Add liveness probe to Antrea Controller #1839
Conversation
This is motivated by antrea-io#1837, in which the Antrea Controller got in a bad state (presubmably because of a bug in the apiserver library), and didn't recover. The endpoint was marked as non-Ready for the Antrea Service, which means the APIService became broken, since the Antrea Service has a single Endpoint (single Controller replica). With the liveness probe, the Antrea Controller would have been restarted by kubelet and the APIService would hopefully have recovered. The /healthz endpoint is deprecated since K8s v1.16, so we switch to /readyz and /livez.
antoninbas
force-pushed
the
add-liveness-probe-for-antrea-controller
branch
from
7684406 to
b535177
Compare
Codecov Report
@@ Coverage Diff @@
## main #1839 +/- ##
=======================================
Coverage ? 61.80%
=======================================
Files ? 196
Lines ? 16655
Branches ? 0
=======================================
Hits ? 10294
Misses ? 5270
Partials ? 1091
Flags with carried forward coverage won't be shown. Click here to find out more. |
They're not, but that reminds me I need to update generate-manifest.sh since we have the |
We are relying on apiserver library and not the K8s API of the cluster, so the paths should be there even in K8s 1.15 cluster? |
good point, I can remove my commit :) |
antoninbas
force-pushed
the
add-liveness-probe-for-antrea-controller
branch
from
e752d43 to
b535177
Compare
|
/test-all |
|
/test-e2e |
This is motivated by #1837, in which the Antrea Controller got in a bad
state (presubmably because of a bug in the apiserver library), and
didn't recover. The endpoint was marked as non-Ready for the Antrea
Service, which means the APIService became broken, since the Antrea
Service has a single Endpoint (single Controller replica). With the
liveness probe, the Antrea Controller would have been restarted by
kubelet and the APIService would hopefully have recovered.
The /healthz endpoint is deprecated since K8s v1.16, so we switch to
/readyz and /livez.