[
https://ovirt-jira.atlassian.net/browse/OVIRT-2782?page=com.atlassian.jir...
]
Daniel Belenky updated OVIRT-2782:
----------------------------------
Description:
By default, Jenkins is configured to wait a few seconds before allocating a new node in a
hope that a node in use to be freed. We can control this setting by setting
{code:java}
hudson.slaves.NodeProvisioner.initialDelay=0
{code}
We can also control the excessive workload threshold and force Jenkins to allocate slaves
for jobs in the queue... but, because Jenkins computes the excess workload value (which
decides if we need to allocate a new node) using an EMA, we set it's margins to higher
values and it effectively lowers its threshold so that Jenkins will allocate nodes faster
and in advance.
The recommended settings by the k8s plugin to spawn a node for every build in the queue:
{code:java}
-Dhudson.slaves.NodeProvisioner.initialDelay=0
-Dhudson.slaves.NodeProvisioner.MARGIN=50
-Dhudson.slaves.NodeProvisioner.MARGIN0=0.85
{code}
was:
By default, Jenkins is configured to wait a few seconds before allocating a new node in a
hope that a node in use to be freed. We can control this setting by setting
{code:java}
hudson.slaves.NodeProvisioner.initialDelay=0
{code}
We can also control the excessive workload threshold and force Jenkins to allocate slaves
for jobs in the queue... but, because Jenkins computes the excess workload value (which
decides if we need to allocate a new node) using an EMA, we set it's margins to higher
values and it effectively lowers its threshold so that Jenkins will allocate nodes faster
and in advance.
The recommended settings by the k8s plugin to spawn a node for every build in the queue:
```
By default, Jenkins is configured to wait a few seconds before allocating a new node in a
hope that a node in use to be freed. We can control this setting by setting
`hudson.slaves.NodeProvisioner.initialDelay` to `0`. But, because Jenkins computes the
excess workload value (which decides if we need to allocate a new node) using an EMA, we
also can lower its threshold so that Jenkins will allocate nodes faster and in advance.
{code:java}
-Dhudson.slaves.NodeProvisioner.initialDelay=0
-Dhudson.slaves.NodeProvisioner.MARGIN=50
-Dhudson.slaves.NodeProvisioner.MARGIN0=0.85
{code}
Configure Jenkins to wait less and allocate more
------------------------------------------------
Key: OVIRT-2782
URL:
https://ovirt-jira.atlassian.net/browse/OVIRT-2782
Project: oVirt - virtualization made easy
Issue Type: Improvement
Reporter: Daniel Belenky
Assignee: infra
By default, Jenkins is configured to wait a few seconds before allocating a new node in a
hope that a node in use to be freed. We can control this setting by setting
{code:java}
hudson.slaves.NodeProvisioner.initialDelay=0
{code}
We can also control the excessive workload threshold and force Jenkins to allocate slaves
for jobs in the queue... but, because Jenkins computes the excess workload value (which
decides if we need to allocate a new node) using an EMA, we set it's margins to higher
values and it effectively lowers its threshold so that Jenkins will allocate nodes faster
and in advance.
The recommended settings by the k8s plugin to spawn a node for every build in the queue:
{code:java}
-Dhudson.slaves.NodeProvisioner.initialDelay=0
-Dhudson.slaves.NodeProvisioner.MARGIN=50
-Dhudson.slaves.NodeProvisioner.MARGIN0=0.85
{code}
--
This message was sent by Atlassian Jira
(v1001.0.0-SNAPSHOT#100107)