailable following problem on hypervisor.
Open GGUS Tickets (Snapshot during morning of meeting)
GGUS ID Level Urgency State Creation Last Update VO Subject
119456 Green Urgent In Progress 2016-02-10 2016-02-10 CMS File waiting for staging
119450 Green Less Urgent In Progress 2016-02-09 2016-02-09 Atlas RAL_LCG2: SOURCE SRM_GET_TURL error on the turl request
119389 Green Urgent In Progress 2016-02-05 2016-02-05 LHCb Data transfer problem to RAL BUFFER
118809 Green Urgent On Hold 2016-01-05 2016-01-13 Towards a recommendation on how to configure memory limits for batch jobs
117683 Green Less Urgent In Progress 2015-11-18 2016-01-05 CASTOR at RAL not publishing GLUE 2
116864 Red Urgent In Progress 2015-10-12 2016-01-29 CMS T1_UK_RAL AAA opening and reading test failing again...
109358 Green Less Urgent Waiting Reply 2015-02-05 2016-02-09 SNO+ RAL WMS unavailable
Availability Report

Key: Atlas HC = Atlas HammerCloud (Queue ANALY_RAL_SL6, Template 508); CMS HC = CMS HammerCloud

Day OPS Alice Atlas CMS LHCb Atlas HC CMS HC Comment
27/01/16 100 100 99 96 96 77 100 Atlas & CMS: VM crash took out one of site BDIIs. Tests could not find resources.; LHCb: Single SRM test failure.
28/01/16 100 100 100 100 100 76 100
29/01/16 100 100 100 100 100 72 N/A
30/01/16 100 100 100 96 100 54 N/A Single SRM Test failure (“User timeout error”)
31/01/16 100 100 100 96 100 58 100 Single SRM Test failure (“User timeout error”)
01/02/16 100 100 100 100 100 62 100 Atlas HC failing with "Trf exit code 40. trans: Athena crash - consult log file"
02/02/16 100 100 100 100 100 63 100
03/02/16 100 100 100 100 100 61 N/A
04/02/16 100 100 100 100 96 48 100 Single SRM Test failure ([SRM_INVALID_PATH] No such file or directory)
05/02/16 100 100 100 100 100 42 100
06/02/16 100 100 100 100 100 88 N/A
07/02/16 100 100 100 100 100 36 100
0/02/16 100 100 100 100 100 78 100
09/02/16 100 100 100 100 100 83 100