Fix machine-api resources #118
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -3,7 +3,10 @@ | |
| # Resource List | ||
| resources=() | ||
|
|
||
| # machine-api resources | ||
| machineresources=() | ||
|
|
||
| # Cluster Version Information | ||
| resources+=(clusterversion ns/openshift-cluster-version) | ||
|
|
||
| # Operator Resources | ||
|
|
@@ -12,8 +15,11 @@ resources+=(clusteroperators) | |
| # Certificate Resources | ||
| resources+=(certificatesigningrequests) | ||
|
|
||
| # MCO/Node Resources | ||
| resources+=(nodes machineconfigs machineconfigpools) | ||
|
|
||
| # machine-api resources | ||
| machineresources+=(machines machinesets) | ||
|
|
||
| # Namespaces/Project Resources | ||
| resources+=(ns/default ns/openshift ns/kube-system ns/openshift-etcd) | ||
|
|
@@ -25,13 +31,18 @@ resources+=(storageclasses persistentvolumes volumeattachments) | |
| resources+=(clusternetworks hostsubnets) | ||
|
|
||
| # Autoscaler Resources | ||
| machineresources+=(clusterautoscaler machineautoscaler) | ||
|
|
||
| # Run the Collection of Resources using must-gather | ||
| for resource in ${resources[@]}; do | ||
| /usr/bin/openshift-must-gather inspect ${resource} | ||
| done | ||
|
|
||
| # Run the Collection of MachineAPIResources using must-gather | ||
|
There was a problem hiding this comment. Why collect this separate? Why not just included this in resources? There was a problem hiding this comment. @sferich888 I'm not sure if it's intended to work as-is, but without this patch, we don't get any machines or machinesets from must-gather utility. Is there a better way? Note, 'machines' is already included in Machine/Node resources, but it doesn't collect any machines (aka, a bug). There was a problem hiding this comment. According to https://docs.openshift.com/container-platform/4.1/machine_management/applying-autoscaling.html the Should we just do some loop or update the loop above to get things from all namespaces? There was a problem hiding this comment. Let's consider https://github.com/openshift/must-gather/pull/127/files over this PR. |
||
| for resource in ${machineresources[@]}; do | ||
| /usr/bin/openshift-must-gather inspect ${resource} --namespace openshift-machine-api | ||
| done | ||
|
|
||
| # Collect System Audit Logs | ||
| /usr/bin/gather_audit_logs | ||
|
|
||
|
|
||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Do we have a relation ship between these?
IE: If we get a node; do we need to know its machineconfigs or machineconfigpools?
If so - it might be better to put this relationship into to the must-gather binary so every time a node is collected *(or one of the other objects); the proper related objects are collected.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I can't speak for everyone though I believe
machineconfigsandmachineconfigpoolsare namespace-less. That's why we need to query for machines separately since they live only underopenshift-machine-apinamespace.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@runcom FYI ^^
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
See https://github.com/openshift/machine-config-operator/blob/master/docs/MachineConfigServer.md#detailed-design
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@sferich888 I have a feeling it's impossible to know which machine config was used for a given node just by quering node. Based on posted [1], I don't think node object has any field/label/annotation that can say which endpoint was used. Also, worker nodes have different image than master nodes. Each image knows an endpoint (most likely it's hard-coded in it when it's baked). Due to security reason so you can't pretend to be master when you are worker. Or, am I wrong @runcom @ashcrow ?
[1] https://github.com/openshift/machine-config-operator/blob/master/docs/MachineConfigServer.md#detailed-design
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This patch has little to do with machineconfigs and everything to do with machines.
Could we automatically collect these resources when collecting nodes? Yes, but if what you're trying to troubleshoot is a machine that never became a node, that's not going to help. We should explicitly collect all these types.