Time-stamp: <2004-12-01 21:26:55 gokieli>

Hints & Tips for the Use of

CASTOR in DELPHI

What is CASTOR?

CASTOR is an implementation of a Managed Storage system, developed at CERN, which allows users to store and retrieve large files (please do not use it for files below ~20MB!). It presents you with a file name space, which means you can store and manage your own files. All DELPHI users have their own CASTOR home directory, with an associated environment variable

$CASTOR_HOME = /castor/cern.ch/user/u/username

There is a pool of tapes behind this name space, to which your files are copied. All the administrative details of tape names, file sequences etc. are dealt with by the CASTOR software.

Basic usage

In order to store your file in CASTOR, use the command rfcp:

      rfcp myfile.dst $CASTOR_HOME/dst/file.1.dst

The usage of CASTOR files is normally very simple. Many CERN applications, in particular CERNLIB (thus PAW, HBOOK etc.) and also our PHDST program can read CASTOR files directly, so that you can use them - e.g.
- in PAW:
```
h/fil 1 $CASTOR_HOME/file.1.paw
```
- in the PDLINPUT file for PHDST:
```
FILE=$CASTOR_HOME/file.1.dst
```
For applications that are not (yet...) CASTOR-aware, you again use rfcp and create a local file:
```
      rfcp $CASTOR_HOME/archive/myfiles.tar recov.tar
    
```

Main CASTOR commands

CASTOR comes with its own set of commands ns*, with their man pages. A (non-exhaustive) list of commands is presented here:

nsls: - lists the contents of your CASTOR home directory. Option -l gives a long listing (like standard ls, also some other ls options are available - although not all).
nsrm: - removes a CASTOR file. Options -R will remove a directory and everything below it.
nsmkdir: - creates a subdirectory
nsrename: - changes the file name
nschmod: - changes mode bits of file/directory. Useful if you share a directory with your colleagues.
Good old rfcp transfers a file to CASTOR - but make sure that the file has Unix read permission for "others"! In case of doubt precede this command by chmod a+r myfile.

More information can be found on IT Castor pages. In particular you may want to see the User Guide.

Some simple tricks

Group files

If you want to give write permission for your CASTOR-files to other DELPHI people, you should create an CASTOR subdirectory with the -m option:

  nsmkdir -m 775 $CASTOR_HOME/groupfiles

All DELPHI people will be able to create (and delete!) files in this directory.
For more details: man nsmkdir.

Removing complete sub-directories

The support for wild-characters in Castor is on the list of things to come, but the time scale is yet unknown. So whenever you want to remove a complete sub-directory from your Castor structure, the easiest way is to use the yes command:

  yes | nsrm $CASTOR_HOME/subdir

Emulating the wild-characters usage

If you want to list or delete only files of a specific kind, the lack of wild characters is sometimes sorely felt. The easiest way around is to use the combination of foreach and grep commands. Suppose you want to remove all your files with extension hbk - something you would like to do with a command
nsrm $CASTOR_HOME/subdir/*.hbk:

  foreach i ( `nsls $CASTOR_HOME/subdir | grep '[.]hbk$'` )
  nsrm $CASTOR_HOME/subdir/$i
  end

(please note two kinds of quotes being used).
In a similar manner some very complicated selections can be imposed.
If you never used foreach: you type first line as is, and after a brief moment the machine will prompt you for the following lines. After typing end and its carriage-return sign, the execution starts immediately. It may take quite a while, if there are many files.

Copying CASTOR-files to/from non-CERN machines

For not too big amounts of data, and specially for files from your CASTOR home directory (where no other method is possible), the simplest way is to use temporary files in the /tmp directory on a CERN machine and the scp command locally on your home machine. If you use the public cluster at CERN (like LXPLUS), you will have to remember on which particular machine you stored your file!
- Login to the CERN machine and store your file in /tmp (we will suppose that in the first step below you connected to lxplus013)
  - ssh lxplus.cern.ch
  - rfcp ${CASTOR_HOME}/myfile /tmp/myfile
- Once your files has been read from CASTOR, in another window on your home machine initiate the transfer
  - scp -pv myuserid@lxplus013.cern.ch:/tmp/myfile /local/myfile
- When the transfer finished, on the CERN machine delete the temporary file and possibly do the same operation with the next file to be transferred.
  - rm /tmp/myfile
  - logout
- Please note that:
  
  You should rather avoid using AFS (like $SCRATCH or $SCRATCH_WEEK areas) for temporary storage - whenever possible, use a local file (like /tmp): AFS often has very serious problems with handling big files, and this can even lead to the AFS server crash!
  
  You must take care not to fill the /tmp area completely, because this may very likely crash your host, or else your files may be removed by the system clean-up: after ~50% of /tmp being used (command df -h /tmp), transfer the files and remove them from /tmp.
  
  In the scp command above you may use the wild-characters for the remote files, but they must be enclosed in double-quotes:
  
  scp -pv myuserid@lxplus013.cern.ch:/tmp/myfile"*" /localdir/
  
  The rfcp command does not accept any wild-characters (see however the previous chapter on a trick that emulates them).

For big amounts of data from /castor/cern.ch/delphi/ CASTOR directory, you will probably have to use the new service, GridFTP. Please see the IT prepared documentation for details. Please note that for the authorisation for transferring the DELPHI data you should write directly to CERN (see the list of bodies issuing the certificates), and not to your country representative (which will give the authorisation for the LHC-connected activities).

Some known problems
(should disappear in the future CASTOR versions)

At present CASTOR files (and specially those newly created!) should have world read permission (022 for user's mask) in order to be correctly handled by the CASTOR (and stager) software. This limitation will be removed soon, but today it is imperative to comply. Anyway, with your home directory being on AFS, there is reason to have such protection at all! Please do not change the default value!

Do not use the shortcuts like ${CASTOR_HOME}/../../p/pubxx when you want to stage other people's CASTOR files. Resolving this kind of names confuses the stager!

Please note that CASTOR is a complex system, developing very rapidly. The recommendations made on this page reflect our knowledge, but it may well be that users will run into problems that we have not realized. Please give us feedback!

As always, questions, complaints, remarks:
delphi-core@cern.ch Jan van Eldik, Dec 6,2000
R.Gokieli, Nov 2, 2001