Memory usage grows when using StackdriverExporter for Flask app

I am using opencensus-python library to trace requests to my Python Flask application.

I have noticed that when using a StackdriverExporter instance, the memory usage of my app grows steadily for each traced request. When observed using memory_profiler, the memory usage grows roughly 700KB to 1MB per request. I have also noticed similar usage growth on GKE metrics dashboard when running the app on GKE.

All other factors being the same, if I use a LoggingExporter instance (instead of StackdriverExporter), the memory usage stays constant (as expected).

I have put together a sample app along with commands to repro this issue: https://gist.github.com/Amey-D/ebbf7acb77446e81998cf325c7737e73

Any ideas?

[c24t]

Reports of this bug’s death may have been exaggerated. After adding OC instrumentation to the GCP microservices demo app, I’m seeing GKE containers leaking memory. This only happens with the stackdriver exporter, and seems to happen regardless of the transport and sampler.

I don’t know if it’s worth reopening yet, but I’m looking into it now and will let you know what I find.

[Amey-D]

Thanks @c24t for investigating this. Per your suggestion I upgraded the version of opencensus-python package to 0.1.8, and I no longer see the memory leaks.

Side notes:

1. I had to add opencensus-python[stackdriver]==0.1.8 to my requirements.txt file because the newer version treats stackdriver exporter as an extra functionality.
2. I also had to install grpcio separately because apparently the transitive dependency is not handled correctly by the package dependencies.