Details
-
Bug
-
Resolution: Done
-
Undefined
-
None
-
False
-
None
-
False
-
NEW
-
VERIFIED
-
Logging (LogExp) - Sprint 215
Description
Some of the deployments created by loki-operator currently produce errors when kube-probe does the readiness or liveness probe, for example:
level=warn ts=2022-03-08T11:38:34.155974618Z caller=logging.go:61 msg="GET /metrics 707.684µs, error: write tcp 10.244.0.11:3100->10.244.0.1:35664: write: connection reset by peer ws: false; Accept: */*; Connection: close; User-Agent: kube-probe/1.21; " level=warn ts=2022-03-08T11:38:34.155992326Z caller=logging.go:61 msg="GET /metrics 757.453µs, error: write tcp 10.244.0.11:3100->10.244.0.1:35662: write: broken pipe ws: false; Accept: */*; Connection: close; User-Agent: kube-probe/1.21; "
While these errors do not seem to be producing any errors on Kubernetes' side they are unnecessary output.
The error seems to be related to kube-probe only accepting a limited response size and closing the connection after that. This should be fixable by using the /ready endpoint for determining readiness instead.