Cleaning Up Spans in Phoenix Server Without Restarting Docker

·Mar 12, 2024 12:47 PM

Hi, is there a way of cleaning up the spans stored in my phoenix server without having to restart the server/container? I'm not using px.launch_app , I using the docker image.

👀

1

5 comments

· Sorted by Oldest

Xander S.
·
Hey Arthur M., thanks for using Phoenix! It sounds like you have in mind removing spans that you don't need. Can you tell me more about what kind of span curation you're trying to accomplish? Is it just a matter of reducing the number of spans you need to look at, or are you trying to curate a dataset, for example?
Arthur M.
·
As I develop my LLM app locally with a Docker container hosting a Phoenix service, I'd like to reduce the number of span messages I need to look at, as you pointed out. But thinking about it a bit more now, I'm also concerned about the amount of memory the container will take up when I eventually put this into production. Will I ever have to worry about the memory consumption of traces in my Phoenix container? Thanks for your reply!
👍1
Xander S.
·
Hey Arthur M., thanks for the clarification. At the moment, we don't support a way to selectively remove spans and traces from Phoenix. We do support filtering in the UI to display and query only a subset of spans, but the filter does not permanently remove unselected spans. It might be possible to configure your OTel exporter to filter the kinds of spans you export, although I have not tried that myself. It might result in orphaned sub-spans, since the tree-like structure of traces is reconstructed from the span parent ID. We hear you on the concern about spans consuming too much memory if Phoenix is left running indefinitely in production. We're working on solutions for persisting spans and traces. Stay tuned 🙂
1
Arthur M.
·
That's awesome. Thanks, Xander!
👍1
Swaraj R.
·
+1 to this I was wondering this myself, we'd like to have our px app have some degree of freshness for the spans available Was thinking of using the Phoenix client to export the spans from the server and store it for adhoc explorations on older data on a periodic basis
👍1

Xander S.
·
Hey Arthur M., thanks for using Phoenix! It sounds like you have in mind removing spans that you don't need. Can you tell me more about what kind of span curation you're trying to accomplish? Is it just a matter of reducing the number of spans you need to look at, or are you trying to curate a dataset, for example?
Arthur M.
·
As I develop my LLM app locally with a Docker container hosting a Phoenix service, I'd like to reduce the number of span messages I need to look at, as you pointed out. But thinking about it a bit more now, I'm also concerned about the amount of memory the container will take up when I eventually put this into production. Will I ever have to worry about the memory consumption of traces in my Phoenix container? Thanks for your reply!
👍1
Xander S.
·
Hey Arthur M., thanks for the clarification. At the moment, we don't support a way to selectively remove spans and traces from Phoenix. We do support filtering in the UI to display and query only a subset of spans, but the filter does not permanently remove unselected spans. It might be possible to configure your OTel exporter to filter the kinds of spans you export, although I have not tried that myself. It might result in orphaned sub-spans, since the tree-like structure of traces is reconstructed from the span parent ID. We hear you on the concern about spans consuming too much memory if Phoenix is left running indefinitely in production. We're working on solutions for persisting spans and traces. Stay tuned 🙂
1
Arthur M.
·
That's awesome. Thanks, Xander!
👍1
Swaraj R.
·
+1 to this I was wondering this myself, we'd like to have our px app have some degree of freshness for the spans available Was thinking of using the Phoenix client to export the spans from the server and store it for adhoc explorations on older data on a periodic basis
👍1