langgraph: seg fault #20
Comments
|
ok it's def loguru mixed w/ the enqueue setting. I just need to find a way to disable the intercept logger. |
|
Ok it looks like langsmith introduced the multiprocessing module in this merge langchain-ai/langsmith-sdk@0243f79, it's possible that reverting to a version before this avoids everything. |
|
Delgan/loguru#1264 |
this is a very good point to realize Delgan/loguru#108 (comment) |
great insight, i'm new to multiprocessing module Delgan/loguru#108 (comment) |
|
also good info re: understanding langraph_api module https://langchain-ai.github.io/langgraph/concepts/langgraph_server/#cron-jobs old rendered Dockerfile source from Docker Desktop: |
|
ok https://medium.com/@muh.bazm/how-i-unified-logging-in-fastapi-with-uvicorn-and-loguru-6813058c48fc |
Got the following message when I tried to talk to my graph:
Current thought process is the following (thx perplexity):
NOTE: Here is the potentially problematic file: https://github.com/langchain-ai/langsmith-sdk/blob/9a2e4961e389f4bdbe4b822c193e0b917e71d651/python/langsmith/_internal/_background_thread.py#L40C62-L40C71
Inside a Docker container, the behavior of
os.cpu_count()(which is typically used to get the CPU count) can be misleading. Python'scpu_count()function usually reports the number of CPUs available on the host machine, not the number of CPUs allocated to the container[5][8]. This means that usingcpu_count()directly in your ThreadPoolExecutor configuration might not accurately reflect the resources available to your container.To address this issue, you have a few options:
Use container-aware methods:
Instead of relying on
cpu_count(), you can use container-aware methods to determine the number of CPUs allocated to your container. For example:This method reads the CPU quota and period from the cgroup filesystem, which reflects the container's CPU allocation[5].
Use environment variables:
You can pass the number of CPUs as an environment variable when running your Docker container and use that in your code:
Then, when running your container:
Hardcode the value:
If you know the exact number of CPUs you'll allocate to your container, you can hardcode this value:
Remember that the optimal number of threads doesn't always directly correlate with the number of CPUs, especially for I/O-bound tasks. You might need to experiment to find the best configuration for your specific use case[7].
Also, keep in mind that if you're using the
--cpuset-cpusflag when running your Docker container, this will limit which CPUs your container can use, but it won't change whatos.cpu_count()reports inside the container[9].Citations:
[1] https://ppl-ai-file-upload.s3.amazonaws.com/web/direct-files/39473573/54ebe82e-61c3-4079-ae0b-c388be238b27/paste.txt
[2] https://stackoverflow.com/questions/76415968/docker-container-using-only-one-cpu-core/76416074
[3] moby/moby#43587
[4] https://stackoverflow.com/questions/56195679/why-is-threadpoolexecutors-default-max-workers-decided-based-on-the-number-of-c
[5] python/cpython#80235
[6] https://www.tutorialspoint.com/how-many-cpus-does-a-docker-container-use
[7] https://superfastpython.com/threadpoolexecutor-number-of-threads/
[8] https://www.geeksforgeeks.org/python-os-cpu_count-method/
[9] https://stackoverflow.com/questions/47545960/how-to-check-the-number-of-cores-used-by-docker-container/51998044
The text was updated successfully, but these errors were encountered: