When using local repo no progress is made on the parsing

[WTK]

I've followed the instructions to setup local potpie and pointed it to local repo to parse it. The project is created in the database with "status=submitted" but then nothing happens, I see no errors, but it seems like repository is not processed.

How can I debug this? How can I verify if the "parser" is running?

[dhirenmathur]

Hey @WTK , the parsing happens on the celery worker, can you please confirm that the celery worker is up, you should see a log like celery@192.168.1.6 ready.

[WTK]

@dhirenmathur It's there it seems - [2025-02-23 11:19:43,312: INFO/MainProcess] celery@Mac ready.

A little update, 1st issue was on my side - I have another stack running locally, so I've changed REDISPORT to be 6380, but I didn't update BROKER_URL to also use that number (so I guess it was using another of my redises).

Now that was fixed I started seeing some messages in celery, but only ones similar to [2025-02-23 13:10:05,620: DEBUG/MainProcess] pidbox received method hello(from_node='celery@ip-192-168-255-14.ec2.internal', revoked={}) [reply_to:{'exchange': 'reply.celery.pidbox', 'routing_key': '25c81e34-f92f-3023-9060-70b2f28c502a'} ticket:26271ee0-c962-452b-a714-f36232ec98c9] -- no other celery tasks afaics.

And one more thing -- neo4j was crashing for me (I'm on Apple M4 Pro), I had to add JAVA_TOOL_OPTIONS: "-XX:UseSVE=0" in the environment it (docker-compose.yaml).

I also had to change a line in inference_service.py self.embedding_model = SentenceTransformer("all-MiniLM-L6-v2") to self.embedding_model = SentenceTransformer("all-MiniLM-L6-v2", device='cpu'), due to issues with M4 without that.

Update 2: after all that I restarted using start.sh and it started processing! 🎉

[WTK]

I can't get it to work in the end. The last error I'm getting is that node_ids is None (when it's expected to be list), exception originating here

potpie/app/modules/intelligence/agents/agents/code_gen_agent.py

Line 281 in d9c4f01

if len(node_ids) > 0:

My guess is that this node_ids is holding some pieces of "knowledge graph" but they're empty?

[dhirenmathur]

That is weird @WTK, that agent usually works well without node ids passed to it.

My guess is that this node_ids is holding some pieces of "knowledge graph" but they're empty?

Yes we enforce this for the testing agents but not the others, so I'm not sure why this is failing here, could you please share the full trace?

[WTK]

INFO:app.modules.intelligence.agents.chat_agents.code_gen_chat_agent:messages we will use [AIMessage(content='You can now ask questions about the testme repository.', additional_kwargs={}, response_metadata={}), HumanMessage(content='I would like to add concept of tasks (similar to google calendar events)', additional_kwargs={}, response_metadata={})]
INFO:root:Development mode enabled. Using environment variable for API key.
INFO:root:Initializing Openai LLM
INFO:root:Development mode enabled. Using environment variable for API key.
ERROR:app.modules.intelligence.agents.chat_agents.code_gen_chat_agent:Error in code generation: object of type 'NoneType' has no len()
INFO:app.modules.conversations.conversation.conversation_service:Generated and streamed AI response for conversation 019539e0-198c-7e31-8ce4-e452a95d1b9a for user defaultuser using agent LLD_agent

There's no true stack trace, only this logged message "Error in code generation: object of type 'NoneType' has no len()", but I traced it to the place I linked to above. I'm using main branch, perhaps it's not the one I should try?

Some screenshots

[dhirenmathur]

Hey @WTK, weird - I'm getting an empty list at that point. But you're right, we should atleast have a nullcheck there, we'd appreciate a PR if you could raise one! Thanks!

[WTK]

What could be the reason for that to be None for me? Like how is that list constructed? Perhaps I'm missing some obvious piece?