Overview

The Dataset Stager (afdsmgrd) is a daemon that coordinates the transfer of data from a remote storage to your local storage.

For each file to transfer, a script is called. The script can be customized to support your source and destination protocol.

Staging requests are issued from the ROOT console, where you can also control the progress of your staging. How to request stagings and how to check current transfer progress from ROOT is explained in the PROOF interface to AliEn file catalog documentation.

Installation

The Dataset Stager is distributed both on a repository on its own and as part of ROOT. The easiest way to compile it is to do it inside ROOT.

Installing from ROOT

When configuring the ROOT source, enable the Dataset Stager by adding --enable-afdsmgrd. Check in the list of enabled features if you have "afdsmgrd".

After running make (and, optionally, make install) you'll find the daemon in the same directory of root.exe.

The configuration file and init.d startup script will be in $ROOTSYS/etc/proof. The daemon can and must run as unprivileged user.

Configuration

The Dataset Stager can share its configuration file with PROOF, as some directives are the same and unknown directives are just ignored.

Directives are one per line and lines beginning with a pound sign (#) are used for comments.

The configuration file is automatically checked at each loop: this means you can change configuration without restarting the daemon or stopping your current transfers.

A detailed description of each directive follows.

set VARIABLE=value : This statement will substitute every occurrence of $VARIABLE with its value in the rest of the configuration file. You can have multiple set statements.

xpd.stagereqrepo [dir:]*directory* : This directive is shared with PROOF: directory is the full path to the dataset repository. Defaults to empty: without this directive the daemon is not operative.

The dir: prefix is optional.

dsmgrd.purgenoopds true|false : Set it to true **(default is false)** to remove a dataset when no file to stage is found. If no file to stage is found, but corrupted files exist, the dataset is kept to signal failures. Used in combination with xpd.stagereqrepo makes it "disposable": only the datasets effectively needed for signaling the staging status will be kept, improving scalability and stability.

dsmgrd.urlregex regex subst : Each source URL present in the datasets will be matched to regex and substituted to subst. regex supports grouping using parentheses, and groups can be referenced in order using the dollar sign with a number ($1 for instance) in subst.

Matching and substitution for multiple URL schemas are supported by using in addition directives dsmgrd.urlregex2 up to dsmgrd.urlregex4 which have the same syntax of this one.

Example of URL translation via regexp:

> - Configuration line: > > dsmgrd.urlregex alien://(.*)$ root://xrd.cern.ch/$1 > > - Source URL: > > alien:///alice/data/2012/LHC12b/000178209/ESDs/pass1/12000178209061.17/AliESDs.root > > - Resulting URL: > > root://xrd.cern.ch//alice/data/2012/LHC12b/000178209/ESDs/pass1/12000178209061.17/AliESDs.root > dsmgrd.sleepsecs secs : Seconds to sleep between each loop. The dataset stager checks at each loop the status of the managed transfers. Defaults to 30 seconds.

dsmgrd.scandseveryloops n : Every n loops, the dataset repository is checked for newly incoming staging requests. Defaults to 10.

dsmgrd.parallelxfrs n : Number of concurrent transfers. Defaults to 8.

dsmgrd.stagecmd shell_command : Command to run in order to stage each file. It might be whatever you want (executable, shell script...). If you add $URLTOSTAGE and/or $TREENAME in the shell_command, they'll be substituted respectively with the destination URL and the default ROOT tree name in the file (as specified in the dataset staging request from ROOT).

An example:

dsmgrd.stagecmd /path/to/afdsmgrd-xrd-stage-verify.sh "$URLTOSTAGE" "$TREENAME"

Return value of the command is ignored: standard output is considered, as explained here.

Defaults to /bin/false.

dsmgrd.cmdtimeoutsecs secs : Timeout on staging command, expressed in seconds: after this timeout, the command is considered failed and it is killed (in first place with SIGSTOP, then if it is unresponsive with SIGKILL). Defaults to 0 (no timeout).

dsmgrd.corruptafterfails n : Set this to a number above zero to tell the daemon to mark files as corrupted after a certain number of either download or verification failures. A value of 0 (default) tells the daemon to retry forever.

Configuring the MonALISA monitoring plugin

The Dataset Stager supports generic monitoring plugins. The only plugin distributed with the stager is the MonALISA monitoring plugin.

dsmgrd.notifyplugin */path/to/libafdsmgrd_notify_apmon.so* : Set it to the path of the MonALISA plugin shared object. By default, notification plugin is disabled.

dsmgrd.apmonurl apmon://apmon.cern.ch : This variable tells the ApMon notification plugin how to contact one or more MonALISA server(s) to activate monitoring via ApMon. It supports two kinds of URLs:

http[s]://host/path/configuration_file.conf (a remote file where to fetch the list of servers from)
apmon://[:password@]monalisahost[:8884] (a single server to contact directly)

If the variable is not set, yet the plugin is loaded, MonALISA monitoring is inhibited until a valid configuration variable is provided.

dsmgrd.apmonprefix MY::CLUSTER::PREFIX : Since MonALISA organizes information in "clusters" and "hosts", here you can specify what to use as cluster prefix for monitoring datasets information and daemon status. If this variable is not set, MonALISA monitoring is inhibited. Please note that the suffix _datasets or _status is appended for each of the two types of monitoring.

A sample configuration file

xpd.stagereqrepo /opt/aaf/var/proof/datasets
dsmgrd.purgenoopds true
dsmgrd.urlregex alien://(.*)$ /storage$1
dsmgrd.sleepsecs 20
dsmgrd.scandseveryloops 30
dsmgrd.parallelxfrs 10
dsmgrd.stagecmd /opt/aaf/bin/af-xrddm-verify.sh "$URLTOSTAGE" "$TREENAME"
dsmgrd.cmdtimeoutsecs 3600
dsmgrd.corruptafterfails 0