-- MichaelHynes - 2015-01-27

Himrod cluster

This is a cluster belonging to Ashraf Aboulnaga and Hans DeSterck, purchase in May 2014, running since October 2014

Sysadmin notes

Admin tools

  • IMPORTANT Currently I have a number of tools located under /cscf-adm/src/cluster - the plan is to move all of them to /usr/local/bin in the very near future.
  • This list below has been moved there

add_users

  • What:
    • * Add users from CVS optionally specify home directory, email address, password and groups*
    • If the group is admin - the user will be get all of the admin groups
  • See:
    • sync-users to sync all user changes to all of the nodes
  • Notes:
    • needs root

del-user

  • Delete a user on all of the nodes
  • Usage:
    • del-user userid
  • Notes:
    • needs root perms

sync-users

  • sync user accounts, passwords, ssh keys and group setting to all of the nodes
  • Usage:
    • sync-users
    • This command can be run any time, and more then once without harm
  • Notes:
    • needs root

Checking nodes and NFS mounts

  • /cscf-adm/src/cluster/fix-mount
    • Verifies NFS mounts are working - mounts them if not

Reset node using iDrac

  • /cscf-adm/src/IiDrac/reset_node ilom-NODE - where NODE is lised in the table below

ILOM managment

  • You can do full KVM management of each node from himrod (himrod has to be up)
    • This includes powering up, console, booting remote media from your own desktop, etc
  • Log onto himrod with X forwarding enabled: ssh -X userid@himrod.cs.uwaterloo.ca
    • Start firefox one himrod using these options: firefox --no-remote
    • Each node has an ILOM address 9listed below) open that url, for example, as https://ilom-himrod-1
    • Accept the security certificate
    • Login userid: cscf-adm - fall 2013 password - see CSCF staff
      • WE will be created a himrod user shortly
172.19.128.10 ilom-himrod-storage
172.19.128.11 ilom-himrod-big-1
172.19.128.12 ilom-himrod-big-2
172.19.128.13 ilom-himrod-big-3
172.19.128.14 ilom-himrod-big-4
172.19.128.101 ilom-himrod-1
172.19.128.102 ilom-himrod-2
172.19.128.103 ilom-himrod-3
172.19.128.104 ilom-himrod-4
172.19.128.105 ilom-himrod-5
172.19.128.106 ilom-himrod-6
172.19.128.107 ilom-himrod-7
172.19.128.108 ilom-himrod-8
172.19.128.109 ilom-himrod-9
172.19.128.110 ilom-himrod-10
172.19.128.111 ilom-himrod-11
172.19.128.112 ilom-himrod-12
172.19.128.113 ilom-himrod-13
172.19.128.114 ilom-himrod-14
172.19.128.115 ilom-himrod-15
172.19.128.116 ilom-himrod-16
172.19.128.117 ilom-himrod-17
172.19.128.118 ilom-himrod-18
172.19.128.119 ilom-himrod-19
172.19.128.120 ilom-himrod-20
172.19.128.121 ilom-himrod-21
172.19.128.122 ilom-himrod-22
172.19.128.123 ilom-himrod-23

check-nodes

  • check that each node is on line or not
  • Usage:
    • check-nodes
  • Notes:
    • This can be used as a common template we use to perform a task on all nodes
    • Check to See if all of the nodes are online

Packages

finding packages

  • Example: apt-cache search postgres -n
    • Search for postgres in the one line description
  • Example: apt-cahe search postgres
    • Searchj for postgres in the entire description

example script to install packages on all of the nodes

  • /cscf-adm/src/cluster/install-mpi
    • The script first installed the packages listed in the script on himrod and then on the nodes
    • The script is only 27 lines long and you will only have to change 2 lines!
    • NODES and "common_vars* is pulled in from the search path - in this case: /usr/local/bin
      • (ie. they doNOT have to be in the current directory)
!/bin/bash
#
# Mike Gore, 10 Oct 2014
#
# Install openmpi on the nodes and headnode

. common_vars
. NODES

update_list
update_packages netpipe-openmpi openmpi-bin openmpi-checkpoint openmpi-common openmpi-doc 


for i in $NODES
do 
   if ping -c 1 $i >/dev/null 2>&1
   then
   cat <<EOF | ssh root@"$i" 
. common_vars
. NODES
update_list
update_packages netpipe-openmpi openmpi-bin openmpi-checkpoint openmpi-common openmpi-doc 
EOF
   else
      echo $i is down
   fi
done

Disks on nodes

  • Each node has disks mounted with names /localdiskN where N is 0 .. 5

Tools

OPENMPI

  • I have installed openmpi on himrod and all of the nodes
  • As of 10 Oct 2014 only the configuration part has not been completed

Limits

I have added the following lines to the /etc/security/limits.conf file for now (until I am given a better idea of the other entries we must set)
(I used the /cscf-adm/src/cluster/sync-users script to update this file on the nodes - I ran it as root)
Note: the nodes may need to be restarted
There is a script that does this restart correctly is under /cscf-adm/src/cluster called restart_all_nodes
(Must be run as root)

# system defaults
*          hard    cpu             unlimited
*          hard    nproc           unlimited
*          hard    as              unlimited
*          hard    data            unlimited
*          hard    sigpending      unlimited
*          hard    nofile          unlimited
*          hard    msqqueue        unlimited
*          hard    locks           unlimited
*          hard    file            unlimited
Topic revision: r1 - 2015-01-27 - MichaelHynes
 
This site is powered by the TWiki collaboration platform Powered by PerlCopyright © 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback