Tags:
dcache1Add my vote for this tag create new tag
view all tags

Service Card for dCache

dCache is our mass Grid Storage resource. It serves more than 1 Petabyte of space to all VOs, mainly using GridFTP and dcap access protocols.

Definition

This is the list of nodes with related dCache services:

Nodes Services hosted
storage01 srm, utility, httpd, info, gridftp, dcap, BDII
storage02 lm, dCache, gPlazma, chimera, chimera-nfs, dir, admin
se[01-10,13-22] gridftp
se[01-10,13-22] pool

Please note that the only dependency is that storage01 (srm) needs to have /pnfs mounted from storage02 (chimera-nfs)

This page shows the space reservations in place (currently a bit obsolete): Dcache_SpaceTokens_sm

The most up to date information can be found in the dCache web interface

Hardware overview:

  • Head nodes storage[01-02] are installed on IBM hardware. SL 6.x x86_64 with dCache 2.6.x series. IB IB stack
  • There are two types of storage pools:
    1. se[01-08] are IBM hosts connected to IBM DC3500 controllers configured with SL 5.x x86_64, dCache 2.6 series, Mellanox IB packages and mpp for fibrechannel multipath. There is one FC card per host connected to the two channels of the IBM DC3500 controller.
    2. se[09-16] are IBM hosts connected to IBM DCS3700 controllers. SL 6.x, dCache 2.6, SL IB and multipath packages.
    3. se[17-22] are IBM or HP hosts connected to NetApp controllers. SL 6.x, dCache 2.6, SL IB and multipath packages.

Operations

Client tools

There are several ways to connect to dCache internal interface. Using the graphical pcells:

  • You can run the GUI (pcells) If you have good bandwidth with pub or ui64, by ssh-ing into it (as a user) and run dcache-gui. The first time you need to configure a new connection to storage02.
  • Alternatively, you can run pcells in your desktop, configure it to connect to localhost, and first create a tunnel through pub: ssh user@pub.lcg.cscs.ch -L 22223:storage02:22223
Or shell-based console
  • To be able to access the ssh shell console in dCache, you need to connect to it with: ssh -2 admin@storage02 -i /usr/libexec/nagios-plugins/.ssh/ssh_host_dsa_key -p 22224
  • You can use Derek's DcacheShellutils
For the shell-based console, you can use a password, or an ssh1 key pair and copy it to cfengine controlled special dCache authorized_keys. The process is described here: http://trac.dcache.org/projects/dcache/wiki/manuals/admin_interface

Working with space reservations

  • Note: Space displayed in the web interface is shown as MiB whereas when using the SSH interface it appears to be listed in bytes. As show in later commands when performing updates it is possible to specify units.
    • 1Mib = 1.04858MB
    • Documentation for the commands used with reservations can be found by running help inside the SrmSpaceManager cell.

  • If you want to see the space tokens in the system with detailed information you need to perform an 'ls' in the SrmSpaceManager, e.g.

[storage02.lcg.cscs.ch] (local) admin > cd SrmSpaceManager
[storage02.lcg.cscs.ch] (SrmSpaceManager) admin > ls
Reservations:
34718295  voGroup:/lhcb     voRole:null          retentionPolicy:OUTPUT   accessLatency:ONLINE  linkGroupId:0  size:0                created:Wed  Jun  19  14:43:49  CEST  2013  lifetime:-1ms       expiration:NEVER  description:LHCb_USER            state:RESERVED  used:0                allocated:0
223       voGroup:/atlas    voRole:production    retentionPolicy:REPLICA  accessLatency:ONLINE  linkGroupId:4  size:780000000000000  created:Tue  Feb  19  10:33:10  CET   2008  lifetime:-1ms       expiration:NEVER  description:ATLASDATADISK        state:RESERVED  used:670321496694602  allocated:0
54172751  voGroup:/cms      voRole:priorityuser  retentionPolicy:REPLICA  accessLatency:ONLINE  linkGroupId:2  size:663107670        created:Tue  Sep  30  10:42:53  CEST  2014  lifetime:2399985ms  expiration:Tue    Sep                              30              11:22:53              CEST                  2014  description:null  state:RESERVED  used:0  allocated:663107670
1941364   voGroup:atlas/ch  voRole:null          retentionPolicy:OUTPUT   accessLatency:ONLINE  linkGroupId:4  size:10000000000000   created:Tue  Sep  09  14:02:26  CEST  2008  lifetime:-1ms       expiration:NEVER  description:ATLASLOCALGROUPDISK  state:RESERVED  used:8322044239556    allocated:0
1631505   voGroup:atlas     voRole:production    retentionPolicy:OUTPUT   accessLatency:ONLINE  linkGroupId:4  size:10000000000000   created:Fri  Jul  18  12:09:44  CEST  2008  lifetime:-1ms       expiration:NEVER  description:ATLASPRODDISK        state:RESERVED  used:4919552759109    allocated:3551320456
222       voGroup:/atlas    voRole:production    retentionPolicy:REPLICA  accessLatency:ONLINE  linkGroupId:4  size:100000000000     created:Tue  Feb  19  10:31:56  CET   2008  lifetime:-1ms       expiration:NEVER  description:ATLASMCDISK          state:RESERVED  used:39394445860      allocated:0
34718259  voGroup:/lhcb     voRole:null          retentionPolicy:OUTPUT   accessLatency:ONLINE  linkGroupId:0  size:230000000000000  created:Wed  Jun  19  14:41:05  CEST  2013  lifetime:-1ms       expiration:NEVER  description:LHCb-Disk            state:RESERVED  used:85891952795361   allocated:0
3205575   voGroup:/atlas    voRole:production    retentionPolicy:OUTPUT   accessLatency:ONLINE  linkGroupId:4  size:38000000000000   created:Mon  Apr  06  15:14:02  CEST  2009  lifetime:-1ms       expiration:NEVER  description:ATLASSCRATCHDISK     state:RESERVED  used:17218172543300   allocated:0
   total number of reservations: 8
   total number of bytes reserved: 1068100663107670
   

  • To show only the reservations for a specific link group

[storage02.lcg.cscs.ch] (SrmSpaceManager) admin > ls -lg=atlas-linkGroup

Found LinkGroup:
4        Name:atlas-linkGroup  FreeSpace:137832997369994  ReservedSpace:137260646735722  AvailableSpace:572350634272  VOs:{atlassgm:*}{/atlas:*}{/atlas/lcg1:*}{altasprd:*}{/atlas/ch:*}{atlas001:*}  onlineAllowed:true    nearlineAllowed:false  replicaAllowed:true  custodialAllowed:false  outputAllowed:true  UpdateTime:Tue  Sep   30             11:07:30          CEST                             2014(1412068050125)
223      voGroup:/atlas        voRole:production          retentionPolicy:REPLICA        accessLatency:ONLINE         linkGroupId:4                                                                   size:780000000000000  created:Tue            Feb                  19                      10:33:10            CET             2008  lifetime:-1ms  expiration:NEVER  description:ATLASDATADISK        state:RESERVED       used:670321496694602  allocated:0
222      voGroup:/atlas        voRole:production          retentionPolicy:REPLICA        accessLatency:ONLINE         linkGroupId:4                                                                   size:100000000000     created:Tue            Feb                  19                      10:31:56            CET             2008  lifetime:-1ms  expiration:NEVER  description:ATLASMCDISK          state:RESERVED       used:39394445860      allocated:0
1941364  voGroup:atlas/ch      voRole:null                retentionPolicy:OUTPUT         accessLatency:ONLINE         linkGroupId:4                                                                   size:10000000000000   created:Tue            Sep                  09                      14:02:26            CEST            2008  lifetime:-1ms  expiration:NEVER  description:ATLASLOCALGROUPDISK  state:RESERVED       used:8322044239556    allocated:0
3205575  voGroup:/atlas        voRole:production          retentionPolicy:OUTPUT         accessLatency:ONLINE         linkGroupId:4                                                                   size:38000000000000   created:Mon            Apr                  06                      15:14:02            CEST            2009  lifetime:-1ms  expiration:NEVER  description:ATLASSCRATCHDISK     state:RESERVED       used:17218298580729   allocated:0
1631505  voGroup:atlas         voRole:production          retentionPolicy:OUTPUT         accessLatency:ONLINE         linkGroupId:4                                                                   size:10000000000000   created:Fri            Jul                  18                      12:09:44            CEST            2008  lifetime:-1ms  expiration:NEVER  description:ATLASPRODDISK        state:RESERVED       used:4938119303531    allocated:14892375

  • To create a new token perform in the same Cell:
    • Syntax is as follows; reserve [-vog=voGroup] [-vor=voRole] [-acclat=AccessLatency] [-retpol=RetentionPolicy] [-desc=Description] [-lgid=LinkGroupId] [-lg=LinkGroupName]
    • The below example creates a reservation for 10TB under link group "atlas-linkGroup" that will not expire due to the "-1" lifetime which mean infinite.

   reserve -vog=/atlas -vor=production -lg=atlas-linkGroup -acclat=ONLINE -retpol=OUTPUT -desc=ATLASPRODDISK 10TB "-1"
   

  • To increase the size of a token (aka set the new max size), note the "222" is the id of the reservation we intend to update.

   update space reservation -size=15TB 222
   

Example for Swiss users

  • ATLAS: We need to increase the ATLASLOCALGROUPDISK (ID 1941364) reservation by 150TB, it already exists as a 10TB reservation so we need to update its space to 160TB

   update space reservation -size=160TB 1941364
   

  • CMS: There are no space reservations for CMS, however CH users can be identified with the "/cms/chcms" attribute.

   reserve -vog=/cms/ch -vor=chcms -lg=cms-linkGroup -acclat=ONLINE -retpol=OUTPUT -desc=CHCMS 150TB "-1"
   

  • LHCb: As LHCb can't easily distinguish CH users we just need to update the LHCb-Disk (ID 34718259) reservation. This was 230TB and we are adding 60TB

   update space reservation -size=290TB 34718259
   

Testing

There are two tools installed in ui64 to test dCache. They are in the path (/opt/cscs/bin) and you need a valid grid proxy to use them.

  • chk_SE-dcache. It tests the basic grid protocols with non-lcg tools, like gridftp, dccp and srmcp. It does not depend on external tools like the LFC or the BDII.
  • chk_SE-lcgtools. It performs more complicated operations, using lcg-* tools. It requires proper BDII publishing and access to the LFC at CERN/FZK.. It also checks space tokens (with -t)

Start/stop procedures

  • service dcache.This is the basic dcache (link to /opt/d-cache/bin/dcache) and can be used for most operations. It is automatically started on boot. For example, you can restart all services in one host with service dcache restart or only one of the services with service dcache restart gPlazma
There is a more detailed document about a Full restart here: DcacheFullRestart

Adding new pools

  1. Create the arrays and mountpoints
  2. Create the pools as follows (showing an example of a 30T system with 2 pools for ATLAS):
    Jul 15 14:40 [root@se18:~]# for i in 6 7; do echo dcache pool create --size=31990000000000   /data_phaseH_storage_r6_1_18_array${i}/pool data_phaseH_storage_r6_1_18_array${i}_atlas data_phaseH_storage_r6_1_18_array${i}_atlas; done
  3. Make sure the file storage01:/etc/dcache/layouts/${HOSTNAME}.conf is correct and put into Configuration Management Software
  4. Add the pools to the corresponding pool groups and pool Linkgroups on storage01:/var/lib/dcache/config/poolmanager.conf
  5. Reload the pool list using dCache interface

Checking logs

Please go to /var/log/dcache/ to find relevant logs.

Chimera migration

This procedure is already done, and not really necessary anymore. Check ChimeraMigration

Solaris to Linux migration

This procedure is already done, and not really necessary anymore. Check SolarisToLinuxMigrationDcache

Dealing with File Loss

Normally, file loss is not a terrible loss. You need to inform about this, and hopefully they can be able to give you back the files.

  • Prepare a pnfs name list of the lost files and send this list to the site contacts. They responsible to take it from there to his experiment.
  • reestablish dcache/pnfs consistency by deleting leftover pnfs names with no on-site replicas.
    • SQL to obtain pnfsids for files on one pool that are only in that pool (to see what's lost):
      select ipnfsid, from t_locationinfo where ipnfsid in (select ipnfsid from t_locationinfo where ilocation='ibm01_data1_cms') group by ipnfsid having (count(ilocation) = 1);
    • SQL to delete the trash table (deletions on pools) after doing a massive RM on the /pnfs tree.
      delete  from t_locationinfo_trash where ilocation='ibm01_data1_cms';
CMS site contacts:
  • Follow this procedure written by Leo for the central CERN wiki in order to map the files to data sets, and delete the replica information from the phedex DB.

Migrate data out of a set of pools.

  1. You probably want start with the default configuration before touching things
    cd PoolManager
    reload -yes
    
  2. In dCache pcells interface, set a pool read-only:
    cd PoolManager
    psu set pool se12_cms rdonly
  3. Create a special group of pools to use as a target for migration. Check below for the whole list
    cd PoolManager
    psu create pgroup cms_migration
    psu addto pgroup cms_migration se31_cms
  4. In dCache pcells interface, start migration of data out of that pool
    cd se12_cms
    migration move -target=pgroup -concurrency=5 cms_migration 
    1. Another option here is to migrate data of a specific VO off a certain pool with this:
      migration move -target=pgroup -storage=cms:cms -concurrency=4 pgroup_of_CMS_pools
  5. Check the state of the migration
    migration info 1
  6. You may want to disable the gridftp door on that node, so that, at the end of the migration, you can just shut down the dcache service.
    cd LoginBroker
    disable GFTP-se12 
  7. Before shutting down the pool, you may want to verify that dCache is not aware of any file that could be left inside the pool - there are always some inconsistencies. You can check this asking the chimera node:
    [root@storage02:~]# psql chimera -U chimera -c "select count(*) from t_locationinfo where ilocation='se15_cms';"

Enable/Disable a door

$ ssh phoenix1
$ ssh -l admin -p 22224 storage02.lcg.cscs.ch
cd LoginBroker
disable GFTP-se18
disable GFTP-se17 

Change the number of transfers (movers) that a door accepts

$ export DCACHE_NUM_MOVERS_DOOR=300
$ echo -e 'cd PoolManager\npsu ls pool\n..\nlogoff\n' | ssh -l admin -p 22224 storage02.lcg.cscs.ch  2>&1 | grep data_ | tr -d $'\r' &> /tmp/pools
$ for pool in $(cat /tmp/pools); do echo "cd ${pool}"; echo "mover set max active ${DCACHE_NUM_MOVERS_DOOR} -queue=regular"; echo ".."; done | ssh -l admin -p 22224 storage02.lcg.cscs.ch  2>&1

Move a set of pools to a different dCache server

Sometimes it is required to move pools from a machine (A) to another system (B) to do tasks such as HW maintenance. This is a relatively easy task with little downtime.

  1. Disable the door running on A as shown before
  2. Stop dCache on A
    # dcache stop
  3. Umount the Logical Volumes to move on A
    # umount /data_phaseG_storage_10_array0 # umount /data_phaseG_storage_10_array1 # umount /data_phaseG_storage_10_array2
  4. On A comment out the mounts on /etc/fstab
  5. On B add the mounts to /etc/fstab
  6. Stop cfengine on B ('service cfexecd stop' and 'service cfenvd stop')
  7. On B edit the option 'volume_list' in /etc/lvm/lvm.conf adding the storage from host A
  8. Deactivate the Logical Volumes on A
    # lvchange -a n phaseG_storage_10_array0 phaseG_storage_10_array1 phaseG_storage_10_array2
  9. Make sure volumes can be seen by B by running
    # multipath -ll
  10. Activate the Logical Volumes deactivated before, on server B
    # lvchange -a y phaseG_storage_10_array0 phaseG_storage_10_array1 phaseG_storage_10_array2
  11. Make sure the correct mount points exist on B
  12. Mount the volumes with
    # mount -a
  13. Copy the proper entries of /etc/dcache/layouts/A.conf to /etc/dcache/layouts/B.conf
  14. Start all pools in dcache on B.
    # dcache start

List the files on a Space Token

Connect to the SrmSpaceManager and issue the following commands:

  1. Find out which is the ID of the Space Token you want (first column under Reservations):
    # ls
    Reservations:
    3205575 voGroup:/atlas voRole:production retentionPolicy:OUTPUT accessLatency:ONLINE linkGroupId:4 size:28000000000000 created:Mon Apr 06 15:14:02 CEST 2009 lifetime:-1ms expiration:NEVER description:ATLASSCRATCHDISK state:RESERVED used:21799242579873 allocated:0 
    222 voGroup:/atlas voRole:production retentionPolicy:REPLICA accessLatency:ONLINE linkGroupId:4 size:100000000000 created:Tue Feb 19 10:31:56 CET 2008 lifetime:-1ms expiration:NEVER description:ATLASMCDISK state:RESERVED used:39394445860 allocated:0 
    4254214 voGroup:/atlas voRole:production retentionPolicy:OUTPUT accessLatency:ONLINE linkGroupId:4 size:50000000000000 created:Wed Sep 16 16:21:31 CEST 2009 lifetime:-1ms expiration:NEVER description:ATLASGROUPDISK state:RESERVED used:1322228694829 allocated:0 
    1631505 voGroup:atlas voRole:production retentionPolicy:OUTPUT accessLatency:ONLINE linkGroupId:4 size:10000000000000 created:Fri Jul 18 12:09:44 CEST 2008 lifetime:-1ms expiration:NEVER description:ATLASPRODDISK state:RESERVED used:1831550835057 allocated:0 
    1941364 voGroup:atlas/ch voRole:null retentionPolicy:OUTPUT accessLatency:ONLINE linkGroupId:4 size:10000000000000 created:Tue Sep 09 14:02:26 CEST 2008 lifetime:-1ms expiration:NEVER description:ATLASLOCALGROUPDISK state:RESERVED used:3490336858842 allocated:0 
    223 voGroup:/atlas voRole:production retentionPolicy:REPLICA accessLatency:ONLINE linkGroupId:4 size:710000000000000 created:Tue Feb 19 10:33:10 CET 2008 lifetime:-1ms expiration:NEVER description:ATLASDATADISK state:RESERVED used:605825455405365 allocated:0 
    4254201 voGroup:/atlas voRole:production retentionPolicy:OUTPUT accessLatency:ONLINE linkGroupId:4 size:3000000000000 created:Wed Sep 16 16:20:39 CEST 2009 lifetime:-1ms expiration:NEVER description:ATLASHOTDISK state:RESERVED used:1060540539738 allocated:0 
    total number of reservations: 7
    total number of bytes reserved: 811100000000000
    
    LinkGroups:
    3 Name:dteam-linkGroup FreeSpace:20008531276 ReservedSpace:0 AvailableSpace:20008531276 VOs:{dteamsgm:*}{/dteam:*}{dteam001:*} onlineAllowed:true nearlineAllowed:false replicaAllowed:true custodialAllowed:false outputAllowed:true UpdateTime:Fri Apr 12 14:03:42 CEST 2013(1365768222283)
    1 Name:dech-linkGroup FreeSpace:21474667014 ReservedSpace:0 AvailableSpace:21474667014 VOs:{dechsgm:*}{/dech:*}{dechprd:*}{dech001:*} onlineAllowed:true nearlineAllowed:false replicaAllowed:true custodialAllowed:false outputAllowed:true UpdateTime:Fri Apr 12 14:03:42 CEST 2013(1365768222283)
    5 Name:ops-linkGroup FreeSpace:42949660822 ReservedSpace:0 AvailableSpace:42949660822 VOs:{/ops/NGI/Germany:*}{opssgm:*}{/ops:*}{opsprd:*}{ops001:*} onlineAllowed:true nearlineAllowed:false replicaAllowed:true custodialAllowed:false outputAllowed:true UpdateTime:Fri Apr 12 14:03:42 CEST 2013(1365768222283)
    6 Name:hone-linkGroup FreeSpace:313104425916 ReservedSpace:0 AvailableSpace:313104425916 VOs:{/hone:*}{honeprd:*}{honesgm:*}{hone001:*} onlineAllowed:true nearlineAllowed:false replicaAllowed:true custodialAllowed:false outputAllowed:true UpdateTime:Fri Apr 12 14:03:42 CEST 2013(1365768222283)
    4 Name:atlas-linkGroup FreeSpace:178634678467080 ReservedSpace:175731250640436 AvailableSpace:2903427826644 VOs:{atlassgm:*}{/atlas:*}{/atlas/lcg1:*}{altasprd:*}{/atlas/ch:*}{atlas001:*} onlineAllowed:true nearlineAllowed:false replicaAllowed:true custodialAllowed:false outputAllowed:true UpdateTime:Fri Apr 12 14:03:42 CEST 2013(1365768222283)
    0 Name:lhcb-linkGroup FreeSpace:1288458125084 ReservedSpace:0 AvailableSpace:1288458125084 VOs:{lhcb001:*}{lhcbsgm:*}{lhcbprd:*}{/lhcb:*} onlineAllowed:true nearlineAllowed:false replicaAllowed:true custodialAllowed:false outputAllowed:true UpdateTime:Fri Apr 12 14:03:42 CEST 2013(1365768222283)
    2 Name:cms-linkGroup FreeSpace:252388091630454 ReservedSpace:0 AvailableSpace:252388091630454 VOs:{cms001:*}{cmssgm:*}{/cms/itcms:*}{/cms:*}{cmsprd:*} onlineAllowed:true nearlineAllowed:false replicaAllowed:true custodialAllowed:false outputAllowed:true UpdateTime:Fri Apr 12 14:03:42 CEST 2013(1365768222283)
    total number of linkGroups: 7
    total number of bytes reservable: 256977514867210
    total number of bytes reserved  : 175731250640436
    last time all link groups were updated: Fri Apr 12 14:03:42 CEST 2013(1365768222283)
  2. For the specified Space Token ID, issue this command (assuming id is 4254214):
    # listFilesInSpace 4254214
    21196092 /atlas production 4254214 20 1339879906842 119991 /pnfs/lcg.cscs.ch/atlas/atlasgroupdisk/SAM/testfile-put-ATLASGROUPDISK-1339879765-0d8b925d623c.txt 0000FB317A1A58F7420481C01EA74176DBCF Stored 0
    7991666 /atlas production 4254214 242835872 1287175575623 14399992 /pnfs/lcg.cscs.ch/atlas/atlasgroupdisk/perf-egamma/data10_7TeV/NTUP_TRT/f295_p246/data10_7TeV.00166658.physics_Egamma.merge.NTUP_TRT.f295_p246_tid177325_00/NTUP_TRT.177325._000020.root.1 0000D11E14AF004B4BE9A0DAD90351DCDFE4 Stored 0
    [...]

Removing and modifying space tokens

Find the space token size and ID

cd SrmSpaceManager
ls

Ensure no files are present

listFilesInSpace 4254201

Remove the space token

release 4254201

After waiting for at least 30 minutes add the free space to another token. The size is the total size for the space token (add freed to old existing space).

update space reservation -size=703000000000000 223

ATLAS FAX

The ATLAS FAX service depends of two elements:

  • xrootD door in storage01
    [root@storage01:~]# dcache status |grep xrootd-storage01Domain
    xrootd-storage01Domain  running (for 8 weeks) 8985  root
    Do not confuse it with the other rooted domains running on the system, that are for CMS.
  • xrootD service in atlas01
    [root@atlas01:~]# service xrootd status
If either of both is not working, the system will show RED status in the ATLAS FAX monitoring page for CSCS.

Reference: https://twiki.cern.ch/twiki/bin/view/AtlasComputing/FAXdCacheN2Nstorage

Note: Please note that the certificate in /etc/grid-security/xrd/xrdcert.pem

needs to be equal to the one in /etc/grid-security/hostcert.pem

Set up

Installation (kept for historical reasons)

You need to perform some steps, some taken from http://www.dcache.org/manuals/Book/start/in-install.shtml

  • After the installation, you need to install postgres 8.3 by hand but first make use there is a separate partition for /var/lib/pgsql:
    lvcreate -L 55G -n lv_pgsql vg_root
    mkfs.ext3 /dev/vg_root/lv_pgsql
    mv /var/lib/pgsql /var/lib/pgsql.backup
    mkdir /var/lib/pgsql
    echo '/dev/vg_root/lv_pgsql   /var/lib/pgsql          ext3    defaults        1 2' >> /etc/fstab
    mount -a
    chmod 700 /var/lib/pgsql
    chown postgres.postgres /var/lib/pgsql
    cp -a /var/lib/pgsql.backup/. /var/lib/pgsql/
    
    rpm -e --nodeps postgresql-libs.x86_64 postgresql-libs.i386
    rpm -U ftp://ftp.postgresql.org/pub/binary/v8.3.9/linux/rpms/redhat/rhel-5-x86_64/compat-postgresql-libs-4-1PGDG.rhel5.x86_64.rpm \
    ftp://ftp.postgresql.org/pub/binary/v8.3.9/linux/rpms/redhat/rhel-5-x86_64/postgresql-8.3.9-1PGDG.rhel5.x86_64.rpm \
    ftp://ftp.postgresql.org/pub/binary/v8.3.9/linux/rpms/redhat/rhel-5-x86_64/postgresql-libs-8.3.9-1PGDG.rhel5.x86_64.rpm \
    ftp://ftp.postgresql.org/pub/binary/v8.3.9/linux/rpms/redhat/rhel-5-x86_64/postgresql-server-8.3.9-1PGDG.rhel5.x86_64.rpm \
    ftp://ftp.postgresql.org/pub/binary/v8.3.9/linux/rpms/redhat/rhel-5-i386/compat-postgresql-libs-4-1PGDG.rhel5.i686.rpm
    
    # yum install postgresql-server compat-postgresql-libs.i686 compat-postgresql-libs.x86_64
    service postgresql initdb
    sed -i 's/ident sameuser/trust/' /var/lib/pgsql/data/pg_hba.conf
    service postgresql restart
    createuser -U postgres --no-superuser --no-createrole --createdb srmdcache
    createdb -U srmdcache dcache
    createdb -U srmdcache billing
  • Prepare chimera in the right node (Node02)
    createdb -U postgres chimera
    createuser -U postgres --no-superuser --no-createrole --createdb chimera
    psql -U chimera chimera -f /opt/d-cache/libexec/chimera/sql/create.sql
    createlang -U postgres plpgsql chimera
    psql -U chimera chimera -f /opt/d-cache/libexec/chimera/sql/pgsql-procedures.sql
    echo "/ localhost(rw)
    /pnfs *.lcg.cscs.ch(rw)" > /etc/exports
    
    ### We create the basics
    /opt/d-cache/libexec/chimera/chimera-cli.sh Mkdir /pnfs
    /opt/d-cache/libexec/chimera/chimera-cli.sh Mkdir /pnfs/lcg.cscs.ch
    echo "chimera" | /opt/d-cache/libexec/chimera/chimera-cli.sh Writetag /pnfs/lcg.cscs.ch sGroup
    echo "StoreName sql" | /opt/d-cache/libexec/chimera/chimera-cli.sh Writetag /pnfs/lcg.cscs.ch OSMTemplate
    ln -s /opt/d-cache/libexec/chimera/chimera-nfs-run.sh /etc/init.d/chimera-nfs
    service chimera-nfs start
    # You should be able to mount it from outside with:  mount storage22.lcg.cscs.ch:/pnfs /pnfs
    
    ### If we want we can populate the VO dirs: 
    mount localhost:/ /mnt
    cd /mnt/pnfs/lcg.cscs.ch 
    for i in atlas cms dech dteam hone lhcb ops; do
    mkdir $i
    chmod 775 $i
    chown ${i}001.$i $i
    echo "StoreName $i" >  ./$i/".(tag)(OSMTemplate)"
    echo "$i" >  ./$i/".(tag)(sGroup)"
    echo "OUTPUT" > ./$i/".(tag)(RetentionPolicy)"
    echo "ONLINE" > ./$i/".(tag)(AccessLatency)"
    done
    
    ### And finally: 
    cd
    umount /mnt 
    

  • On Node01 we have to install the information system and link it to the dcache info script:
    yum-with-glite install glite-BDII
    ln -s /opt/d-cache/libexec/infoProvider/info-based-infoProvider.sh /opt/glite/etc/gip/provider/info-based-infoProvider.sh
    service bdii restart
    ldapsearch -x -H ldap://storage12.lcg.cscs.ch:2170 -b 'o=grid'
  • /opt/d-cache/install/install.sh to make the service ready to start.

Pool creation

  1. Using the dcache layouts templates, create the pools. Here is an example:
    # cat layouts/se14.conf
    [gridftp-se14Domain]
    [gridftp-se14Domain/gridftp]
    
    [data_phaseG_storage_10_array0_cms]
    [data_phaseG_storage_10_array0_cms/pool]
    name=data_phaseG_storage_10_array0_cms
    path=/data_phaseG_storage_10_array0/pool
    waitForFiles=${path}/data
    lfs=precious
    
    [data_phaseG_storage_10_array2_cms]
    [data_phaseG_storage_10_array2_cms/pool]
    name=data_phaseG_storage_10_array2_cms
    path=/data_phaseG_storage_10_array2/pool
    waitForFiles=${path}/data
    lfs=precious
    
    [data_phaseG_storage_10_array4_cms]
    [data_phaseG_storage_10_array4_cms/pool]
    name=data_phaseG_storage_10_array4_cms
    path=/data_phaseG_storage_10_array4/pool
    waitForFiles=${path}/data
    lfs=precious
    
Add pools, obviously replace $VO and $N with whatever is applicable as well as storage name. Double check this before adding.

#Once the layout files are added 

ssh storage01

ssh -2 admin@storage02 -i /usr/libexec/nagios-plugins/.ssh/ssh_host_dsa_key -p 22224

(local) admin > cd PoolManager
(PoolManager) admin > psu create pool  data_phaseH_storage_14_array_$N_$VO
(PoolManager) admin > psu addto pgroup $VO data_phaseH_storage_14_array_$N_$VO
(PoolManager) admin > save

  1. Once this is done, do not forget to assign each of these pools a maximum capacity ( set max diskspace) or dCache will not compute well the available space:
    # cat /data_phaseG_storage_5_array1/pool/setup
    #
    # Created by data_phaseG_storage_5_array1_atlas(Pool) at Tue Jan 15 11:00:29 CET 2013
    #
    csm set checksumtype ADLER32
    csm set policy -frequently=off
    csm set policy -onread=off -onwrite=on -onflush=off -onrestore=off -ontransfer=off -enforcecrc=on -getcrcfromhsm=off
    #
    # Flushing Thread setup
    #
    flush set max active 1000
    flush set interval 60
    flush set retry delay 60
    #
    # HsmStorageHandler2(org.dcache.pool.classic.HsmStorageHandler2)
    #
    rh set max active 0
    st set max active 0
    rm set max active 1
    rh set timeout 14400
    st set timeout 14400
    rm set timeout 14400
    jtm set timeout -queue=regular -lastAccess=0 -total=0
    jtm set timeout -queue=p2p -lastAccess=0 -total=0
    jtm set timeout -queue=wan -lastAccess=0 -total=0
    jtm set timeout -queue=io -lastAccess=0 -total=0
    set heartbeat 30
    set report remove on
    set breakeven 0.7
    set gap 4294967296
    set duplicate request none
    mover set max active -queue=regular 100
    mover set max active -queue=p2p 10
    mover set max active -queue=wan 10
    #
    # MigrationModule
    #
    #
    #  Pool to Pool (P2P) [$Revision: 1.61 $]
    #
    pp set port 0
    pp set max active 10
    pp set pnfs timeout 300
    set max diskspace 23992761057280
  2. NOTE: Keep in mind that there is one domain per dCache pool, which creates a java process per domain. This affects the available memory for each java process. For example, on a machine with 64GB of RAM and 6 pools, you would have this:
    # free -m
                 total       used       free     shared    buffers     cached
    Mem:         64402      59502       4899          0          3      50833
    -/+ buffers/cache:       8665      55736
    Swap:        16383          0      16383
    # df |grep data_phaseG
                         23435843584 17307245868 6128597716  74% /data_phaseG_storage_5_array0
                         23435843584 18500647744 4935195840  79% /data_phaseG_storage_5_array1
                         23435843584 18531993564 4903850020  80% /data_phaseG_storage_5_array2
                         23435843584 17422559744 6013283840  75% /data_phaseG_storage_5_array3
                         23435843584 17864687500 5571156084  77% /data_phaseG_storage_5_array4
                         23435843584 17771529276 5664314308  76% /data_phaseG_storage_5_array5
    # grep java.memory /opt/d-cache/etc/dcache.conf
    dcache.java.memory.heap=4096m
    dcache.java.memory.direct=4096m

Pool distribution

This is the current pool distribution across the dCache instance:
dCache pool distribution

Config files

Below is a litsing of common dcache configuration files and a brief overview of their function and syntax. For more details please the the full documentation on the dcache website http://www.dcache.org/manuals/index.shtml

Default values can be found under /usr/share/dcache/defaults some examples are also found in /usr/share/dcache/examples

dcache.conf

The main configuration file for dcache, you should not have to edit this unless you are making a major change

Example

chimera.db.user = postgres 

logback.xml

This file controls the default logging for dcache and is makes use of XML syntax. Note logging can be altered on the fly within the dcache cli so this may not reflect the current logging of a running system.

http://www.dcache.org/manuals/Book-2.6/config/cf-tss-monitor-fhs.shtml

    <threshold>
      <appender>pinboard</appender>
      <logger>root</logger>
      <level>info</level>
    </threshold>

LinkGroupAuthorization.conf

This file specifies the link group and who has permission to it.

http://www.dcache.org/manuals/Book-2.6/config/cf-write-token-spacereservation-fhs.shtml

TODO Example below is taken from production, need to clarify syntax

LinkGroup cms-linkGroup 
/cms
/cms/itcms
/cms/uscms
cmssgm
cmsprd
cms001

gplazma.conf

Controls what methods are valid for authentication. This file has a similar syntax to PAM files

http://www.dcache.org/manuals/Book-2.6/config/cf-gplazma-setup-fhs.shtml

Example

# step          modifier        plugin          params k=v
auth            optional        x509
map            requisite       authzdb        authzdb=/etc/grid-security/authzdb

layouts/*.conf

These files are host specific. In the case of storage01/ storage02 they control the dcache services that are run, for the SE machines they denote which pools are available.

Storage01 snippet

[srm-storage01Domain]
[srm-storage01Domain/srm]
srmJettyThreadsMax=1500
srmJettyThreadsMaxQueued=1500
srmCopyReqThreadPoolSize=2000
remoteGsiftpMaxTransfers=2000

SE snippet

[se01Domain_ibm01_data1_cms]
[se01Domain_ibm01_data1_cms/pool]
name=ibm01_data1_cms
path=/data1/cms
waitForFiles=${path}/setup
lfs=precious

Storage01 cells

Storage01 runs the following

  • dcap-storage01Domain
  • gridftp-storage01Domain
  • utilityDomain
  • httpdDomain
  • infoDomain
  • srm-storage01Domain
  • xrootd-storage01Domain (port 1094)
  • xrootd-CMS-Domain (port 1096)
  • xrootd-CMS-Domain2 (port 1095)
  • webdav-storage01Domain

Storage02 cells

  • dCacheDomain
  • gPlazma-storage02Domain running
  • namespaceDomain
  • dirDomain
  • adminDoorDomain
  • nfs-storage02Domain

OS notes

se[01-08] (mpp multipath module)

On a Storage Element (pool), NOT on a Head Node, after the new kernel has booted, we need to do the following:

  • Disconnect ONE of the FC cables BEFORE YOU BOOT so the OS sees only one path to the storage controllers.
  • recreate the mpp rdac files:
    # cd /root/linuxrdac-09.03.0C05.0331
    # make uninstall
    # make install
  • This should create a mpp initrd file on /boot/ for the current booted kernel:
    # ls -l /boot/mpp*
    -rw------- 1 root root 4075301 Jan 27  2011 /boot/mpp-2.6.18-128.29.1.el5.img
    -rw------- 1 root root 4165199 May 10 11:38 /boot/mpp-2.6.18-274.12.1.el5.img
  • Now you need to change the grub.conf file to boot next time using the mpp initrd created:
    # cat /boot/grub/grub.conf 
    # grub.conf generated by anaconda
    #
    # Note that you do not have to rerun grub after making changes to this file
    # NOTICE:  You have a /boot partition.  This means that
    #          all kernel and initrd paths are relative to /boot/, eg.
    #          root (hd0,0)
    #          kernel /vmlinuz-version ro root=/dev/vg_root/lv_root
    #          initrd /initrd-version.img
    #boot=/dev/sda
    default=0
    timeout=5
    splashimage=(hd0,0)/grub/splash.xpm.gz
    hiddenmenu
    title Scientific Linux SL (2.6.18-308.8.2.el5)
       root (hd0,0)
       kernel /vmlinuz-2.6.18-308.8.2.el5 ro root=/dev/vg_root/lv_root rhgb
       initrd /initrd-2.6.18-308.8.2.el5.img
    title Scientific Linux SL (2.6.18-274.12.1.el5)
       root (hd0,0)
       kernel /vmlinuz-2.6.18-274.12.1.el5 ro root=/dev/vg_root/lv_root rhgb
       initrd /mpp-2.6.18-274.12.1.el5.img
  • Reboot the machine and connect the FC cable previously disconnected.

se[09-22] (SL6x standard multipath module)

These machines are SL6 with standard multipath and infiniband packages from the distribution. This allows us to be able to update the kernel and or the system packages without worrying about compiling anything else. Just the standard yum update should work fine.

yum install device-mapper-multipath lssci sg3_utils
chkconfig multipathd on
cfagent -qv #to get multipath.conf
reboot

A number of modifications have been done to these machines in order to make operations as smooth as possible:

  • /etc/multipath.conf has been configured to name the arrays properly, based on the WWID of each of them:
     multipath {
                    wwid            360080e50002ddd4c0000053c50766586
                    alias           phaseG_storage_7_array0
            }
    for more details about how to find arrays' WWIDs on IBM storage systems please refer to this page
  • /etc/multipath.conf blacklists the machine's own RAID controller:
    blacklist {
     device {
       product "ServeRAID M5110e"
       }
    }
  • a LVM Physical Volume (PV) has to be created on each device associated with a logical array on the storage system:
        [root@se15:~]# for i in $(seq 1 5); do pvcreate /dev/mapper/phaseH_storage_11_array${i}; done
        [root@se16:~]# for i in $(seq 1 5); do pvcreate /dev/mapper/phaseH_storage_12_array${i}; done
        
    and then a LVM Volume Group (VG) must be created for each storage system belonging to a storage server and including all the respective PV created in the previous step (in the example Storage11 belongs to se15 and Storage16 belongs to se16:
        [root@se15:~]# vgcreate phaseH_storage_11 /dev/mapper/phaseH_storage_11_array*
        [root@se16:~]# vgcreate phaseH_storage_12 /dev/mapper/phaseH_storage_12_array*
        
    At this point a Logical Volume (LV) can be defined for each logical array in the VG corresponding to the storage system belonging to the server where the LVs are created from:
        [root@se15:~]# lvcreate --name phaseH_storage_11_array0 phaseH_storage_11 -L 21.828T
        [...]
        [root@se15:~]# lvcreate --name phaseH_storage_11_array5 phaseH_storage_11 -L 21.828T
        
    and
     
        [root@se16:~]# lvcreate --name phaseH_storage_12_array0 phaseH_storage_12 -L 21.828T
        [...]
        [root@se16:~]# lvcreate --name phaseH_storage_12_array5 phaseH_storage_12 -L 21.828T
        
    (information about the size to specify can be retrieved using pvdisplay)

  • As defined in the previous step, on each machine there are 2 LVM Volume Groups, the one "belonging" to the machine and the one of the failover one:
    # vgdisplay | grep 'phaseG_'
      VG Name               phaseG_storage_7
      VG Name               phaseG_storage_8
    Then, for each VG we have several LV, some active (host that owns) and some inactive (failover host). The activation of the LV on each host is done via /etc/lvm/lvm.conf, especifically the line volume_list = [ "vg_root", "phaseG_storage_7" ]. Look below for more information:
    # lvdisplay
    [...]
      --- Logical volume ---
      LV Path                /dev/phaseG_storage_7/phaseG_storage_7_array0
      LV Name                phaseG_storage_7_array0
      VG Name                phaseG_storage_7
      LV UUID                F4pLAN-2ogh-MHie-cmVg-GFq0-ymEz-NtivUU
      LV Write Access        read/write
      LV Creation host, time se11.lcg.cscs.ch, 2013-02-13 16:26:12 +0100
      LV Status              available
      # open                 1
      LV Size                21.83 TiB
      Current LE             5722080
      Segments               1
      Allocation             inherit
      Read ahead sectors     auto
      - currently set to     256
      Block device           253:14
    [...]
      --- Logical volume ---
      LV Path                /dev/phaseG_storage_8/phaseG_storage_8_array5
      LV Name                phaseG_storage_8_array5
      VG Name                phaseG_storage_8
      LV UUID                uZlUQt-43zL-r3DQ-t4rc-D03M-jOIi-q3jvou
      LV Write Access        read/write
      LV Creation host, time se12.lcg.cscs.ch, 2013-02-13 16:27:42 +0100
      LV Status              NOT available
      LV Size                21.83 TiB
      Current LE             5722080
      Segments               1
      Allocation             inherit
      Read ahead sectors     auto
    This behaviour can be changed modifying the following lines on /etc/lvm/lvm.conf under the section devices {
        obtain_device_list_from_udev = 1
        [...]
        volume_list = [ "vg_root", "phaseG_storage_7" ]
        [...]
        filter =[ "a|mapper|", "a|/dev/system_raid*|", "r|/dev/sd*|",  "r|.*|" ]
    
  • Overall this is a neat way of identifying what belongs where, but this extensive LVM modification has a problem: when installing a new kernel, during the rebuild of vmlinuz and initramfs images, dracut includes by default the file /etc/lvm/lvm.conf which in turn, disallows the booting kernel to use any device on the path /dev/sd* (rule "r|/dev/sd*|"). This results on the system not knowing where vg_root is and, therefore, throwing a kernel panic. The solution is very simple:
    1. Modify /etc/dracut.conf so it does not include any FC drivers or LVM configuration:
      # grep "lvmconf\|omit_drivers" /etc/dracut.conf
      omit_drivers+="qla2xxx qla3xxx qla4xxx"
      lvmconf="no"
    2. Rename the devices presented by the system's RAID controller so they do not match the remove /dev/sd* rule on /etc/lvm/lvm.conf. This is done to avoid LVM not knowing where vg_root even after running /sbin/init. Obviously, an udev rule needs to be written here:
      # cat /etc/udev/rules.d/10-local.rules
      # This is used to rename the block device sd?* of which parent has the attributes SUBSYSTEMS=="scsi", ATTRS{vendor}=="IBM", ATTRS{model}=="ServeRAID*" (system RAID)
      # to a new name: system_raid_%k (%k makes it sdy/sda, the original name). So in the end, we would have /dev/system_raid_sda[1-16]
      KERNEL=="sd?*", SUBSYSTEMS=="scsi", ATTRS{vendor}=="IBM", ATTRS{model}=="ServeRAID*",IMPORT{program}="/lib/udev/rename_device", NAME="system_raid_%k"
      As you can see, the devices presented to the OS by the system's RAID are renamed /dev/system_raid_sdXY. This allows the rule "a|/dev/system_raid*|", of /etc/lvm/lvm.conf to process the devices and, in turn, find the vg_root and the system /sbin/init
      # pvdisplay
          Logging initialised at Thu Feb 21 10:35:32 2013
          Set umask from 0022 to 0077
          Scanning for physical volume names
        --- Physical volume ---
        PV Name               /dev/system_raid_sdy2
        VG Name               vg_root
        PV Size               278.27 GiB / not usable 19.00 MiB
        Allocatable           yes (but full)
        PE Size               32.00 MiB
        Total PE              8904
        Free PE               0
        Allocated PE          8904
        PV UUID               pYEgl2-sS1w-LvDF-mY7D-mXJE-szwr-FivgTT
      [...]
      

  • After completing the LVM configuration, filesystems can be created on LVs using xfs:
        [root@se15:~]# mkfs.xfs /dev/phaseH_storage_11/phaseH_storage_11_array0 -b size=4096 -d su=128k,sw=10 -l version=2,su=128k -f
        [...]
        [root@se15:~]# mkfs.xfs /dev/phaseH_storage_11/phaseH_storage_11_array5 -b size=4096 -d su=128k,sw=10 -l version=2,su=128k -f
    
        [root@se16:~]# mkfs.xfs /dev/phaseH_storage_12/phaseH_storage_12_array0 -b size=4096 -d su=128k,sw=10 -l version=2,su=128k -f
        [...]
        [root@se16:~]# mkfs.xfs /dev/phaseH_storage_12/phaseH_storage_12_array5 -b size=4096 -d su=128k,sw=10 -l version=2,su=128k -f
        
  • and eventually mount points can be created and fstab edited to implement a manual failover:
        [root@se15:~]# mkdir /data_phaseH_storage_11_array{0,1,2,3,4,5}
        [root@se16:~]# mkdir /data_phaseH_storage_12_array{0,1,2,3,4,5}
    
        [root@se15:~]# vim /etc/fstab
        [...]
        /dev/phaseH_storage_11/phaseH_storage_11_array0           /data_phaseH_storage_11_array0           xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier      0 1
        [...]
        /dev/phaseH_storage_11/phaseH_storage_11_array5           /data_phaseH_storage_11_array5           xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier      0 1
        #/dev/phaseH_storage_12/phaseH_storage_12_array0           /data_phaseH_storage_12_array0           xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier   0 1
        [...]
        #/dev/phaseH_storage_12/phaseH_storage_12_array5           /data_phaseH_storage_12_array5           xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier   0 1
    
        [root@se16:~]# vim /etc/fstab
        [...]
        #/dev/phaseH_storage_11/phaseH_storage_11_array0           /data_phaseH_storage_11_array0           xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier      0 1
        [...]
        #/dev/phaseH_storage_11/phaseH_storage_11_array5           /data_phaseH_storage_11_array5           xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier      0 1
        /dev/phaseH_storage_12/phaseH_storage_12_array0           /data_phaseH_storage_12_array0          xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier      0 1
        [...]
        /dev/phaseH_storage_12/phaseH_storage_12_array5           /data_phaseH_storage_12_array5          xfs           largeio,inode64,swalloc,noatime,nodiratime,logbufs=8,allocsize=131072k,nobarrier      0 1
        
    please note how the first 6 lines are de-commented in se15 's fstab while the last 6 ones are commented, and vice-versa in se16 's fstab since Stoarge11 belongs to se15 and Storage12 belongs to se16. Now a mount -a should mount all the LVs on the respective mount points and a df should show the available space. At this point the pools' configuration can be performed.

Install notes for se[17-18]

Because these two nodes share only one pair of controllers (and two enclosures), the configuration is slightly different in the sense that there are two VG, with PV and LV assigned as follows:

  • PV: all on both machines ( /dev/mapper/phaseH_storage_r6_1_array[0-11] ).
    # for i in $(seq 0 11); do [[ -e /dev/mapper/phaseH_storage_r6_1_array${i} ]] && echo pvcreate /dev/mapper/phaseH_storage_r6_1_array${i}; done
  • Volume Groups:
    • VolumeGroup on se17 phaseH_storage_r6_1_17 (array 0-5)
      [root@se17:~]# vgcreate phaseH_storage_r6_1_17 /dev/mapper/phaseH_storage_r6_1_array0 /dev/mapper/phaseH_storage_r6_1_array1 /dev/mapper/phaseH_storage_r6_1_array2 /dev/mapper/phaseH_storage_r6_1_array3 /dev/mapper/phaseH_storage_r6_1_array4 /dev/mapper/phaseH_storage_r6_1_array5
    • VolumeGroup on se18 phaseH_storage_r6_1_18 (array 6-11)
      [root@se18:~]# vgcreate phaseH_storage_r6_1_18 //dev/mapper/phaseH_storage_r6_1_array6 /dev/mapper/phaseH_storage_r6_1_array7 /dev/mapper/phaseH_storage_r6_1_array8 /dev/mapper/phaseH_storage_r6_1_array9 /dev/mapper/phaseH_storage_r6_1_array10 /dev/mapper/phaseH_storage_r6_1_array11
  • Logical Volumes:
    [root@se17:~]# for i in $(seq 0 5); do [[ -e /dev/mapper/phaseH_storage_r6_1_array${i} ]] &&  lvcreate --name phaseH_storage_r6_1_17_array${i} --extents 100%PVS phaseH_storage_r6_1_17 /dev/mapper/phaseH_storage_r6_1_array${i}; done
    [root@se18:~]#  for i in $(seq 6 11); do [[ -e /dev/mapper/phaseH_storage_r6_1_array${i} ]] &&  lvcreate --name phaseH_storage_r6_1_18_array${i} --extents 100%PVS phaseH_storage_r6_1_18 /dev/mapper/phaseH_storage_r6_1_array${i}; done
  • Filesystems:
    [root@se17:~]# for i in $(ls /dev/mapper/ |grep phaseH_storage_r6_1_17|awk -F'-' '{print $2}'); do mkfs.xfs /dev/phaseH_storage_r6_1_17/${i} -b size=4096 -d su=128k,sw=10 -l version=2,su=128k -f; mapperdevice=$(echo ${i} | sed 's/_17//g'); echo "/dev/phaseH_storage_r6_1_17/${i}      /${mapperdevice}         xfs      defaults 0 0" >> /etc/fstab ; mkdir -p /${mapperdevice};done
    [root@se18:~]# for i in $(ls /dev/mapper/ |grep phaseH_storage_r6_1_18|awk -F'-' '{print $2}'); do mkfs.xfs /dev/phaseH_storage_r6_1_18/${i} -b size=4096 -d su=128k,sw=10 -l version=2,su=128k -f; mapperdevice=$(echo ${i} | sed 's/_18//g'); echo "/dev/phaseH_storage_r6_1_18/${i}      /${mapperdevice}         xfs      defaults 0 0" >> /etc/fstab ; mkdir -p /${mapperdevice};done

Install notes for se[19-22]

These nodes share only one set of controllers (and 3 enclosures) per pair

  • se19/20: PV all / VG phaseJ_storage_r2_1_se19 and phaseJ_storage_r2_1_se20 / LV odd PVs active (1 3 5 7 9 11 13 15 17) on se19 and even PVs active (0 2 4 6 8 10 12 14 16) on se20. Space available is 270TiB per machine after RAID, partitioning and formatting (XFS). Total 540TiB per pair.
  • se21/22: PV all / VG phaseJ_storage_r1_1_se21 and phaseJ_storage_r1_1_se22 / LV odd PVs active (1 3 5 7 9 11 13 15 17) on se21 and even PVs active (0 2 4 6 8 10 12 14 16) on se22. Space available is 270TiB per machine after RAID, partitioning and formatting (XFS). Total 540TiB per pair.
This is how the LVM and filesystem are distributed and created:

  • Physical Volumes are shared across pairs ( /dev/mapper/phase*array* ): Run this on all 4 servers:
    # for i in $(ls /dev/mapper/phase*array* |grep -v -E 'p1|p2|p3'); do [[ -e $i ]] && pvcreate $i; done
  • Volume Groups are shared across pairs:
    • VolumeGroup on se19 phaseJ_storage_r2_1_se19
      # vgcreate phaseJ_storage_r2_1_se19 /dev/mapper/phaseJ_storage_r2_1_array17 /dev/mapper/phaseJ_storage_r2_1_array15 /dev/mapper/phaseJ_storage_r2_1_array13 /dev/mapper/phaseJ_storage_r2_1_array11 /dev/mapper/phaseJ_storage_r2_1_array9 /dev/mapper/phaseJ_storage_r2_1_array7 /dev/mapper/phaseJ_storage_r2_1_array5 /dev/mapper/phaseJ_storage_r2_1_array3 /dev/mapper/phaseJ_storage_r2_1_array1
    • VolumeGroup on se20 phaseJ_storage_r2_1_se20
      # vgcreate phaseJ_storage_r2_1_se20 /dev/mapper/phaseJ_storage_r2_1_array16 /dev/mapper/phaseJ_storage_r2_1_array14 /dev/mapper/phaseJ_storage_r2_1_array12 /dev/mapper/phaseJ_storage_r2_1_array10 /dev/mapper/phaseJ_storage_r2_1_array8 /dev/mapper/phaseJ_storage_r2_1_array6 /dev/mapper/phaseJ_storage_r2_1_array4 /dev/mapper/phaseJ_storage_r2_1_array2 /dev/mapper/phaseJ_storage_r2_1_array0
      * VolumeGroup on se21 phaseJ_storage_r1_1_se21
      # vgcreate phaseJ_storage_r1_1_se21 /dev/mapper/phaseJ_storage_r1_1_array17 /dev/mapper/phaseJ_storage_r1_1_array15 /dev/mapper/phaseJ_storage_r1_1_array13 /dev/mapper/phaseJ_storage_r1_1_array11 /dev/mapper/phaseJ_storage_r1_1_array9 /dev/mapper/phaseJ_storage_r1_1_array7 /dev/mapper/phaseJ_storage_r1_1_array5 /dev/mapper/phaseJ_storage_r1_1_array3 /dev/mapper/phaseJ_storage_r1_1_array1
      * VolumeGroup on se22 phaseJ_storage_r1_1_se22
      # vgcreate phaseJ_storage_r1_1_se22 /dev/mapper/phaseJ_storage_r1_1_array16 /dev/mapper/phaseJ_storage_r1_1_array14 /dev/mapper/phaseJ_storage_r1_1_array12 /dev/mapper/phaseJ_storage_r1_1_array10 /dev/mapper/phaseJ_storage_r1_1_array8 /dev/mapper/phaseJ_storage_r1_1_array6 /dev/mapper/phaseJ_storage_r1_1_array4 /dev/mapper/phaseJ_storage_r1_1_array2 /dev/mapper/phaseJ_storage_r1_1_array0
  • Logical Volumes are shared across pairs, but only active on one of the two systems of the pair at the same time. For this to work make sure the entry volume_list in lvm.conf has the correct version of [ "vg_root","phaseJ_storage_r2_1_se19" ]. Run this on all 4 servers:
    # for id in $(pvdisplay 2>/dev/null |grep phaseJ_storage_r2_1_$(hostname -s) -B 1 |grep 'PV Name' | awk {'print $3}'); do lvcreate --name $(basename $id) --extents 100%PVS phaseJ_storage_r2_1_$(hostname -s) $id; done
  • Filesystems. Run the following on all 4 servers:
    # for device in $(lvdisplay 2>/dev/null |grep $(hostname -s) |grep 'LV Path' | awk '{print $3}'); do mkfs.xfs $device -b size=4096 -d su=128k,sw=10 -l version=2,su=128k -f ;echo "$device     /data_$(basename $device)    xfs   defaults 0 0" >> /etc/fstab; mkdir /data_$(basename $device); done

Tips and tricks for PreProduction

Some of these solutions can be checked in http://trac.dcache.org/wiki/TroubleShooting

  • If you get this message when trying to send files to the SE with error Best pool too high : Infinity, you need to change the gap in each pool:
    set gap 10m
    The reason why this happens is that the default gap is set by default to 4G, and if you are using pools with less free space than that (very likely in preproduction), transfers will never reach the pools. More info here (release 1.9.5): http://www.dcache.org/manuals/Book-1.9.5/config/cf-pm-cm.shtml#cf-pm-cm-space = set gap /size[] # unit = k|m|g=

  • If you get this message when starting the dcache services [JGLOBUS-8] Failed to load credentials. [Caused by: [JGLOBUS-22] Algorithm not supported.], you must check if the hostkey.pem file is RSA. The following fixes it:
    mv /etc/grid-security/hostkey.pem /etc/grid-security/hostkey-pk8.pem
    openssl rsa -in /etc/grid-security/hostkey-pk8.pem -out /etc/grid-security/hostkey.pem

  • se[-9-14] FC connections:
  • PhaseG_DCS3700_connections.png

HP FC connections

The physical layout of the FC ports on the HP machines goes from left to right, example image below. If in doubt use 'systool -c fc_host -v' to verify.

hp_fc.png

The connections are made as follows;

In future expansion additional controllers can be connected to port 1 of the cards on the SE machines, connecting to both controllers on ports 0 and 2 in the same fashion.

  • SE21

FC Host   | Controller
----------|------------
C0P0      | C0P0
C1P0      | C1P2

  • SE 22

FC Host   | Controller
----------|------------
C0P0      | C1P0
C1P0      | C0P2

DBs backups in /var/lib/pgsql/

Show... Hide

Dec 02 17:33 [root@storage02:store]# dcache database ls
DOMAIN              CELL             DATABASE HOST      USER     MIN- MAX-CONNS MANAGEABLE AUTO 
namespaceDomain     PnfsManager      chimera  localhost postgres 30   90        Yes        Yes  
namespaceDomain     cleaner          chimera  localhost postgres                No         No   
nfs-storage02Domain NFSv41-storage02 chimera  localhost postgres                No         No   
TOTAL                                                            30   90                        
Dec 02 17:33 [root@storage02:store]# exit
logout
Connection to storage02 closed.

Dec 02 17:34 [fmartinelli@pub:~]$ ssh root@storage01 dcache database ls
DOMAIN              CELL                  DATABASE HOST      USER      MIN- MAX-CONNS MANAGEABLE AUTO 
utilityDomain       PinManager            dcache   localhost srmdcache                Yes        Yes  
srm-storage01Domain SRM-storage01         dcache   localhost srmdcache                No         Yes  
srm-storage01Domain SrmSpaceManager       dcache   localhost srmdcache                No         Yes  
srm-storage01Domain RemoteTransferManager dcache   localhost srmdcache                No         Yes  
TOTAL                                                                  0    0     

Dec  2 05:16:50 storage02 backup[31936]: backup 3 starting at Mon Dec  2 05:16:50 CET 2013
Dec  2 05:16:50 storage02 backup[31936]:   Archival destination: archive=lcgarc@ela.cscs.ch:/store/system/Phoenix/
Dec  2 05:16:50 storage02 backup[31936]:   Base directory: basedir=/opt/backup
Dec  2 05:16:50 storage02 backup[31936]:   Log files directory: logdir=/opt/backup/log
Dec  2 05:16:50 storage02 backup[31936]:   Incremental backup state directory: statedir=/opt/backup/state
Dec  2 05:16:50 storage02 backup[31936]:   Tar command: tar=/bin/tar
Dec  2 05:16:50 storage02 backup[31936]:   Additional tar options: tar_options=
Dec  2 05:16:50 storage02 backup[31936]:   SSH command: ssh=/usr/bin/ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no
Dec  2 05:16:50 storage02 backup[31936]:   SSH identity key: /opt/backup/etc/id_rsa_backup
Dec  2 05:16:50 storage02 backup[31936]: Creating backup directory '/store/system/Phoenix//2013/12/02' ...
Dec  2 05:16:50 storage02 backup[31936]: Warning: Permanently added 'ela.cscs.ch,148.187.1.6' (RSA) to the list of known hosts.#015
Dec  2 05:16:51 storage02 backup[31936]: Would backup 'storage02:/var/lib/pgsql/dumpall-c_complete.sql' to remote file 'storage02_var_lib_pgsql_dumpall-c_complete.sql.tar'.
Dec  2 05:16:51 storage02 backup[31936]: Running a full backup (requested by '-f' command-line option).
Dec  2 05:16:51 storage02 backup[31936]: Warning: Permanently added 'ela.cscs.ch,148.187.1.6' (RSA) to the list of known hosts.#015
Dec  2 05:17:13 storage02 backup[31936]: Warning: Permanently added 'ela.cscs.ch,148.187.1.6' (RSA) to the list of known hosts.#015
Dec  2 05:17:38 storage02 backup[31936]: Warning: Permanently added 'ela.cscs.ch,148.187.1.6' (RSA) to the list of known hosts.#015
Dec  2 05:17:55 storage02 backup[31936]: Warning: Permanently added 'ela.cscs.ch,148.187.1.6' (RSA) to the list of known hosts.#015
Dec  2 05:18:19 storage02 backup[31936]: Warning: Permanently added 'ela.cscs.ch,148.187.1.6' (RSA) to the list of known hosts.#015
Dec  2 05:18:45 storage02 backup[31936]: Backup of 'storage02:/var/lib/pgsql/dumpall-c_complete.sql' finished (with exitcode 0)
Dec  2 05:18:45 storage02 backup[31936]: backup done at Mon Dec  2 05:18:45 CET 2013

Materialized views

The storage02 machine has Fabio's materialized views installed (https://bitbucket.org/fabio79ch/v_pnfs/wiki/Home). This provides a very usefull table of pnfsid's, paths, access times and creation times.

Refresh the view

The view is generated from a query so the data contained within it drifts as time goes on. The view can be refreshed with the following, this took around 20 minutes when the system was under minimal load. On a a test system this took 45 minutes.

su - postgres
psql

  psql> \c chimera
  psql> REFRESH MATERIALIZED VIEW v_pnfs;

Space monitor

  • Every night there is a script that runs ( storage02:/etc/cron.daily/run_space_monitor2.sh ) and generates a few pages on Ganglia that show the internal space utilisation per folder for each of the three VOs:
    1. ATLAS
    2. CMS
    3. LHCb
  • Please note that this script *refreshes the materialised views( as required!

Issues

Can't remove file from NFS

This was due to root squashing in the NFS export options of storage02. I was able to change the options in /etc/exports and restart in NFS domain in production on the fly without issue. Would not recommend doing this outside of a downtime unless you really need to.

lcg-getturl returns file:/pnfs/.... on WN

This was due to the nfs4.1 service offering files.

Note the "loginBroker=srm-LoginBroker" does not mean this is broker used by the SRM but the login broker the SRM itself registers with.

#dCache CLI

cd LoginBroker
disbale NFSv41-storage02@nfs-storage02Domain

#To fix permanently change the login broker

vim /etc/dcache/layouts/storage01.conf
  [nfs-${host.name}Domain]
  [nfs-${host.name}Domain/nfsv41]
  loginBroker=srm-LoginBroker

Before and after

lcg-getturls -p file,gsiftp srm://storage01.lcg.cscs.ch/pnfs/lcg.cscs.ch/dteam/test1244                                                                                                                    
file:/pnfs/lcg.cscs.ch/dteam/test1244

lcg-getturls -p file,gsiftp srm://storage01.lcg.cscs.ch/pnfs/lcg.cscs.ch/dteam/test1244                                                                                                                  
gsiftp://se10.lcg.cscs.ch:2811/pnfs/lcg.cscs.ch/dteam/test1244

This caused a temporary issue where some CMS failed as the above change removed the following information published from the BDII. As such this change took over 12 hours to propagate to the top BDII.

#The following entries were removed the published information

NFSv41-storage02@nfs-storage02Domain, storage01.lcg.cscs.ch, CSCS-LCG2, local, grid
dn: GlueSEAccessProtocolLocalID=NFSv41-storage02@nfs-storage02Domain,GlueSEUni
queID=storage01.lcg.cscs.ch,Mds-Vo-name=CSCS-LCG2,Mds-Vo-name=local,o=grid
GlueSEAccessProtocolLocalID: NFSv41-storage02@nfs-storage02Domain
: file transfer
objectClass: GlueSETop
objectClass: GlueSEAccessProtocol
objectClass: GlueKey
objectClass: GlueSchemaVersion
GlueSEAccessProtocolMaxStreams: 5
GlueSchemaVersionMinor: 3
GlueSEAccessProtocolVersion: file
GlueSEAccessProtocolEndpoint: file://storage02.lcg.cscs.ch:2049
GlueChunkKey: GlueSEUniqueID=storage01.lcg.cscs.ch
GlueSchemaVersionMajor: 1
GlueSEAccessProtocolType: file

Unable to write to NFS mount

If you get IO errors when attempting to write to the NFS mount of /pnfs or doing transfers and see the following in the logs check the pool you are writing to isn't low on free space. Check the pool logs for the "Too high infinity crap blah blah blah" message.

May 12 09:34:32 storage02 kernel: NFS: v4 server returned a bad sequence-id error on an unconfirmed sequence ffff880037059a68!
May 12 09:34:32 storage02 kernel: NFS: v4 server storage02  returned a bad sequence-id error!

Changing movers on the fly

The settings can be for the number of movers can be changed on the fly via the dCache ssh interface. This is usefull if the setup file was not added to the root of the pool in the initial install for example.

# cd to the pool you wish to edit

cd data_phaseG_storage_7_array0_cms

# View current settings

mover queue ls -l
regular 0 100 0
p2p 0 10 0
wan 0 2 0

# Change the settings

mover set max active 10 -queue=wan
save

# Confirm change

mover queue ls -l
regular 0 100 0
p2p 0 10 0
wan 0 10 0
ServiceCardForm
Service name dCache
Machines this service is installed in storage[01-02],se[01-18]
Is Grid service Yes
Depends on the following services -
Expert Miguel Gila
CM CfEngine
Provisioning PuppetForeman
Topic attachments
I Attachment History Action Size Date Who Comment
PNGpng PhaseG_DCS3700_connections.png r1 manage 33.4 K 2013-02-21 - 10:42 MiguelGila se[-9-14] FC connections
PNGpng dCache_pool_distribution_1.1.png r2 r1 manage 138.4 K 2015-05-07 - 08:11 MiguelGila dCache Pool distribution
PNGpng hp_fc.png r1 manage 413.3 K 2014-11-06 - 15:24 GeorgeBrown  
Edit | Attach | Watch | Print version | History: r61 < r60 < r59 < r58 < r57 | Backlinks | Raw View | Raw edit | More topic actions
Topic revision: r61 - 2015-07-31 - MiguelGila
 
This site is powered by the TWiki collaboration platform Powered by Perl This site is powered by the TWiki collaboration platformCopyright © 2008-2020 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback