pubsub: memory leak

[jeanbza]

1. Specify the API at the beginning of the title (for example, "BigQuery: ...")

Pubsub

2. OS type and version

alpine 3.6

3. Python version and virtual environment information python --version

3.6.1

4. google-cloud-python version pip show google-cloud, pip show google-<service> or pip freeze

google-api-core (0.1.4)
google-auth (1.4.1)
google-cloud-pubsub (0.31.0)
googleapis-common-protos (1.5.3)
grpc-google-iam-v1 (0.11.4)

5. Stacktrace if available
6. Steps to reproduce

Stick code provided below on a k8s pod, spam pubsub with messages, watch memory eventually top out at 2.5G and crash the pod.

7. Code example

https://gist.github.com/jadekler/99524a3a8425c45e9299948032dd4e71

(pink is python, green is go - python crashes the pod at 2.5gb, go continues at around 30mb as expected)

[theacodes]

@jadekler your sample keeps a dictionary of seen messages - it expect this is at least contributing to memory growth. Can you remove it or use a bloom filter instead?

[jeanbza]

I bet you're right! I shall do so, sorry for the dummy mistake :)

[jeanbza]

Ill remove it entirely in fact; it's testing something out of scope for this issue.

[theacodes]

It's fine, I don't think it's the sole cause but definitely will contribute to it.

[jeanbza]

Happy to report that using grpcio (1.10.0rc2), I'm not seeing any memory leaks.

Full list of deps:

google-api-core (0.1.4)
google-auth (1.4.1)
google-cloud-pubsub (0.32.0)
googleapis-common-protos (1.5.3)
grpc-google-iam-v1 (0.11.4)
grpcio (1.10.0rc2)