P103870. An enhancement to LSF fairshare
component. The fairshare run time field type has
been changed to a long from an int to avoid a potential overflow
situation.
Readme documentation for IBM Spectrum LSF
10.1.0.10 Fix 558923 including installation-related
instructions, prerequisites and
co-requisites.
The fairshare run time field type has been changed to a long
from an int to avoid a potential overflow situation.
Readme file for: IBM® Spectrum LSF
Product/Component Release: 10.1.0.10
Update Name: Fix 558923
PMR/APAR: P103870
Fix ID: lsf-10.1-build558923-csm
Publication date: 17 September 2020
Last modified date: 17 September 2020
Contents:
1. List of fixes
2. Previous fixes
3. Download location
4. Product notifications
5. Products or components affected
6. System requirements
7. Installation and configuration
8. List of files
9.
Copyright and trademark
information
ID
|
Description
|
P103735 |
An enhancement to LSF that allows you to use the -f parameter with the 'brsvmod addhost' command in a similar way to using the brsvadd command. |
P103620
|
This fix resolves an
issue with mbschd and preemption that might cause a
core dump.
|
P103447
|
Fixed preemption calculations in the
allocation planner related to NO_PREEMPT_INTERVAL,
NO_PREEMPT_FINISH_TIME, and PREEMPT_DELAY
|
P103322
|
Fixed incorrect
calculation of hierarchical farishare
factor for absolute priority scheduling.
|
RFE#140975
|
An enhancement to LSF
to change the default behavior of what happens to a
job that fails during setup. This enhancement allows
you to configure LSF so that it repeats the attempt to
schedule or dispatch a job if the setup fails with a
specific CSM API error code.
|
7.1 Before
installation
1.
Shutdown LSF on all work load
manager (WLM) and launch node (LN) hosts.
2.
Back up the LSF
configuration from the conf
($LSF_ENVDIR) directory and any scripts or binary files that you
added to the
LSF_SERVERDIR directory (for example, customized elims,
esubs, stage in pre-scripts, stage
in post scripts).
3.
Run rpm -qa| grep lsf
to list the currently installed rpm files.
4.
Use yum erase or rpm -evh
to unistall the existing LSF
packages.
rpm -ev
--allmatches -notriggers
'list of filenames as they appear in step 3'
7.2 Installation
steps
7.2.1 Server
Package
Instructions
Download
lsf-10.1.0.10-558923.ppc64le_csm.bin package
Run
lsf-10.1.0.10-558923.ppc64le_csm.bin to extract the RPM files.
Accept the
license agreement when prompted to continue with the file
extraction.
lsf-common-10.1.0.10-558923.ppc64le.rpm
lsf-master-10.1.0.10-558923.ppc64le.rpm
lsf-misc-10.1.0.10-558923.ppc64le.rpm
lsf-server-10.1.0.10-558923.ppc64le.rpm
lsf-python2-api-1.0.6-10.1.0.9.ppc64le.rpm
ibm_jsm-10.4.0.00rtm3-rh8_20200604.ppc64le.rpm
Use rpm -ivh
or yum install commands to deploy the common, server, and master
RPM packages.
The installation is relocatable (--prefix options supported)
On the work
load manager
rpm -ivh
lsf-common-10.1.0.10-558923.ppc64le.rpm
lsf-server-10.1.0.10-558923.ppc64le.rpm
lsf-master-10.1.0.10-558923.ppc64le.rpm
On the launch node
rpm -ivh
lsf-common-10.1.0.10-558923.ppc64le.rpm
lsf-server-10.1.0.10-558923.ppc64le.rpm
Verify that the
installation is
successful:
rpm -qa
| grep lsf
lsf-server-10.1.0.10-558923.ppc64le
lsf-common-10.1.0.10-558923.ppc64le
lsf-master-10.1.0.10-558923.ppc64le
7.2.2 X86_64
Client Package
Instructions
Obtain and run
lsf-client-10.1.0.10-558923.x86_64.bin package to extract the
rpm packages.
This should be done on x86_64 client host.
lsf-conf-10.1.0.10-558923.noarch.rpm
lsf-man-pages-10.1.0.10-558923.noarch.rpm
lsf-client-10.1.0.10-558923.x86_64.rpm
lsf-python2-api-1.0.6-10.1.0.9.x86_64.rpm
The installation
directory is
/opt/ibm/spectrumcomputing/lsf.
The user lsfadmin
must be added if it does not exist on the host before installing
the lsf-conf rpm package.
rpm -ivh
./lsf-conf-10.1.0.10-558923.noarch.rpm
./lsf-man-pages-10.1.0.10-558923.noarch.rpm
./lsf-client-10.1.0.10-558923.x86_64.rpm
./lsf-python2-api-1.0.6-10.1.0.9.x86_64.rpm
Verify that the
installation is
successful:
rpm -qa
| grep lsf
lsf-conf-10.1.0.10-558923.noarch
lsf-man-pages-10.1.0.10-558923.noarch
lsf-client-10.1.0.10-558923.x86_64
lsf-python2-api-1.0.6-10.1.0.9.x86_64
7.3 After
installation
7.3.1 Server
Package
Restore previously
backed up LSF
conf directory and any customized scripts or binaries in
LSF_SERVERDIR
(Optional) Under
LSF_TOP:
Rename work.rpmsave
to work
Using bash: On
each WLM and LN,
run the following commands as root:
. /opt/ibm/spectrumcomputing/lsf/conf/profile.lsf
or if you have
used the --prefix
to define your own LSF_TOP
. /conf/profile.lsf
Start up
LSF
lsf_daemons start
7.3.2 X86_64
Client Package
Configuring client
hosts within
your cluster.
Log on to the LSF master host as the user lsfadmin.
Static Client
It is recommended
that the x86_64
hosts be configured as static clients.
Edit the
LSF_CONFDIR/lsf.cluster.cluster_name
file and add host information for
the new host to the Host section.
•
Add the name of the host.
•
Add model or type.
o
If you enter the ! keyword in
the model and
type columns, the host model and type reported by LSF will be
UNKNOWN.
o
Alternatively, X86_64 can be specified for type.
• Specify
LSF server or client in the server
column:
o
You must specify 0 (zero) to indicate an LSF client-only
host.
Begin Host
HOSTNAME
model
type
server
RESOURCES
<< snip >>
clienthost !
X86_64
0 ()
End Host
It is important that all host names of client hosts added to the cluster are resolvable by the management nodes. Depending on your site’s DNS, you can configure a $LSF_ENVDIR/hosts file that includes all official names and IP addresses of the client hosts. If LSF cannot resolve the host name of a client host it will reject requests originating from that host.
Float Client
Alternatively, you
could
configure the x86_64 hosts as floating clients. You do not add
the host
information for the new host to the Host section of the
LSF_CONFDIR/lsf.cluster.cluster_name
file. You must add the parameters
FLOAT_CLIENTS_ADDR_RANGE and FLOAT_CLIENTS to the Parameters
section of the
LSF_CONFDIR/lsf.cluster.cluster_name
file.
The FLOAT_CLIENT
parameter sets
the maximum allowable size for floating clients in a cluster. If
FLOAT_CLIENTS
is not specified in lsf.cluster.cluster_name,
the floating LSF client feature is disabled. When the LSF
floating client
feature is enabled, any host can submit jobs to the cluster. You
can limit
which hosts can be LSF floating clients with the parameter
FLOAT_CLIENTS_ADDR_RANGE.
The
FLOAT_CLIENTS_ADDR_RANGE
specifies IP address or range of addresses of domains from which
floating
client hosts can submit requests.
A floating client can join the cluster only when an LSF command is run from this host and LSF has verified it against FLOAT_CLIENTS (total number of floating clients) and FLOAT_CLIENTS_ADDR_RANGE (IP range). After verification, it is listed by the lshosts command. A floating client expires at 12:00 midnight or when the master LIM is reconfigured or restarted.
Begin Parameters
FLOAT_CLIENTS_ADDR_RANGE=
IP_address
...
FLOAT_CLIENTS= number_of_floating_clients
End Parameters
Now reconfigure
the cluster.
• Reconfigure
lim
to enable the new host in the cluster.
% lsadmin
reconfig
• Restart mbatchd.
% badmin mbdrestart
Check your
connectivity
Run the lsid and bhosts command from a client host. these commands should return the same information as if they were run from a CORAL server machine.
Submitting Jobs
When submitting jobs from an x86_64 client host and your cluster’s execution hosts are not x86_64 you should specify the resource requirement -R “type==any”. This tells LSF to ignore host type when selecting hosts for the job. Note that when submitting CSM job in easy mode the resource requirement is automatically set.
7.4 Uninstallation
7.4.1 Server
Package
Shut down LSF on all WLM and LN hosts.
lsf_daemons
stop
Back up the LSF
conf directory
and any customized scripts or binaries as stated in previous
steps.
Use yum erase or
rpm -evh commands to unistall LSF
following the same previous steps.
7.4.2 X86_64 Client Package
Log on to the
x86_64 client host.
Use the rpm uninstall command to uninstall the rpm packages.
# rpm -ev --allmatches --notriggers lsf-client-10.1.0.10-558923.x86_64 lsf-man-pages-10.1.0.10-558923.noarch lsf-conf-10.1.0.10-558923.noarch lsf-python2-api-1.0.6-10.1.0.9.x86_64