request
set. Kubernetes uses kernel throttling to implement CPU limits, and we currently don't set CPU limits
. If an application goes above the limit, it gets throttled (aka fewer CPU cycles). CPU requests
are used during scheduling of pods.OOMKilled
.prometheus-node-exporter
- exporter for machine metricstestground-sidecar
- the daemon that monitors containers on a given host and executes network updates, traffic shaping, etc. for test plan instancesgenie-plugin
- Kubernetes CNI plugin for choosing pod network during deployment timeflannel
- the control network daemonkube-proxy
- the Kubernetes network proxyweave
- the data network daemonc5.2xlarge
worker node instance type)master
node, usually it takes ~5min. for all 10k test plan instances (pods) to be scheduled on a given work node and to be in Running
state. Depending on various factors (i.e. number of hosts, whether the test plan image is already downloaded, etc.) we get around 30-50 pods scheduled per second for a given test plan run.sync service
is an integral part of Testground, used to communicate out-of-band information between test plan instances (such as endpoint addresses, peer ids, etc.)