Only 1 slave provisioned per template

[alexschwantes]

It is a little unclear what the expected outcome is, however the actual behaviour differs from what I expected.

Given that I have the max Virtual machines set to a high number ie 5, when I start a job that is restricted to run on a configured Azure template label, the Azure plugin provisions a new slave. However if I start two jobs that are restricted to run on the same Azure template label, one job will start on the provisioned slave and the other one will just queue. What I expected to happen was that a new slave is provisioned to cater for the second job as the number of virtual machines in the subscription is less than my configured max.

Edit: After further investigation I did manage to get a second slave to be provisioned but it is not clear why.

With the azure slaves configured to only run 1 job at a time.

• If I use the Buildflow plugin to schedule 2 builds in parallel, all being restricted to run on the Azure label, only 1 slave will be provisioned.
• If I manually start 2 jobs that are all restricted on the same Azure slave, again only 1 slave will be provisioned.
• If I manually start 3 jobs, or use the Buildflow plugin to schedule 3 jobs then a second slave is provisioned. The Jenkins log showed the following reference to starting based on workload. Apr 24, 2016 12:09:31 PM INFO hudson.slaves.NodeProvisioner$StandardStrategyImpl apply Started provisioning rqa-swarm-linux from Azure-c55e2022-bc1f-46be-ab45-654433cca28f with 1 executors. Remaining excess workload: -0.197

It seems that the workload value is the crucial part in initiating a new slave to be provisioned. But why this isn't triggered when there is a queued job or from the Buildflow plugin I'm not sure.

I want to avoid queuing jobs, when there is room for more azure slaves to be provisioned.

[alexschwantes]

So digging in a bit more I found some information on this over at https://cloudbees.zendesk.com/hc/en-us/articles/204690520-Why-do-slaves-show-as-suspended-while-jobs-wait-in-the-queue- which points to the internal Jenkins provisioning strategy used to decide if a new node should be provisioned or not. It seems to get greater control over this they created their own NodeProvisioner Strategy.

Further, https://github.com/jenkinsci/mesos-plugin#over-provisioning-flags explains that Jenkins spawns slaves conservatively, and provides the following flags that can be overridden during Jenkins Startup to spawn an executor for each build in queue immediately without waiting: -Dhudson.slaves.NodeProvisioner.MARGIN=50 -Dhudson.slaves.NodeProvisioner.MARGIN0=0.85

Doing this so far has produced better results for my needs. It would be nicer however if this could be configured from within Jenkings rather than editing java opts...,