Lower the system metrics logging frequency to reduce MLflow server's load

[chenmoneygithub]

What does this PR do?

Lower the system metrics logging frequency to reduce MLflow server's load. System metrics logging is using a different async logging queue from normal metrics, so the time buffering doesn't apply to system metrics. This design was by intention so that MLflow can terminate system metrics separately. 30s is the logging frequency of wandb, so there should be no user experience regression.

What issue(s) does this change relate to?

Before submitting

• Have you read the contributor guidelines?
• Is this change a documentation change or typo fix? If so, skip the rest of this checklist.
• Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• Did you update any related docs and document your change?
• Did you update any related tests and add any new tests related to your change? (see testing)
• Did you run the tests locally to make sure they pass?
• Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[chenmoneygithub]

@mvpatel2000 please let me know if you prefer to have it be a configurable argument, thanks!

[mvpatel2000]

LGTM. Every 30 seconds seems a bit infrequent... what about 10s?

[chenmoneygithub]

@mvpatel2000 Thanks for reviewing! I am shortening the gap to 15s, which matches the wandb behavior: Lower the system metrics logging frequency to reduce MLflow server's load