[Spacewalk-list] Machines upgrade behavior within spacewalk

[Michal Bruncko]

Hello list,

I am happily using Spacewalk for managing our systems for years and I 
have question about clients machine upgrade process.

- We are using virtualization environment where on single host are 
running ~20 virtual systems (and several virtualization hosts are 
available).
- I am updating package channels one every week (just RHEL 6.x x64_86 
systems) by crond script.
- And for upgrading machines, I simply choose all client systems from 
list, click on "manage" and then on "upgrade" links and apply all 
relevant upgrade packages at once.

but within some weeks there are lot of upgraded packages available 
including kernel and selinux packages, which consumes more resources of 
virtualization hosts like standard/other/smaller packages. If I use same 
upgrade process described bellow for upgrade package bundle including 
kernel packages, the virtualization host is obviously utilized and disk 
utilization increased to maximum.This results to slower service 
availability and I am also afraid about disk arrays if this very high 
utilization can cause some troubles for them.

Now I am wondering if there is way to use serialized 
(machine-per-machine) upgrade process instead of parallel (all machine 
at once) process without manual intervention by admin. As I can see from 
upgrade wizard I can only choose to "Install updates ASAP" or "Schedule 
upgrade process to specified time". I also wondered about some manual 
upgrade process in way that I choose only some virtuals from list 
instead of all, but I need to identify which virtuals is hosted on which 
virt. host. Yes, I can create groups (of uncollision machines from 
different virtual host) for such upgrade to minimize disk utilization, 
but it is not very handy and more time by admin needs to be consumed to 
update all machines.

Is there any other options that I have not yet recognized (i.e. using 
via CLI instead of WebGUI) that can help me with such problems? I am 
wondering about two options:

1. serializing installation per virtual machine instead of all-at-once 
behavior which is available now. this have advantage mainly in cases - 
when two redundant machines are serving same service (i.e. coroporate 
directory) and directory process needs to be restarted in oder to 
upgrade this package - which cause service unavailability at all as both 
virtuals will be in upgrade process at same time.

2. second way can be work similar like with clam.d process upgrade like 
we know - it us running on daily basis, but the upgrade process is 
executed in random time within next four hour from executing this script 
by cron process. So I am talking about choice "Upgrade randomly within 
XX hours/minutes per every selected machine from now". This will spread 
upgrade process of all selected machines randomly over defined time 
interval. And of course the disk utilization will be not so high utilized.

What do you think about those proposals?

thank you

-- 
Ing. Michal Bruncko, PhD., CCNP, RHCSA™
IT systems and network administrator
Coupled school of business and services Ruzomberok
Slovak Republic