Provide way to collect start/end read/write time inside the IOs
- Dominant language
- Java
- Stars
- 8.7k
- Forks
- 4.7k
- Avg merge
- 1d 20h
- Merged PRs (30d)
- 196
Description
Currently, IO tests measure time using Metrics API but collect start/end time from ParDo transforms that are adjacent to the IO. It's fine for some tests but maybe could be done better. The drawback of the current solution is that we cannot collect time before PBegin and after PDone. Other than that the time we collect now is still not the exact time of read/write start/end but only the time at which first/last record appeared in the DoFn.
See: [TimeMonitor.java](https://github.com/apache/beam/blob/957b7cc7746aa626d2eb4dea341f668ec19d5d39/sdks/java/testing/test-utils/src/main/java/org/apache/beam/sdk/testutils/metrics/TimeMonitor.java) as an example of such DoFn.
Possible solution: save metrics in startBundle / finishBundle method in IOs whenever a dedicated pipelineOption is set to true.
In general, maybe it's a good idea to place some other metrics inside IOs too? wdyt?
Imported from Jira [BEAM-6969](https://issues.apache.org/jira/browse/BEAM-6969). Original Jira may contain additional context.
Reported by: ŁukaszG.
Contributor guide
Assessment
This issue has not been assessed yet.