xCAT / Bugs / #1157 IBMhpc on AIX diskfull: Requisite Failures on PE,ESSL,HPC

yan feng han - 2010-05-28

This issue only happened when I isntalled ESSL following "Setting up ESSL and PESSL in a Stateful Cluster " guide, and the similar issue in the process of installing PE folloowing "Setting up PE in a Stateful Cluster "- it missing bos.loc.utf.EN_US for the xlfrte.msg.EN_US lpp. So fix this for PE together.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

yan feng han - 2010-05-28

I tried this "updatenode" for PE and ESSL on a clean node (it is installed no any HPC bundles with OS) and found comilpers lpps xlfrte.msg.en_US and xlsmp.msg.en_US.rte requires bos.loc.iso.en_US. So both bos.loc.iso.en_US and bos.loc.utf.EN_US were not installed with OS automatically, and when ran "updatenode" for PE or ESSL which including compilers bundle, the "-g" couldn't install the two bos lpps too, the reason was the two bos lpps are "I" instead of "U":

c711f1ec01: Requisite Failure Key:
c711f1ec01: "I" fileset must be installed prior to or at the same time as its
c711f1ec01: dependent filesets. This requisite cannot be installed automatically
c711f1ec01: (using -g option). It must be explicitly selected for installation.

So I must add them into lpp_source and install them by "nim -o cust". It is not
convenient for users, since HPC integration likes to install every thing related with HPC automatically, I think we need to make a better way for it. I suggest that we make a bos_msg bundle to include them, and add this bundle into the osimage.installp_bundle before we begin installing OS, and that will install these lpps with OS. I tried this way on my cluster it woked well. And a advantage of this way is we can easily add any other bos prereqs as needed later.

The same failures happened when I installed PE(compilers,pe bundles), or ESSL(compilers,essl bundles) , or IBMhpc_all bundle, with the OS together, so we may need to add the bos_msg bundle before all hpc bundles before for installing a new node. Then the issue will be fixed. I tried it on my cluster too, it worked.

Since all HPC integrations components related, the defect may effect some components. My suggstion is:
1> for PE and ESSL, we need to modify "Setting up ESSL and PESSL in a Stateful Cluster " and "Setting up PE in a Stateful Cluster "guides for ppe.poe, to let user add it into lpp_source.
2> for PE, ESSL and HPC together, we need to make a bos_msg bundle to include the two bos lpps and add it before any hpc bundles into the osimage.

Thx.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Brian Croswell - 2010-05-28

Linda, Han Yan ..

Another possibility would be to include the bos.loc.utf.EN_US and the bos.loc.iso.en_US file sets as part of the IBMhpc_base.bnd which contains the I base requisite and is already a base requirement for the HPC LPPs ....
It looks like the bos.loc.iso.en_US fileset is already part of the base AIX NIM LPPSOURCE.
We will need to instruct the admin to copy/install the bos.loc.utf.EN_US file set into the AIX NIM LPPSOURCE with nim -o update .
I also noticed that the bos.cpr fileset listed in the HPC bundle in also not part of the base AIX NIM LPPSOURCE and will need the admin to copy/install ..

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

yan feng han - 2010-05-28

When installing ppe.poe, two requisites are rsct.lapi.rte and mcr, so we need to put this two lpps into lpp_source, then with "-g", they can be installed. So need to modify PE and ESSL guide for them.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

yan feng han - 2010-05-28

the mcr should be mcr.rte in the comments above.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Brian Croswell - 2010-05-28

Linda, Han Yan,

We need to include AIX bos.loc.com.utf file set to be included in LPPSOURCE and bundle since it is a bos.loc.utf.EN_US a co-req ..
I had a good implementation with HPC compilers, pe, essl working with my FVT configuration xCAT MN hv32lpar05 (xcat20) and node hv32lpar11 ...
Here was my update and setup as part of the HPC implementation ..

1) As part of the HPC pre-req activity . I needed to include IBMhpc_base.bnd as part of the HPC implementation. I copied the IBMhpc_base bundle down to /install/nim/installp_bundle. I then made updates to this bundle file to include bos.loc.utf.EN_US and bos.loc.com.utf images . I also included bos.cpr image in this bundle file. I did the nim -o define to include bundle as IBMhpc_base
and then included it as part of the xCAT AIX osimage used with LPAR
[xcat20][/]> cat /install/nim/installp_bundle/IBMhpc_base.bnd

AIX bundle file containing base AIX prereq for all HPC software

I:bos.adt.debug
I:bos.adt.libm
I:bos.adt.syscalls
I:bos.adt.prof
I:bos.cpr
I:bos.loc.iso.en_US
I:bos.pmapi
I:bos.loc.utf.EN_US
I:bos.loc.com.utf
I:perfagent.tools
I:sysmgt.sguide.rte
I:Java5

[xcat20][/]> lsdef -t osimage -l
Object name: AIX61JSP1
bosinst_data=AIX61JSP1_bosinst_data
imagetype=NIM
installp_bundle=xCATaixSSL,xCATaixSSH,IBMhpc_base,compilers,pe,essl
lpp_source=AIX61JSP1_lpp_source
nimmethod=rte
nimtype=standalone
osname=AIX
spot=AIX61JSP1

I agree with Han Yan that we do need to make sure that rsct.lapi and mcr need to be included in the LPPSOURCE as part of the HPC ppe implementation
We also should plan to include the pe bundle as part of the essl implementation. I am not sure if the pe_install script is required though.
For my implementation I included compilers, pe and essl as one implementation set for the HPC node installation.
[xcat20][/]> lsdef -l hv32lpar11

Object name: hv32lpar11
cons=hmc
groups=lpar,all,computehpc
hcp=w1m6hmc01
id=12
mac=00215EA68CDB
mgt=hmc
nodetype=lpar,osi
os=AIX
parent=c918f9fsp01-8236-E8C-SN1004E1P
postbootscripts=otherpkgs
postscripts=syslog,aixremoteshell,syncfiles,IBMhpc.postscript,compilers_license,pe_install,essl_install
pprofile=hv32lpar11
profile=AIX61JSP1
provmethod=AIX61JSP1
status=booted
statustime=05-28-2010 11:43:38

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Anonymous - 2010-05-30

Brian, Han Yan,
Thank you both for all your investigation on this issue.
I did not realize that PESSL requires PE for the MPI libraries.
I have added PE to all the ESSL/PESSL wiki pages, and included the two bos.loc.utf entries in the AIX base bundle file.
I see I also had forgotten to include the base bundle in all the instruction pages. This has been added.

I think I have all the parts of this fixed now. Please reopen this defect if you still see problems.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

yan feng han - 2010-05-31

Linda, is there a new build with the fix of this defect? I'd like to verify it now on my c711 cluster, this cluster may be taken away by other teams at some point, then I may have no cluster to verify it. If the build is not ready, can you put the efix on my c711? Thanks.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

yan feng han - 2010-06-09

Verified in xcat 2.4.2 0608 build on c711, and all packages in the bundle files can be installed now.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

IBMhpc on AIX diskfull: Requisite Failures on PE,ESSL,HPC

An extreme cluster/cloud administration toolkit

Milestone

Searches

Help

#1157 IBMhpc on AIX diskfull: Requisite Failures on PE,ESSL,HPC

Updatenode c711f1ec03 -S installp_bundle=compilers,essl installp_flags="-agXY" -V

Discussion

AIX bundle file containing base AIX prereq for all HPC software