Randomize Maintenance Window Start Time
Currently, randomization only applies to deadlines and does not take maintenance windows into account. If you have a heavily virtualized environment with shared storage and want to use maintenance windows there is no good way to prevent a resource storm at the start of the maintenance windows. The only way is to break up the devices into several collections with staggered maintenance windows. This is cumbersome to maintain at a level that would provide the desired granularity.
Ideally there would be an option to randomize the maintenance window start time. Bonus points to making that randomization configurable.
bdam
shared this idea
2 comments
-
Bryan Dam
commented
Worth noting here is that the randomization period should not equal the MW period itself ... it's not just a checkbox. Randomly starting the install process 30 seconds before the end of the MW would b bad. We need some way of specifying that for the first X period of the MW the randomize spread will occur. Would probably be wise to put in some validation/warning if the user leaves less than say 2 hours for actual patching to happen.
What if there's multiple windows? Debatable but I think the ideal would be to use whatever the latest randomization end is and use that. On that basis that more spread is better and that there should always be a buffer at the end for actual patching.
-
Anonymous
commented
we had a production issue on a datacenter when the 1500 servers, that share a common maintenance window, restarted at the same time after a software update distribution, you can imagine the result. The feature to randomize the maintenance window would be helpful. Thanks