Currently, the OSH uses main uWSGI app to serve responses to the
Kubernetes readiness and liveness probes. Unfortunately, this is not
sustainable during load. When all of the uWSGI workers are occupied with
work for longer than the probe timeout, the liveness probe fails as the
request is queued up for too long.
This change proposes alternative solution of running the liveness probes
against an uWSGI stats endpoint which is a lightweight endpoint served
by the master process and is not affected by the workers being busy.
It enables the uWSGI stats server on port 1717 in each of the
relevant pods and updates the deployments to use the port exposed by
those endpoints.
This change allows the deployment to use a liveness port that is
different from the one dynamically looked up in service catalog.
Readiness probes will remain unchanged as it makes sense to check actual
application on start.
Change-Id: Ie466aafeb4edef72ae1591d91a0f1583636a757c
Signed-off-by: Marek Skrobacki <marek.skrobacki@rackspace.co.uk>
9a28908d54