Currently the gpfs xrootd space is handled by a single xrootd server that connects directly to the redirectors. The list below shows how to move the server from one gpfs machine to another one without causing clients to fail (there will be a short stalling of the clients).

Before Moving

  1. remove taylor restart for xroots,cmsd  on <old>  (in /etc/taylor.opts)
  2. update crontab on <new>
  3. add <new> to xrootd.cf to distribute proper xrootd file
  4. backup xrootd.cf on <old>  backup/xrootd.cf
  5.  edit (in place) xrootd.cf on <old> add:
    export /fgstdummy/ if cmsd

Restart

  1. activate xrootd on <new>
    This will set the prod link and install the xrootd.cf file
  2.  Stop cmsd on <old>
    RDR will suspend writes and reads for files not found. Xrootd will continue to run and clients can read/write.
  3. Start xrootd/cmsd on <new>
    files on gpfs will be accessible

  4.  Start cmsd on <old>
    RDR will reconfigure its map for <old> and export the  /fgstdummy/ path. All previous entries from <old> are removed and there will be no more redirection to <old>.

  5.  If no more clients connected to <old> stop xrootd/cmsd
  6. rm prod link on <old>
     prod -> vN.N.N
  7. let Charlotte know that Nagios needs to be changed
  8.  change glast,glastall host lists

After Restart

  1.  enable restart in taylor on <new>
  2. make sure the rmdir semaphore file exists (/gpfs/slac/fermi/xrootd/dirlocks/fn)
  3. check that pftp is installed and working

  • No labels