noob wrangling questions

[Theo]

I hope my n00b questions dont offend you 
I've been monitoring a farm for a week now using Qube and I started seeing some weird patterns. It is not a busy time for renders at the moment but I would like to solve my questions before crunch time..

Here are my observations and related questions :

Sometimes jobs are pending even if they are set for a different set of machines than the ones being used. A job with high priority on a specific cluster (lets say /A ) will pend and not start rendering even if all of the /A machines are idle. Instead it will sometimes wait for a job at /B to finish and then use those..

Some other times jobs seem not to be using all the workers available to them. ( even if they are set for a higher number of instances than the available workers) Is there some kind of threshold of instances that can be used at the same time for a single job ? I doubt it so why is this randomly happening every now and then ?

I do understand that there is fundamental difference between the group parameter and the cluster parameter. All of the machines at the moment are set on clusters but not all of them are in groups. Will the wrangling be facilitated if all of the machines where to be put into groups ?

thnx in advance

[jburk]

Do the pending jobs have a reservation or resource requirement that can't be met by the machines that are sitting idle?  They may even be waiting for a global (farm-wide) resource (eg. a prman license).

Jobs can be configured so that the consume 3rd-party licenses that are counted by the supervisor, and if all those license are already consumed either by other jobs or other instances in the same job, the remaining job instances will be held pending.

Groups are used to be able to easily select a set of machines, without having to name each and every machine.  If you set a job's hostgroup, it will only run on hosts in that group.  If you set the hostgroup to multiple groups, it will run on any machine that's in any of those groups, not a host that is in every group (groups are OR'd together - grpA OR grpB).

Clusters are for sorting priorities between jobs.  See the clustering documentation at: http://docs.pipelinefx.com/x/BYN

[Theo]

hi jburk!

Yes , you are right! some of the renders are being limited by the number of available licenses and therefore are either pending or not using all the available workers.

I do get the difference between groups and clusters. However, I was wondering if I should put all my machines into groups whereas at the moment they are divided into clusters but not all of them are in groups.. I don't think I have anything to lose by that, if anything, it will offer me more control ..

Thank you for the help

[jburk]

Clusters allow a set of users to get better priority on machines than other users.  If you take away the clusters, you will likely lose some control.

Clusters and groups are very different, and do different things.  Host groups are used to decide which hosts can or will run a job.  Clusters are used to decide whose job runs on a worker when the farm is full.

You very much should read the docs about clustering and how it relates to job priority:  http://docs.pipelinefx.com/x/-YB

[Theo]

Thnx again jburk
appreciate the help