Alexey Serbin created KUDU-3514:
-----------------------------------
Summary: Introduce per-RPC metric of responses sent past the
client-defined deadline
Key: KUDU-3514
URL: https://issues.apache.org/jira/browse/KUDU-3514
Project: Kudu
Issue Type: Improvement
Components: master, server, tserver
Reporter: Alexey Serbin
Assignee: Alexey Serbin
As of Kudu 1.17.0, there is per-RPC histogram for processing times of remote
calls: {{handler_latency_xxx}}, where the {{xxx}} suffix stands for the RPC
interface and method name. There is also service-wide
{{rpc_incoming_queue_time}} historgram.
However, even using these two metrics above it's hard to definitely tell how
many calls of a particular RPC method timed out while being processed by a Kudu
server, even if the client-defined timeout is known, fixed, and the same for
all the RPCs processed by a Kudu server.
It would be nice to introduce a new per-RPC counter to accumulate stats on the
responses that have been sent back to the client side already past the
deadline. That would help troubleshooting of particular scenarios and provide
a glimpse of what part of requests to a particular RPC method timed out while
being processed.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
