Bug 1878108
| Summary: | OCP 4.6 installation fails in CPU quota check for OSD on GCP | ||
|---|---|---|---|
| Product: | OpenShift Container Platform | Reporter: | Manuel Dewald <mdewald> |
| Component: | Installer | Assignee: | Abhinav Dahiya <adahiya> |
| Installer sub component: | openshift-installer | QA Contact: | To Hung Sze <tsze> |
| Status: | CLOSED ERRATA | Docs Contact: | |
| Severity: | urgent | ||
| Priority: | urgent | CC: | adahiya, gshereme, jeder, lseelye, mdewald, mwoodson, rrackow, yanyang |
| Version: | 4.6 | Keywords: | Regression, ServiceDeliveryBlocker |
| Target Milestone: | --- | ||
| Target Release: | 4.6.0 | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | If docs needed, set a value | |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2020-10-27 16:40:10 UTC | Type: | Bug |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
The default GCP machine names look like this: m2-ultramem-208 where Abhinav's assumption in machineTypeToQuota() holds up. 208 is the vcpu count. But GCP can have custom types too, and we use these extensively in OSD. custom-4-16384 preserving information from Slack: GPC is allowing custom image that don't start with the machine type and then assumes "N1" as type. ```$ gcloud beta compute instances create example-instance-test --machine-type custom-4-3840 Created [https://www.googleapis.com/compute/beta/projects/innate-attic-182119/zones/us-east1-b/instances/example-instance-test]. NAME ZONE MACHINE_TYPE PREEMPTIBLE INTERNAL_IP EXTERNAL_IP STATUS example-instance-test us-east1-b custom (4 vCPU, 3.75 GiB) 10.142.0.4 34.75.155.249 RUNNING``` ```$ gcloud beta compute instances create example-instance-test-1 --machine-type test-custom-4-3840 ERROR: (gcloud.beta.compute.instances.create) Could not fetch machine type: - The resource 'projects/innate-attic-182119/zones/us-east1-b/machineTypes/test-custom-4-3840' was not found``` That is causing a problem when the name of the machine type is analzed here https://github.com/openshift/installer/blob/287658271951b5f8dbf1a77c77ff1557d81c5931/pkg/asset/quota/gcp/gcp.go#L199 @mdewald, did you create the custom type: custom-4-16384? If yes, could you please include more details about the type? I want to add a test case to capture this change and reflect what you have / had. Thanks in advance. test case (stub): https://polarion.engineering.redhat.com/polarion/#/project/OSE/workitem?id=OCP-35198 custom-4-16384 is a shortened alias for n1-custom-4-16384, which is a built-in GCP machine type. It means 4 CPU, 16384 memory. https://cloud.google.com/compute/docs/machine-types#custom_machine_types using 4.6.0-0.nightly-2020-09-21-114202 and
compute:
- architecture: amd64
hyperthreading: Enabled
name: worker
platform:
gcp:
osDisk:
DiskSizeGB: 0
DiskType: ""
type: custom-4-16384
replicas: 3
(same for control)
I am able to bring up a cluster and the machine (as shown in web console) has correct type)
Instance Type
custom-4-16384
From gcp console:
Machine type
custom (4 vCPUs, 16 GB memory)
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory (OpenShift Container Platform 4.6 GA Images), and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2020:4196 |
Description of problem: When installing 4.6 in OSD, the quota check returns failed to generate asset \\\"Platform Quota Check\\\": error(MissingQuota): compute.googleapis.com/cpus is not available in us-east1 because The required number of resources (114688) is more than the limit of 2400\"\nlevel=fatal msg=\"bootstrap host address and at least one control plane host address must be provided\"\n" installID=cz4gg5hr How reproducible: Install a OCP cluster with a install config like the following on GCP: compute: - name: worker platform: gcp: osDisk: DiskSizeGB: 0 DiskType: "" type: custom-4-16384 replicas: 4 controlPlane: name: master platform: gcp: osDisk: DiskSizeGB: 0 DiskType: "" type: custom-4-16384 replicas: 3 kind: InstallConfig metadata: creationTimestamp: null labels: api.openshift.com/environment: integration api.openshift.com/id: some-id api.openshift.com/managed: "true" api.openshift.com/name: some-name hive.openshift.io/cluster-type: managed name: gshereme-test1 namespace: uhc-integration-1fkg61fq9oieabah0b1i03k2es3l1mgs networking: clusterNetwork: - cidr: 10.128.0.0/14 hostPrefix: 23 machineCIDR: 10.0.0.0/16 machineNetwork: - cidr: 10.0.0.0/16 serviceNetwork: - 172.30.0.0/16 platform: gcp: projectID: some-project region: us-east1 pullSecret: "" sshKey: ssh-rsa..... Actual results: failed to generate asset \\\"Platform Quota Check\\\": error(MissingQuota): compute.googleapis.com/cpus is not available in us-east1 because The required number of resources (114688) is more than the limit of 2400\"\nlevel=fatal msg=\"bootstrap host address and at least one control plane host address must be provided\"\n" installID=cz4gg5hr Expected results: The cluster installs successfully Potential cause: The cause is most likely in the installer quota check: https://github.com/openshift/installer/blob/5972c875c3ef1cd13c43c52b7cda2660efdda1b3/pkg/asset/quota/gcp/gcp.go#L164 As we use custom machine types (name: custom-4-16384) The quota check assumes the second number in the string is the number of CPUs, but in our case it is not. It extracts this second number as CPU count here: https://github.com/openshift/installer/blob/5972c875c3ef1cd13c43c52b7cda2660efdda1b3/pkg/asset/quota/gcp/gcp.go#L199 and this is multiplied with the count of machines (compute: 4, controlPlane: 7) which results in 7*16384 = 114688 CPUs.