Details
-
Improvement
-
Resolution: Unresolved
-
Major
-
None
-
None
-
0
Description
Currently, we don't have the capability to attribute KV operations to specific clients. Not having this monitoring capability prevents us from diagnosing falling/bad clients effectively.
We currently keep track of the following metrics:
- Connection Creation Time: Timestamp of when a connection is established.
- Last Execution Time: Timestamp of the most recent command execution by a connection.
- Command Dispatch Loop Metrics:
- Minimum and maximum time spent by a connection in the command dispatch loop.
- Total time spent by a connection in the command dispatch loop.
- CPU yield events due to reaching maximum use count (number of attempts or maxed time slice).
- Bytes Sent and Received: Total amount of data transmitted and received by a connection.
This data is currently collected for the lifetime of each connection, additional work is required to make these usable over time. For instance, to distinguish between a connection that was highly active a day ago and just ran a single operation, versus one that ran one operation a day ago and has been consistently busy within the last hour.