Tracking Phoenix Metrics for Health and Trace Monitoring

Zhuyi X. · 2025-05-13T18:22:01.608Z

Does phoenix emit any metrics, e.g. • how many traces are created for each project in a time series format in grafana • or metrics that can help check if phoenix is running in a healthy state. is there a way to do so?

Zhuyi X.

·May 13, 2025 06:22 PM

Does phoenix emit any metrics, e.g.

how many traces are created for each project in a time series format in grafana
or metrics that can help check if phoenix is running in a healthy state.

is there a way to do so?

6 comments

· Sorted by Oldest

RunLLM
APP
·
Phoenix exports some metrics via a Prometheus endpoint, which can be scraped for health and operational metrics, but there is no built-in integration with Grafana, nor does Phoenix natively provide time series metrics like "traces per project" in a Grafana-ready format. For trace counts and similar statistics, Phoenix's UI and API provide trace overviews, but not direct Grafana dashboards or time series exports for external visualization tools. According to a trusted source, there is no official Grafana integration for Phoenix at this time, but Prometheus metrics are available for health checks and basic monitoring if you scrape the endpoint from the Phoenix container deployment (Slack discussion, Phoenix Dockerfile, Support Staff Knowledge).For time series trace counts per project, Phoenix does not currently expose this as a Prometheus metric or Grafana-ready data source. However, feature requests and issues have been raised to add time series and quantile metrics for traces and latency in the Phoenix UI and API, but these are not yet available for external dashboards (GitHub Issue: trace count, GitHub Issue: timeseries graph). Would you like more detail on how to access Prometheus metrics or alternative ways to monitor Phoenix health?Sources: - Slack discussion: Phoenix metrics and Prometheus - GitHub Issue: trace count - Support Staff Knowledge - GitHub Issue: timeseries graph
💡 Hint: Mention RunLLM in the thread for followups.
Xander S.
·
Hey Zhuyi X., our OpenInference instrumentation does not currently emit metrics, and Phoenix does not support ingesting OTel metrics.
Zhuyi X.
·
I see, thanks! Do you have any advice for how to monitor the status of the phoenix server?
Xander S.
·
Oh, I see. You mean metrics from Phoenix to monitor Phoenix. We do have the ability to emit Prometheus metrics from Phoenix when PHOENIX_ENABLE_PROMETHEUS is set to true. You can then scrape Prometheus metrics with config like the following:
prometheus.io/path: /metrics prometheus.io/port: "9090" prometheus.io/scrape: "true"
1
Zhuyi X.
·
it looks prometheus.io/path: /metrics unneeded, i.e. both
curl localhost:9090 curl localhost:9090/metrics
work, is that expected?
👍1
Xander S.
·
Glad it worked!

RunLLM
APP
·
Phoenix exports some metrics via a Prometheus endpoint, which can be scraped for health and operational metrics, but there is no built-in integration with Grafana, nor does Phoenix natively provide time series metrics like "traces per project" in a Grafana-ready format. For trace counts and similar statistics, Phoenix's UI and API provide trace overviews, but not direct Grafana dashboards or time series exports for external visualization tools. According to a trusted source, there is no official Grafana integration for Phoenix at this time, but Prometheus metrics are available for health checks and basic monitoring if you scrape the endpoint from the Phoenix container deployment (Slack discussion, Phoenix Dockerfile, Support Staff Knowledge).For time series trace counts per project, Phoenix does not currently expose this as a Prometheus metric or Grafana-ready data source. However, feature requests and issues have been raised to add time series and quantile metrics for traces and latency in the Phoenix UI and API, but these are not yet available for external dashboards (GitHub Issue: trace count, GitHub Issue: timeseries graph). Would you like more detail on how to access Prometheus metrics or alternative ways to monitor Phoenix health?Sources: - Slack discussion: Phoenix metrics and Prometheus - GitHub Issue: trace count - Support Staff Knowledge - GitHub Issue: timeseries graph
💡 Hint: Mention RunLLM in the thread for followups.
Xander S.
·
Hey Zhuyi X., our OpenInference instrumentation does not currently emit metrics, and Phoenix does not support ingesting OTel metrics.
Zhuyi X.
·
I see, thanks! Do you have any advice for how to monitor the status of the phoenix server?
Xander S.
·
Oh, I see. You mean metrics from Phoenix to monitor Phoenix. We do have the ability to emit Prometheus metrics from Phoenix when PHOENIX_ENABLE_PROMETHEUS is set to true. You can then scrape Prometheus metrics with config like the following:
prometheus.io/path: /metrics prometheus.io/port: "9090" prometheus.io/scrape: "true"
1
Zhuyi X.
·
it looks prometheus.io/path: /metrics unneeded, i.e. both
curl localhost:9090 curl localhost:9090/metrics
work, is that expected?
👍1
Xander S.
·
Glad it worked!