[change] Reduced network calls in device metric flow #501

[pandafy]

Closes #501

Checks:

• I have manually tested the proposed changes
• I have written new test cases to avoid regressions (if necessary)
• I have updated the documentation (e.g. README.rst)

[pandafy]

The CI is failing due to a mismatch in formatting of timestamp. In the current implementation (master branch), the datetime object gets serialized when passed to celery. Due to this, the time argument of post_metric_write signal is a string. When using UDP for writing timeseries database, the datetime object is not serialized (because the task is executed synchonously) and the formatting of timestamp differs.

Moreover, if we pass a timestamp string to Metric.write method, the post_metric_write signal will use the exact string for time argument which leads to inconsistent results.

I think we should format all incoming timestamps to the ISO format in Metric.write which also includes the timezone information.

[nemesifier]

This looks on the right track to me and I do not see any evident flaw in the code.

I do see 2 areas which I think we can improve: fetching metrics and charts objects now is done each time, both when reading and writing. Charts and metric objects are not written often, I believe they're written only once when they are created. From time to time another metric or chart may be written when new data comes in, therefore I think caching the DB calls to fetch these objects will help considerably to speed up execution time, both for writing and reading metrics, and is worth the bit of extra effort, can you proceed with this?