How to Install OpenPBS v20.0.1 into Gentoo Linux

Amasaki Shinobu — Mon, 18 Apr 2022 01:25:27 +0000

This article is written for people who build their own clusters on Gentoo Linux.

Index

Introduction
Installing
- Preparation
- Dependencies
- Build & Install
- Post-install settings
Launch
- Start Database Server
- Start the Service
- Register a Node
- Check the Startup Status
Conclusion
References

Introduction

OpenPBS is a job scheduling software for a High Performance Computing (HPC) system. This is one of the few open source software for job management system for HPC, and originally an open source version of proprietary products. This article tries installing it into Gentoo Linux.

Installing

Preparation

Following the Gentoo Linux styles, all software used this time are built and installed from the source codes. Installation of OpenPBS on other operating systems except Gentoo is described in openpbs/INSTALL on GitHub.

Note: This experiment is tested on Gentoo virtual machine with Vagrant on openSUSE Tumbleweed . Security settings are not taken into account. When actually using OpenPBS, SET security settings such as firewalld and iptables.

USE Flag Settings

USE Flag settings are described on the page of "HPC on Gentoo" on Gentoo Wiki.

# cat /etc/portage/make.conf | grep USE
USE="${USE} 3dnow fortran gnuplot gpm mmx ncurses nptl nptlonry openmp ompt
   pam perl ssl tcpd unicode vim-syntax xml zlib python networkmanager"

`sudo` and `git` (optionally)

Install the package sudo and git used for the main installation work.

# emerge --ask --verbose app-admin/sudo dev-vcs/git

Dependencies

Installing Packages by Portage

Install necessary packages to build OpenPBS.

#!/bin/bash
# openpbs_deps_install.sh

### Dependencies on Build-time
emerge --ask --verbose 
    sys-devel/gcc sys-devel/make sys-devel/libtool \
    sys-apps/hwloc sys-libs/ncurses dev-lang/perl dev-lang/python \
    dev-libs/openssl sys-devel/autoconf sys-devel/automake \
    dev-libs/libedit dev-db/postgresql dev-lang/swig \
    x11-libs/libX11 x11-libs/libXt x11-libs/libXext x11-libs/libXft \
    media-libs/fontconfig

### Dependencies on Run-time
emerge --ask --verbose
    dev-libs/libical x11-libs/libXrender mail-mta/sendmail

Execute this shellscript.

# ./openpbs_deps_install.sh

Building and Installing Tcl/Tk on Manually

Tcl is one of the script languages, and Tk is GUI toolkit of that. These are able to install by Portage. In that case, however, static libraries that building OpenPBS might depends on are not generated in /usr/lib.

Therefore, these two must build and install from Tarball on manually. The source codes can be downloaded from Tcl Developer Site.

First, install Tcl by executing following shellscript.

/usr is the place of installation and /usr/local is used as working directory.

#!/bin/bash
# tcl_install.sh

### Tcl
cd /usr/local
wget https://prdownloads.sourceforge.net/tcl/tcl8.6.11-src.tar.gz
tar xzf ./tcl8.6.11-src.tar.gz 
cd ./tcl8.6.11/unix
./configure --prefix=/usr --enable-static --disable-shared
make
make install

Execute this.

# ./tcl_install.sh

Next, install Tk.

#!/bin/bash
# tk_install.sh

### Tk
cd /usr/local
wget  https://prdownloads.sourceforge.net/tcl/tk8.6.11.1-src.tar.gz
tar xzf ./tcl8.6.11-src.tar.gz
cd ./tk8.6.11/unix
./configure --prefix=/usr --enable-static --disable-shared
make
make install

Execute this.

# ./tk_install.sh

Build & Install

In this article, the place of installation of OpenPBS is /opt/pbs. First, the git clone command of openpbs/openpbs on GitHub or extend tarball in order to get the source code. And then execute the shellscript of autoge.sh.

The Case of `git clone`

# cd /opt
# git clone https://github.com/openpbs/openpbs /opt/openpbs-20.0.1
# cd openpbs-20.0.1
# git checkout v20.0.1
# ./autoge.sh

The Case of Extending Tarball

# cd /opt
# wget https://github.com/openpbs/openpbs/archive/refs/tags/v20.0.1.tar.gz
# tar xzf v20.0.1.tar.gz -C ./openpbs-20.0.1
# cd openpbs-20.0.1
#./autoge.sh

Executing `configure`

Set the environment variable LDFLAGS="-ltinfo" to link libtinfo.so, and then execute ./configure with the following options.

# cd /opt/openpbs-20.0.1
# export LDFLAGS="-ltinfo"
# ./configure --prefix=/opt/pbs --libexecdir=/opt/pbs/libexec

Executing `make` & `make install`

Next, execute the commands make and make install. To avoid an error when executing the make command, set LDFLAGS as follows.

# export LDFLAGS="${LDFLAGS} -L/usr/lib64 -lfontconfig -lXft"
# make
# make install

Confirmation

Execute the following command, and if the version number is displayed, the installation is successfull.

# /opt/pbs/bin/pbs_hostn --version
pbs_version = 20.0.1

Post-install settings

Executing Post-install script

Execute the following command on each nodes.

# /opt/pbs/libexec/pbs_postinstall

Then, the file of /etc/pbs.conf is generated. The environment variable PBS_HOME is set to /var/spool/pbs and configuration files and log directories are generated.

Next, set the permissions of these two executable file as follows.

# chmod 4755 /opt/pbs/sbin/pbs_iff /opt/pbs/sbin/pbs_rcp

Settings of `/etc/pbs.conf`

/etc/pbs.conf is a file that describes the operation and role of the localhost. See the documents of PBS Pro "PBS Professional Installation & Update Guide" for the role of a host.

PBS_SERVER describes the hostname of the master node.
PBS_START_* describes the role of the host.

On the master node, the role of SERVER, SCHEDULER and COMMUNICATION are required, so set the variables as follows.

PBS_START_SERVER=1
PBS_START_SCHED=1
PBS_START_COMM=1

In this article, calculation is executed on the master node, so set the variable PBS_START_MOM to 1.

PBS_START_MOM=1

If you want to use a different compute nodes, set the above three variables to 0 and set PBS_START_MOM=1.

# cat /etc/pbs.conf
PBS_SERVER=master
PBS_START_SERVER=1
PBS_START_SCHED=1
PBS_START_COMM=1
PBS_START_MOM=1
PBS_EXEC=/opt/pbs
PBS_HOME=/var/spool/pbs
PBS_CORE_LIMIT=unlimited
PBS_SCP=/usr/bin/scpjk

Settings of`server_priv` and `mom_priv`

server_priv/nodes describes the hostnames of the nodes to be added. When adding more nodes, add a hostname that can be resolved in the same way. The settings of server_priv is set only for the node in charge of the server role, that is, the master node.

# cat /var/spool/pbs/server_priv/nodes
master np=1 allnodes

the settings of mom_priv/config are as follows:

# cat /var/spool/pbs/mom_priv/config
$clienthost master
$restrict_user_maxsysid 999

Write the hostname of the master node on the right of $clienthost.

Launch

Start Database Server

Load the configuration of postgresql-13 and start the database service.

## emerge —-config dev-db/postgresql:13
# /etc/init.d/postgresql-13 start

# rc-service postgresql-13 status
 * Checking PostgreSQL 13 status ...
pg_ctl: server is running (PID: XXXX)
/usr/lib64/postgresql-13/bin/postgres "-D" "/etc/postgresql-13"
...

If you want to start the service when the OS boots, execute the following command.

# rc-update add postgresql-13 default

Start the Service

Execute the pbs_habitat command which sets initial configuration in order to start the PBS service.

# /opt/pbs/libexec/pbs_habitat
***
*** Setting default queue and resource limits.
***
*** End of /opt/pbs/libexec/pbs_habitat

# rc-service pbs start
Starting PBS
/opt/pbs/sbin/pbs_comm ready (pid=XXXX), Proxy Name:master:17001, Threads:4
PBS comm
PBS sched
Connecting to PBS dataservice...connected to PBS dataservice@master
PBS server

Register a Node

Use the qmgr command to register and manage nodes.

# /opt/pbs/bin/qmgr -c "create node master"     #-> node registration

# rc-service pbs restart                        #-> service restart

Check the Startup Status

Thepbsnodes command displays the current status of the nodes. If it says state = free, starting pbs is successful. In this state, the nodes is ready to receive a job and compute the calculation.

$ pbsnodes -a
master
     Mom = master
     ntype = PBS
     state = free
     pcpus = 1
     resources_available.arch = linux
     resources_available.host = master
     resources_available.mem = 2033580kb
     resources_available.ncpus = 1
     resources_available.vnode = master
     resources_assigned.accelerator_memory = 0kb
     resources_assigned.hbmem = 0kb
     resources_assigned.mem = 0kb
     resources_assigned.naccelerators = 0
     resources_assigned.ncpus = 0
     resources_assigned.vmem = 0kb
     resv_enable = True
     sharing = default_shared
     license = l
     last_state_change_time = Sat Sep 25 10:05:51 2021

Conclusion

This article describes the procedure for installing the job scheduler OpenPBS on Gentoo Linux. Install most of the dependent packages with Portage, and install some manually to prepare the library files. You can build and install by specifying reference to those library files.

References

Note

This article is translated from a Japanese article on MY website.

DEV Community: Amasaki Shinobu