fix: Updating the vmsize for e2e cilium to avoid resource scarcity #2014
Conversation
vipul-21
force-pushed
the
singhvipul/ciliume2e
branch
6 times, most recently
from
dbbab4b to
2550c63
Compare
|
/azp run |
|
Azure Pipelines successfully started running 2 pipeline(s). |
vipul-21
force-pushed
the
singhvipul/ciliume2e
branch
from
2550c63 to
81108b4
Compare
vipul-21
changed the title
| @@ -32,7 +32,7 @@ steps: | |||
| mkdir -p ~/.kube/ | |||
| echo "Create AKS Overlay cluster" | |||
| make -C ./hack/swift azcfg AZCLI=az REGION=$(REGION_OVERLAY_CLUSTER_TEST) | |||
| make -C ./hack/swift overlay-no-kube-proxy-up AZCLI=az REGION=$(REGION_OVERLAY_CLUSTER_TEST) SUB=$(SUB_AZURE_NETWORK_AGENT_TEST) CLUSTER=${{ parameters.clusterName }}-$(make revision) | |||
| make -C ./hack/swift overlay-no-kube-proxy-up AZCLI=az REGION=$(REGION_OVERLAY_CLUSTER_TEST) SUB=$(SUB_AZURE_NETWORK_AGENT_TEST) CLUSTER=${{ parameters.clusterName }}-$(make revision) VM_SIZE=Standard_DS4_v2 | |||
There was a problem hiding this comment.
Standard_DS4_v2 is a HUGE jump up in VM size and cost (12x!). I originally picked B2s as the cheapest viable option - did you evaluate any other SKUs? Maybe we could start at Standard_B2ms which has double the mem at only double the cost?
There was a problem hiding this comment.
I tried/tested with Standard_D4_v3 for the perf dashboard( which use goldpinger) that why choose a similar one for our e2e tests.
Will try to test with Standard_B2ms and see if that works.
Thanks !
There was a problem hiding this comment.
The Standard_B2ms worked without the resource limit. @rbtr . Thanks for the help
Will try another run to be sure.
| @@ -76,3 +76,10 @@ spec: | |||
| port: 8080 | |||
| initialDelaySeconds: 5 | |||
| periodSeconds: 5 | |||
| resources: | |||
There was a problem hiding this comment.
The resource block is omitted intentionally so that the goldpinger Pods will overprovision on the nodes. When you add this, only (node mem)/100Mi Goldpinger pods can be scheduled on the Node.
There was a problem hiding this comment.
So if the node mem is 8gb( in case Standard_B2ms) the number of pods that can be schedule on that node would be 80. And we need to scale the pods till 100.
Do you recommend adding a smaller limit to accommodate approx ~110 pods OR remove the limit altogether so that we can overprovision on that node
vipul-21
force-pushed
the
singhvipul/ciliume2e
branch
2 times, most recently
from
7dccfc5 to
13e9c22
Compare
vipul-21
changed the title
|
/azp run |
|
Azure Pipelines successfully started running 2 pipeline(s). |
vipul-21
force-pushed
the
singhvipul/ciliume2e
branch
from
13e9c22 to
4b495eb
Compare
…2014) CI: Testing the e2e test for cilium
…2014) CI: Testing the e2e test for cilium
* fix: assume invalid semver CNI has the required dump state command (#2078) * fix: Updating the vmsize for e2e cilium to avoid resource scarcity (#2014) CI: Testing the e2e test for cilium --------- Co-authored-by: Vipul Singh <[email protected]>
Reason for Change:
Issue Fixed:
Requirements:
Notes: