6. LSF Database - UNIX

This chapter is aimed at the LSF administrator running a UNIX cluster or a mixed UNIX/NT cluster. If your cluster contains only Windows NT hosts, see Chapter 7, `LSF Database - Windows NT', on page 49.

Note

LSF database administration is the only part of LSF that requires you to log on using the LSF primary administrator user account. The LSF primary administrator account will be the first cluster administrator specified in the lsf.cluster.cluster file. If your cluster includes any Windows NT hosts, do not attempt to change the LSF primary administrator after LSF has been installed.

In This Chapter

• `LSF Database Installation' on page 42
• `LSF Database Utility Commands' on page 43
• `LSF Data Collection Parameters' on page 45
• `lsb.acct Data Conversion' on page 46

LSF Database Installation

Installation

Install LSF Analyzer using lsfsetup, the UNIX installation program for LSF Suite. For detailed instructions, see the LSF Installation Guide.

When you install LSF Analyzer for the first time, you are prompted to specify the LSF database server. This defines the LSF_DB_HOST parameter in the lsf.conf file. The LSF database files will be installed on the machine you specify.

If you specify the installation host (the machine used to run lsfsetup) as the LSF database server, the LSF database installation program starts automatically.

If you specify another machine as the LSF database server, you have to install the LSF database using a separate installation file that is included in the regular LSF distribution. Complete the regular LSF installation, using lsfsetup, and then log onto the LSF database server and run lsfdbsetup.

When you install the LSF database, you are prompted to specify the LSF database directory. By default, this is /usr/local/lsf_db. All the LSF database files are installed under this directory. The database is created automatically and has the same name as the cluster.

Starting the Database

Once the database is installed, you must start it.

To start the LSF database, log onto the LSF database server as the LSF primary administrator and type:

% lsdbserver start

The lsdbserver command is located in the /bin subdirectory of the LSF database directory.

LSF Database Utility Commands

The utilities described in this section help you monitor and limit the size of the LSF database, to get better performance from LSF Analyzer. They are used to:

• start up and shut down the LSF database
• create and drop LSF databases
• move or delete existing data in the LSF database
• show the status of the LSF database
• show the size of the LSF database
• rebuild the indices of the LSF database

Environment Configuration

The lsf.conf file and the online documentation for database utility commands should be accessible to the working environment of the LSF primary administrator on the LSF database server (LSF_ENVDIR must be set). For example:

% setenv PATH $DB/bin:$PATH
% setenv MANPATH $DB/man:$MANPATH

where $DB is the LSF database directory.

Command Reference

To run any of the commands in this section, you must log onto the LSF database server as the LSF primary administrator. The following commands all support the standard -h option.

lsdbserver start|stop
Starts up or shuts down the LSF database.

lsdbcreate database_name
Creates a new database with the LSF database schema. Specify a unique database name (you cannot use the name of an existing LSF database).

lsdbstatus
Prints the current status of the LSF database server (which configuration file is in use; maximum and actual number of connections; details of all connections).

lsdbrecords database_name [ -t time ]
Prints reference numbers for job-related and resource-related records. If -t is specified, prints another pair of reference numbers, for job and resource records logged before the specified time.

If you want good performance from LSF Analyzer, you should keep the LSF database small. When the reference numbers exceed 50,000 for job records or 400,000 for resource records, you should archive the data using the lsdmove command.

The rate of growth of your LSF database depends on all of the following:

• number of jobs issued daily in the cluster
• number of hosts in the cluster
• number of resource indices logged for the cluster

-t time
To specify time, use the following syntax:

time = [year /] month / day / hour : minute

Use 4 digits for the year, if specified. Use 2 digits for month, day, hour, and minute.

lsdbbuildidx database_name
Reindexes the specified LSF database. If LSF Analyzer users have trouble retrieving data from the database, try this command, as there might be a problem with the database index.

lsdbclear database_name
Deletes all the logged data from the data tables in the specified LSF database, leaving an empty LSF database.

CAUTION!

There is no way to recover the deleted data.

You should not run this command on a working database, such as the online LSF database. If you do, use the lsdbstatus command and make sure there are no open connections to the database before using this command. This will prevent data inconsistency problems.

lsdbdrop database_name
Drops the specified LSF database (deletes the entire LSF database, including all of the logged data).

CAUTION!

There is no way to recover a dropped LSF database.

lsdbmove source_database_name destination_database_name
Moves the contents of the source database (data) to the destination database. This command must be followed by the badmin reconfig command.

LSF Data Collection Parameters

Data collection can be tuned by modifying the parameters configured in the lsb.params file. For example, edit:

LSB_CONFDIR/cluster/configdir/lsb.params

The following parameters can be configured:

DB_DEFAULT_INTVAL
Specifies the time interval to log job data (except load information and resource usage of running and suspended jobs) to the database, in minutes. To stop logging job data, set DB_DEFAULT_INTVAL=-1.

Default: 5

DB_JOB_RES_USAGE_INTVAL
Optional. If defined, specifies the time interval to log resource usage of running and suspended jobs to the database, in minutes.

Default: undefined (job resource is not logged).

DB_LOAD_INTVAL
Specifies the time interval to log load information (internal and external load indices and shared resources), in minutes.

Minimum: 15

Default: 60 (1 hour)

DB_SELECT_LOAD
Optional. Specifies which load information to collect. Possible values are internal load indices, external load indices, and shared resources. This parameter is case sensitive. If more than 4 load values are specified (separated by spaces) only the first 4 will be used. For example,

DB_SELECT_LOAD = ut mem ext_idx1 shared_lic1

will collect CPU utilization, available memory, the user-specified external load index ext_idx1, and the user-specified shared resource shared_lic1.

Default: ut

lsb.acct Data Conversion

The acct2db utility is used to convert job log files (e.g., lsb.acct) into LSF databases, allowing you to analyze LSF data collected before LSF Analyzer was installed. These databases cannot be used as online (active) databases. The license for the acct2db utility expires 30 days after your LSF 3.2 license is generated.

To convert data from an existing job log file, take the following steps:

1. Use the LSF primary administrator user account to log onto the host where the job log file (e.g., lsb.acct) is stored.
2. Create a new LSF database using the lsdbcreate command described in the `Command Reference' on page 44.
3. Issue the acct2db command and specify the new database as the target database. Each finished job record in the job log file is converted into a database record.

Syntax

acct2db [ -h ] [ -V ]
        [ -f lsb.acct.file ]
        [-H database_host ]
        database_name

-h
Prints command usage to stderr and exits.

-V
Prints the LSF release version to stderr and exits.

-f lsb.acct.file
Specifies the job log file on the local host which is to be converted into a database.

Default: LSB_SHAREDIR/cluster/logdir/lsb.acct

-H database_host
Specifies the remote host where the target database is located.

Default: the local host

database_name
Specifies the name of the target LSF database. The target database cannot be the online (active) database.