Included Prometheus interceptor support for gRPC streaming#1858
Included Prometheus interceptor support for gRPC streaming#1858RobertSamoilescu merged 6 commits intoSeldonIO:masterfrom
Conversation
sakoush
left a comment
sakoush
left a comment
There was a problem hiding this comment.
LGTM. I left some minor comments.
| ) | ||
|
|
||
| interceptors = [] | ||
| self._interceptors = [] |
There was a problem hiding this comment.
why are we changing it, was that a bug?
There was a problem hiding this comment.
Was not a bug. Just needed a way to access the interceptors list for testing.
| @@ -0,0 +1,129 @@ | |||
| import pytest | |||
There was a problem hiding this comment.
thanks for adding these tests. ideally we also should test other metrics / errors but can happen as a follow up PR.
| async def get_stream_request(request): | ||
| yield request | ||
|
|
||
| # send 10 requests |
There was a problem hiding this comment.
| # send 10 requests | |
| # send 1 requests |
There was a problem hiding this comment.
That should be 10, but I forgot to update the num_request var.
| num_words = len(request_text.split()) | ||
|
|
||
| assert int(counted_requests) == num_requests | ||
| assert int(counted_requests) * num_words == int(counted_responses) |
There was a problem hiding this comment.
question: how are we actually counting the actual words in the responses?
There was a problem hiding this comment.
The model used is a dummy text one which split words by white space. Each word is going to be streamed back by the model.
RobertSamoilescu
force-pushed
the
fix/metrics-streaming
branch
from
0afebdc to
55a271e
Compare
2 participants
This PR includes Prometheus interceptor support for gRPC streaming. Currently for gRPC streaming, we have to set
"metrics_endpoint": null, thus Prometheus logs cannot be scraped. It also updates docs and test for Prometheus interceptor.