Open jayunit100 opened 6 years ago
1) Interesting. What I found in common was...... these are all HTTP services. So the missing services for envoy issue is a non-issue.
2) However, I'm still going to suggest that maybe the older services (created before I istio'd my cluster) being present is a probblem.
Starting to think maybe its just a config issue. Not sure though !
@jayunit100:
RE: bytes-in/out for HTTP services
We are in-progress of adding total bytes sent/received for http services (current support is request.size
and response.size
, but this will add headers, etc. to the calculation).
RE: Old services
Are the service definitions for these services still in the API server?
I see them for http services.
I don’t see them for https services.
Thanks for the response btw, just curious if that HTTP above was a typo? Curious what challenges there are for https that raw bytes in and out are significantly different then http. I’m also trying to learn more about Istio internals so, could be my own ignorance.
HTTPS (and other TCP protocols) should be there now (connection.sent.bytes_total
and connection.received.bytes_total
). The difference is that for HTTP, we had been relying on the request/response size attributes, which were including strictly reporting on the body. We are adding to HTTP support by adding attributes that include total size as well.
Does that help?
http://35.184.76.105:3000/d/1/istio-dashboard?refresh=5s&orgId=1 , FYI, this is my istio dashboard. Anything look fishy? I only see metrics for HTTP based services, which definetly makes me think something is wrong w/ HTTPS based metrics?
@jayunit100 one thing I notice is that your HTTP services are showing up as TCP services. I think this may be a configuration issue. Are the ports for those services labeled with the prefix of "http" ?
Hi istio ! Have really enjoyed my initial POC w/ this stuff on GKE !!! Now that its time to get serious and try to create some finer grained information, and I ran into some issues, however.
I beleive this issue is a bug, because it makes it impossible, or extremely cumbersome, to use istio to investigate a single namespace of interest... but if not, I'm happy to close it .
It appears that the istio prometheus / grafana defaults
As an example, here are the metrics in my prometheus instance. many of which correspond to namespaces that were created many months before I put Istio on my cluster: (the hub45_istio metrics are valid, all other ones are unwanted entirely, and presumeably, shouldnt even be spotted by envoy as I never labelled those namespaces).
1) Is this a bug at the envoy level ? I assume that Istio by default would only be reporting metrics for the namespaces I actually cared about / labelled.
2) Note that, in this cases, I've actually also deleted all pods in istio-system, just to see if that fixed things. It didnt seem to.