Hi, When I have make requests on UI, on my self host phoenix instance, I have timeout issue. When I make request with api and specify timeout - everything works fine, Where can I change setting for timeout, for UI phoenix application?
There is currently no documented setting to directly increase the UI request timeout for the Phoenix web application. Timeout issues in the UI are typically due to backend query or worker timeouts, not a configurable UI-side timeout. You can increase backend timeouts for API requests (e.g., using the Python client), but the UI itself does not expose a timeout configuration parameter in the official documentation or environment variables [source].If you are experiencing timeouts with large datasets or long-running queries in the UI, consider optimizing your queries, increasing backend worker timeouts (if available), or upgrading to a newer Phoenix version, as some releases have improved handling for large datasets and timeouts [source]. Would you like more detail on backend timeout configuration or workarounds?Sources: - Phoenix GitHub Issue: Loading large datasets timeouts - https://arize.com/docs/phoenix/self-hosting/configuration
💡 Hint: Mention RunLLM in the thread for followups.
To increase backend timeouts in Phoenix, you generally need to adjust the timeout settings in the worker or API client making the requests. For API calls, you can specify a higher timeout value in methods like get_trace_dataset(timeout=...) or query_spans(timeout=...) [source]. For experiment runs, recent Phoenix versions allow setting a worker timeout parameter, but this is not exposed for all operations and may require upgrading to a newer release [source].There is no single global backend timeout environment variable documented for the Phoenix server itself. If you need to handle long-running queries, increase the timeout in your client code or look for a timeout parameter in the relevant API or worker configuration. Would you like step-by-step instructions for a specific API or deployment method?Sources:
- https://github.com/Arize-ai/phoenix/issues/7991
- https://github.com/Arize-ai/phoenix/issues/7486
💡 Hint: Mention RunLLM in the thread for followups.
Hi, 🔒[private user] I am trying to load spans, As I understand, Default Timeout is 5 sec. In api client :
df_all_spans = client.get_spans_dataframe(project_name=project_name, timeout=60)Works fine, but
df_all_spans = client.get_spans_dataframe(project_name=project_name)has same issue, as UI has, I am using default Phoenix instance with default phoenix UI.
okay got it. I'll go ahead and try to reproduce this error. In the meantime could you try: (1) hard refresh / incognito (cache mismatch can cause JSON.parse errors), (2) restart the Phoenix container, (3) upgrade to latest Phoenix, and (4) filter to a smaller time window before loading spans (default pulls all spans) & let me know if the issue is still persistent & we'll try to see what else we can do!
They good approaches, but not for me. Example, I need pull all spans. How can I increase timeouts value?
Hey Andrei L., I was also facing similar issues and you will face same again with your solution when your project will have huge number of traces.
df_all_spans = client.get_spans_dataframe(project_name=project_name)what worked out for me :
set timeout to some huge number
use query filter : SpanQuery().where(....)
use start_time and end_time with very small interval and in batch rather than all in one go
use limit
the UI doesn’t have a timeout per se: it’ll just wait for as long as it needs, so the problem may be something else. Can you tell us which page that you were trying to access that was having a timeout issue resulting in the “Something went wrong” screenshot above? The screenshot seems to be about a parsing issue, which means the server has returned something for it to parse, but if it’s a timeout issue then the server shouldn’t have anything to return. (In past situations, there had been other people experiencing UI timeout, but that was because they had a proxy or load balancer that was timing out the connection)
