Troubleshooting Arize AI: Can't See Traces in Phoenix for LLM Application

Gurdeep S. · 2025-06-12T08:07:34.461Z

Hello Team, I am new to Arize AI and exploring Phoenix for tracing our LLM application. We have added traces in the code and deployed on AWS ECS. I am able to startup logs as below. However, not able to see the traces on https://app.phoenix.arize.com/ 12 June 2025 at 11:02 (UTC+5:30)| Phoenix Project: saphire-evalbluekaktus-ai-chats 12 June 2025 at 11:02 (UTC+5:30)| Span Processor: SimpleSpanProcessorbluekaktus-ai-chats 12 June 2025 at 11:02 (UTC+5:30)| Collector Endpoint: https://app.phoenix.arize.com bluekaktus-ai-chats 12 June 2025 at 11:02 (UTC+5:30)| Transport: HTTP + protobuf

13 comments

· Sorted by Oldest

RunLLM
APP
·
If your logs show the Phoenix collector endpoint as `https://app.phoenix.arize.com` but traces are not visible in the UI, first ensure you are sending traces to the correct endpoint: it must be `https://app.phoenix.arize.com/v1/traces` for OTLP HTTP ingestion. Also, make sure your API key is set in the `PHOENIX_CLIENT_HEADERS` environment variable as `api_key=<YOUR_API_KEY> ` and that `PHOENIX_COLLECTOR_ENDPOINT` is set to `https://app.phoenix.arize.com` (<https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces-1.md|docs>, <https://github.com/Arize-ai/phoenix/blob/main/tutorials/hosted_phoenix/hosted_phoenix_llamaindex_tutorial.ipynb|hosted tutorial>). If you are still not seeing traces, double-check that your exporter is configured to use the `/v1/traces` path, and that your API key is valid and has project access. Example setup: ``` import os os.environ["PHOENIX_CLIENT_HEADERS"] = "api_key=<YOUR_API_KEY> " os.environ["PHOENIX_COLLECTOR_ENDPOINT"] = "https://app.phoenix.arize.com" ``` If using OpenTelemetry directly, set the OTLP endpoint to `https://app.phoenix.arize.com/v1/traces`. Would you like more troubleshooting steps? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces-1.md|Phoenix Tracing Quickstart> - <https://github.com/Arize-ai/phoenix/blob/main/tutorials/hosted_phoenix/hosted_phoenix_llamaindex_tutorial.ipynb|Hosted Phoenix LlamaIndex Tutorial>
💡 Hint: Mention RunLLM in the thread for followups.
Gurdeep S.
·
Hi 🔒[private user], I tried both ways. Still not able to see the traces on the application. I am adding the code snippet for your reference. Kindly assist.

Dustin N.

hi Gurdeep S. can you try:

tracer_provider = register(
    project_name="saphire-eval",
       endpoint="https://app.phoenix.arize.com/v1/traces",
    auto_instrument=False,
    set_global_tracer_provider=True,
    batch=True,
)

Gurdeep S.
·
hi Dustin N. i have tried your suggested code but still not able to see the traces. Attaching the code and logs
Dustin N.
·
you can set the auth directly in the register method, I'm curious to see if it works without overriding the span processor
Gurdeep S.
·
hi Dustin N., I am still facing the issue.
Dustin N.
·
you're still overriding the batch span processor there
Gurdeep S.
·
hi Dustin N., I tried everything. But not able to see the traces. Attaching the screenshot for your reference -
Dustin N.
·
hi Gurdeep S. I don't work on cloud specifically, but PHOENIX_COLLECTOR_ENDPOINT is more generic than the more specific endpoint argument that you pass to register. If passing a URL specifically to register, you'll need to pass in the fully qualified OLTP endpoint you plan on using, which ends in v1/traces for Phoenix cloud. I know that you might need a "space" suffix and ID as well, though I'm not sure if you currently have one. Kiko C. or Anthony P. should be able to tell you As far as I understand your endpoint should end up looking like: endpoint="https://app.phoenix.arize.com/v1/traces" OR endpoint="https://app.phoenix.arize.com/s/{SPACE_ID}/v1/traces"
Anthony P.
·
What LLM library are you attempting to instrument?
Anthony P.
·
Anthony P.
·
I am able to send traces to phoenix cloud using the above setup
Kiko C.
·
Hey Gurdeep S., I have a proposition, we're improving Phoenix Cloud and we are currently under a beta program. I can get you in it, and huddle with you to get your traces in. How does that sound?

Troubleshooting Arize AI: Can't See Traces in Phoenix for LLM Application | Arize AI Community

RunLLM
APP
·
If your logs show the Phoenix collector endpoint as `https://app.phoenix.arize.com` but traces are not visible in the UI, first ensure you are sending traces to the correct endpoint: it must be `https://app.phoenix.arize.com/v1/traces` for OTLP HTTP ingestion. Also, make sure your API key is set in the `PHOENIX_CLIENT_HEADERS` environment variable as `api_key=<YOUR_API_KEY> ` and that `PHOENIX_COLLECTOR_ENDPOINT` is set to `https://app.phoenix.arize.com` (<https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces-1.md|docs>, <https://github.com/Arize-ai/phoenix/blob/main/tutorials/hosted_phoenix/hosted_phoenix_llamaindex_tutorial.ipynb|hosted tutorial>). If you are still not seeing traces, double-check that your exporter is configured to use the `/v1/traces` path, and that your API key is valid and has project access. Example setup: ``` import os os.environ["PHOENIX_CLIENT_HEADERS"] = "api_key=<YOUR_API_KEY> " os.environ["PHOENIX_COLLECTOR_ENDPOINT"] = "https://app.phoenix.arize.com" ``` If using OpenTelemetry directly, set the OTLP endpoint to `https://app.phoenix.arize.com/v1/traces`. Would you like more troubleshooting steps? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces-1.md|Phoenix Tracing Quickstart> - <https://github.com/Arize-ai/phoenix/blob/main/tutorials/hosted_phoenix/hosted_phoenix_llamaindex_tutorial.ipynb|Hosted Phoenix LlamaIndex Tutorial>
💡 Hint: Mention RunLLM in the thread for followups.
Gurdeep S.
·
Hi 🔒[private user], I tried both ways. Still not able to see the traces on the application. I am adding the code snippet for your reference. Kindly assist.

Dustin N.

hi Gurdeep S. can you try:

tracer_provider = register(
    project_name="saphire-eval",
       endpoint="https://app.phoenix.arize.com/v1/traces",
    auto_instrument=False,
    set_global_tracer_provider=True,
    batch=True,
)

Gurdeep S.
·
hi Dustin N. i have tried your suggested code but still not able to see the traces. Attaching the code and logs
Dustin N.
·
you can set the auth directly in the register method, I'm curious to see if it works without overriding the span processor
Gurdeep S.
·
hi Dustin N., I am still facing the issue.
Dustin N.
·
you're still overriding the batch span processor there
Gurdeep S.
·
hi Dustin N., I tried everything. But not able to see the traces. Attaching the screenshot for your reference -
Dustin N.
·
hi Gurdeep S. I don't work on cloud specifically, but PHOENIX_COLLECTOR_ENDPOINT is more generic than the more specific endpoint argument that you pass to register. If passing a URL specifically to register, you'll need to pass in the fully qualified OLTP endpoint you plan on using, which ends in v1/traces for Phoenix cloud. I know that you might need a "space" suffix and ID as well, though I'm not sure if you currently have one. Kiko C. or Anthony P. should be able to tell you As far as I understand your endpoint should end up looking like: endpoint="https://app.phoenix.arize.com/v1/traces" OR endpoint="https://app.phoenix.arize.com/s/{SPACE_ID}/v1/traces"
Anthony P.
·
What LLM library are you attempting to instrument?
Anthony P.
·
Anthony P.
·
I am able to send traces to phoenix cloud using the above setup
Kiko C.
·
Hey Gurdeep S., I have a proposition, we're improving Phoenix Cloud and we are currently under a beta program. I can get you in it, and huddle with you to get your traces in. How does that sound?