Readiness check for uwsgi using its stats socket
Use this check to mark a pod running uwsgi as NotReady whenever the uwsgi queue crosses a threshold.
This will allow uwsgi to finish processing requests in its queue before Kubernetes sends more traffic to it
Because the readiness check needs to run as part of a pod, in order to have functional test we deploy a small python+uwsgi application that sleeps for a few seconds + a second python "client" script that makes requests. Since the application sleeps upon a request and new requests are constantly being sent, uwsgi will eventually queue and we'll readiness checks should fail.
Kind installation Skaffold installation Kubectl installation
With kind, kubectl and skaffold installed, simply run:
skaffold dev
Skaffold builds a docker image, loads into the kind "cluster" and creates the Pods, watches for changes in the filesystem and automatically re-deploys when needed.
The example below will prevent traffic from hitting the Pod once the queue reaches 70% its max capacity:
readinessProbe:
exec:
command:
- uwsgi-is-ready
- --stats-socket
- /tmp/uwsgi-stats
- --queue-threshold
- 0.7
failureThreshold: 2
initialDelaySeconds: 5
periodSeconds: 5
successThreshold: 1
timeoutSeconds: 1
You can find a complete example in k8s/deployment.yaml
In the container image used by your Pod, run:
pip install uwsgi-readiness-check