Skip to content

Commit eebe586

Browse files
authored
test(e2e): run tenant workers on one vCPU with CPU headroom above it (#3863)
## What this PR does Moves the e2e tenant workers from two vCPUs to one, keeping a CPU limit above the vCPU count: `resources.cpu: 1` with `podCpuLimit: 2` and `podCpuRequest: 100m`. This is the next step of the #3513 measurement series. With two vCPUs the failing worker's two vCPU threads burn 1.4 to 1.8 cores between them while the guest kernel never gets past SMP bringup, and raising the CFS ceiling from 2 to 3 cores (#3862) changed neither the burn nor the progress, so the spin is between the two vCPUs rather than against the quota. A single-vCPU guest has no sibling vCPU to spin on, which is exactly what this shape tests. The ceiling of 2 keeps the QEMU emulator and IO threads from eating into the one core the guest computes with, which was the failure shape of the original single-vCPU workers whose quota equalled their vCPU count. The request of 100m is what KubeVirt derived for a one-vCPU worker before, so scheduling does not move. Like its predecessors this changes only what the e2e suites run, not what any chart ships by default. ### Screenshots Not applicable, no UI change. ### Downstream repositories The diff is one values block inside `hack/e2e-chainsaw/_lib/run-kubernetes.sh`. Nothing in the trigger map is touched. - [x] No downstream repository is affected by this change - [ ] [cozystack/website](https://github.com/cozystack/website) - follow-up: - [ ] [cozystack/terraform-provider-cozystack](https://github.com/cozystack/terraform-provider-cozystack) - follow-up: - [ ] [cozystack/ansible-cozystack](https://github.com/cozystack/ansible-cozystack) - follow-up: - [ ] [cozystack/ccp](https://github.com/cozystack/ccp) - follow-up: - [ ] [cozystack/talm](https://github.com/cozystack/talm) - follow-up: - [ ] [cozystack/cozyhr](https://github.com/cozystack/cozyhr) - follow-up: - [ ] [cozystack/cozy-proxy](https://github.com/cozystack/cozy-proxy) - follow-up: - [ ] [cozystack/cozystack-telemetry-server](https://github.com/cozystack/cozystack-telemetry-server) - follow-up: - [ ] [cozystack/external-apps-example](https://github.com/cozystack/external-apps-example) - follow-up: - [ ] [cozystack/examples](https://github.com/cozystack/examples) - follow-up: ### Release note ```release-note test(e2e): tenant workers in the kubernetes e2e suites run on one vCPU with a CPU limit of two, testing whether the node-join freeze is a spin between sibling vCPUs ```
2 parents 6735277 + 73b8afc commit eebe586

1 file changed

Lines changed: 11 additions & 11 deletions

File tree

‎hack/e2e-chainsaw/_lib/run-kubernetes.sh‎

Lines changed: 11 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -3167,24 +3167,24 @@ ${ouroboros_addon}
31673167
# ceiling. Workers failing to join (cozystack/cozystack#3513) have been
31683168
# measured burning their vCPU threads flat out at a ceiling equal to their
31693169
# vCPU count while the guest kernel made no progress, which is what a vCPU
3170-
# spinning on a lock looks like when the vCPU holding that lock is the one
3171-
# preempted out of the quota they share. Whether the spare CPU lets the
3172-
# preempted one back in is what this is here to answer. The request is
3170+
# spinning on a lock looks like. Spare quota above the pair did not change
3171+
# it: under a ceiling of 3 the pair kept burning the same 1.4 to 1.8 cores
3172+
# with no progress, so the spin is not a quota artefact. A single vCPU
3173+
# cannot spin on its sibling, which is what this shape is here to answer,
3174+
# and the ceiling above the vCPU count keeps the emulator and IO threads
3175+
# from eating into the one core the guest computes with. The request is
31733176
# what keeps scheduling where it was: setDefaultResourceRequests, in
31743177
# KubeVirt's VMI mutating webhook, copies a declared CPU limit into an
31753178
# undeclared CPU request, so the limit alone would have every worker ask
31763179
# the scheduler for its whole ceiling and sit Pending on Insufficient cpu.
31773180
# Its value is the one KubeVirt derived for these workers before, from the
31783181
# vCPU count and the cluster CPU allocation ratio.
3179-
podCpuLimit: 3
3180-
podCpuRequest: 200m
3181-
# Two vCPUs at the memory the workers had before. Sized by resources
3182-
# rather than by instanceType: u1.large doubles memory along with the
3183-
# vCPUs, and four 8Gi workers do not fit beside the rest of the suite,
3184-
# so the second worker of each cluster stayed Pending on Insufficient
3185-
# memory and the node group never reached two Ready nodes.
3182+
podCpuLimit: 2
3183+
podCpuRequest: 100m
3184+
# One vCPU at the memory the workers had before. Sized by resources
3185+
# rather than by instanceType, which cannot carry the pod CPU pair above.
31863186
resources:
3187-
cpu: 2
3187+
cpu: 1
31883188
memory: 4Gi
31893189
roles:
31903190
- ingress-nginx

0 commit comments

Comments
 (0)