ericyangliu opened a new issue, #4802:
URL: https://github.com/apache/polaris/issues/4802

   ### Is your feature request related to a problem? Please describe.
   
   Polaris exposes HTTP server request duration metrics through 
Micrometer/Prometheus, but the exported HTTP timer metrics do not include 
histogram bucket series such as `http_server_requests_seconds_bucket`.
   
   Without bucket series, Prometheus-compatible backends cannot calculate 
aggregable p95/p99 latency with `histogram_quantile`. Operators can see request 
count, total duration, and max duration, but cannot build percentile-based 
dashboards or alerts for Polaris API latency.
   
   Observed with Polaris 1.5.0 on Kubernetes using the management metrics 
endpoint on port 8182.
   
   Available series include:
   
   - `http_server_requests_seconds`
   - `http_server_requests_seconds_count`
   - `http_server_requests_seconds_sum`
   - `http_server_requests_seconds_max`
   
   Expected series for percentile monitoring would include:
   
   - `http_server_requests_seconds_bucket`
   
   ### Describe the solution you would like
   
   Polaris should provide a supported way to publish Prometheus histogram 
buckets for HTTP server request duration metrics.
   
   The behavior should be configurable so operators can enable histogram 
buckets when they need percentile-based monitoring, while deployments that do 
not need the extra series can keep the current cardinality profile.
   
   ### Describe alternatives you have considered
   
   Using `http_server_requests_seconds_sum` and 
`http_server_requests_seconds_count` only supports average latency.
   
   Using `http_server_requests_seconds_max` does not provide p95/p99 latency 
and is too sensitive to individual outliers.
   
   Calculating p95/p99 downstream is not possible unless Polaris emits bucketed 
histogram data at the source.
   
   ### Additional context
   
   This is needed for production monitoring and alerting of Polaris API latency 
in Prometheus-compatible systems such as Prometheus, Mimir, or Grafana Cloud.
   
   I searched existing Polaris issues for `_bucket`, `histogram`, 
`http_server_requests_seconds`, `Micrometer`, `Prometheus metrics`, 
`percentile`, `quantile`, and `latency metrics`. I found related metrics 
issues, but not this specific request.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to