Official How-To

Official How-To

Prepare Chimera

We need a working pristine chimera installation in the target host. Chech ServiceDcache to know how to (everything but don't populate the VOs dirs).

Obtain the dump file with all pnfs information

For this you need to know the:

  • pnfsid of the source dir: In the source host look at /pnfs/fs/admin/etc/config/serverRoot and make sure it's the same as cat /pnfs/fs/'.(id)(usr)'
  • pnfsid of the target dir: In the target host perform cat /mnt/pnfs/'.(id)(lcg.cscs.ch)'
Then we need to install the latest pnfs-dump rpm and perform in the source:
time /opt/pnfs/tools/pnfsDump -r <pnfsid of the source dir> -vv -d0 -o chimera /tmp/pnfs.sql -2 -p <pnfsid of the target dir> -o verify /tmp/verify.md5 -r -o files /tmp/files.lst -f

###You can expect something like this:
Final stats:

inodes:
  nDir:       137024
  nFile:      577029
  nUnknown:        0
  nSkipped:        0
  ----------------- 
  Total:      714053

Major DB errors: 0

Cache queries: 13101440 (hits: 6550720, misses: 0)

Time elapsed:  395.9s (00:06:35.9)
    dbserver:  297.1s (00:04:57.1)
    overhead:    4.9s (00:00:04.9)
    pnfsDump:   93.9s (00:01:33.9)

Average inode processing rate: 1808 Hz

real    6m35.866s
user    1m39.162s
sys     0m30.455s

And we copy them to the target host:

scp /tmp/verify.md5 /tmp/pnfs.sql /tmp/files.lst root@storage22:

And also an initialization SQL script from pnfs-dump RPM:

scp /opt/pnfs/share/sql/prep-chimera-for-migration.sql root@storage22:

Import pnfs dumps into Chimera

  • First ensure chimera is unmounted and chimera-nfs service is shut down
  • Prepare chimera for migration (with the script copied from last step)
  • Then dump the metadara and check the output file doesn't contain errors.
  • And finally start the service and mount chimera. Move around the filesystem, should look like the old pnfs.
    umount /mnt
    service chimera-nfs stop
    psql -U postgres -f /root/prep-chimera-for-migration.sql chimera
    time psql -U postgres -f /root/pnfs.sql chimera > pnfs.sql.injection.stdout
    service chimera-nfs start
    mount localhost:/ /mnt 

Check everything worked fine

The first test is the MD5sum. In the Chimera node, with the filesystem mounted in /mnt as usual, perform

cd /mnt/pnfs/lcg.cscs.ch/
time md5sum -c /root/verify.md5 | tee /root/verify.md5.result | grep -v :\ OK$

The second check (StorageInfo) needs to run on the chimera node with chimera unmounted and pnfs mounted:

umount /mnt
mount #and check chimera is not mounted anywhere else
service chimera-nfs stop
mkdir -p /pnfs/fs
cd /pnfs
ln -s /pnfs/fs ftpBase
ln -s fs/usr lcg.cscs.ch
mount storage02.lcg.cscs.ch:/fs /pnfs/fs ## If you can't, probably need to add something to /pnfs/fs/admin/etc/exports pn the pnfs host
time sh /opt/d-cache/libexec/migration-check.sh -k /root/files.lst 2> /root/files.lst.errors > /root/files.lst.stdout

Timing

You have to wait for certain commands to finish. This is an approximation of how much you can leave for a coffee (for almost 600k files):

  • Dump pnfs information: 7 minutes
  • Import metadata into chimera: 80 minutes
  • md5sum check: 75 minutes
  • StorageInfo check: 85 minutes

Issues

We've discovered during the procedure that some directories lack the RetentionPolicy and AccessLatency tags that their parents have. This is the list:

hone/ppraxmar-srmmkdir-test-remove
hone/generated
cms/local_tests
cms/local
cms/trivcat
cms/generated
atlas/borgeg
atlas/instal
atlas/generated
atlas/SAM
ops/generated
dteam/CSCS
dteam/generated

-- PabloFernandez - 2010-03-11


This topic: LCGTier2 > WebHome > ServiceInformation > ServiceCards > ServiceDcache > ChimeraMigration
Topic revision: r10 - 2012-07-18 - FabioMartinelli
 
This site is powered by the TWiki collaboration platform Powered by Perl This site is powered by the TWiki collaboration platformCopyright © 2008-2024 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback