[elasticsearch] set cpu request = cpu limit #458
Conversation
jmlrt
force-pushed
the
request-equals-limits
branch
from
9d3a952
to
c566822
Compare
|
jenkins test this please |
|
Upgrade test is failing when changing the cpu request: https://devops-ci.elastic.co/job/elastic+helm-charts+pull-request+integration-elasticsearch/254/ES_SUITE=upgrade,KUBERNETES_VERSION=1.13,label=docker&&virtual/ I'll investigate next week. |
|
jenkins test this please |
1 similar comment
|
jenkins test this please |
By increasing elasticsearch cpu request in c566822, n1-standard-8 don't provide enough cpu for the tests
| @@ -39,7 +39,7 @@ variable "additional_zones" { | |||
|
|
|||
| variable "machine_type" { | |||
| description = "Machine type for the kubernetes nodes" | |||
| default = "n1-standard-8" | |||
| default = "custom-10-30720" | |||
There was a problem hiding this comment.
The fact that this increase is causing issues for our testing environment might be a sign that this should be announced as a breaking chance in the release notes. At least breaking for anybody using the current default resource limits.
There was a problem hiding this comment.
In our testing environment we are deploying 21 Elasticsearch pods (7 scenarios x 3 nodes) per clusters and we can assume that the pod are never requiring more resources than set in requests resources during the tests.
This change cause resource needed for Elasticsearch pods grow from 2.1 CPU to 21 CPU in our testing environment.
IMHO standard usage is closer to 3 long running Elasticsearch pods per cluster which are consuming CPU and memory limits, so I'm not sure that the impact should be considered as a breaking change. WDYT?
There was a problem hiding this comment.
The impact was big enough to cause us issues in our testing environment because of resource capacity. A 10x increase in the amount of CPUs requested by default could also cause issues for others too. For clusters with autoscaling enabled they might suddenly end up with 10 times the amount of Kubernetes node running after a version bump.
I don't think the change is big enough to avoid doing it, but I do think it would be nice to give all users a heads up including documentation for how to keep the existing configuration.
There was a problem hiding this comment.
That sounds good to me, I'll merge the PR and will add a note in next release's changelog about this change and how to keep existing config.
${CHART}/tests/*.py${CHART}/examples/*/test/goss.yamlRelated to #409 (comment)
I think we can let cpu request to 100m in "workstation" examples like docker-for-mac example.
What about default values for Filebeat, Metricbeat, Kibana and Logstash should we also stay with the request=limit rule?