You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by GitBox <gi...@apache.org> on 2019/06/25 23:09:29 UTC
[GitHub] [flink] NicoK opened a new pull request #8888:
[FLINK-12983][metrics] replace descriptive histogram's storage back-end
NicoK opened a new pull request #8888: [FLINK-12983][metrics] replace descriptive histogram's storage back-end
URL: https://github.com/apache/flink/pull/8888
## What is the purpose of the change
Instead of using `DescriptiveStatistics` and its `ResizableDoubleArray` storage back-end, implement an own circular array based on a fixed-size double array that wraps around. This significantly improves the time needed to add new values. It also ensures that access is synchronized which `ResizableDoubleArray` does not guarantee.
Performance results:
1. only adding values to the histogram
```
Benchmark Mode Cnt Score Error Units
HistogramBenchmarks.dropwizardHistogramAdd thrpt 30 47985.359 ± 25.847 ops/ms
HistogramBenchmarks.descriptiveHistogramAdd thrpt 30 70158.792 ± 276.858 ops/ms
-> #8887:
HistogramBenchmarks.descriptiveHistogramAdd thrpt 30 75303.040 ± 475.355 ops/ms
-> this PR:
HistogramBenchmarks.descrHistogramCircularAdd thrpt 30 200906.902 ± 384.483 ops/ms
-> using a linear array of twice the histogram size (and moving values back to the start once the window reaches the end), for reference:
HistogramBenchmarks.descrHistogramLinearAdd thrpt 30 189788.728 ± 233.283 ops/ms
```
2. after adding each value, also retrieving a common set of metrics:
```
Benchmark Mode Cnt Score Error Units
HistogramBenchmarks.dropwizardHistogram thrpt 30 400.274 ± 4.930 ops/ms
HistogramBenchmarks.descriptiveHistogram thrpt 30 124.533 ± 1.060 ops/ms
-> #8887:
HistogramBenchmarks.descriptiveHistogram thrpt 30 251.895 ± 1.809 ops/ms
-> this PR:
HistogramBenchmarks.descrHistogramCircular thrpt 30 301.068 ± 2.077 ops/ms
-> using a linear array as above:
HistogramBenchmarks.descrHistogramLinear thrpt 30 234.050 ± 5.485 ops/ms
```
Please note that this PR is based on #8887.
## Brief change log
- add `CircularDoubleArray` to `DescriptiveStatisticsHistogram`
- remove extra `AtomicLong elementsSeen` (can be done more efficiently in `CircularDoubleArray`)
## Verifying this change
This change is already covered by existing tests, such as derivatives of `AbstractHistogramTest`.
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): **no**
- The public API, i.e., is any changed class annotated with `@Public(Evolving)`: **no**
- The serializers: **no**
- The runtime per-record code paths (performance sensitive): **maybe** (histograms are used by latency markers only but may be used by users)
- Anything that affects deployment or recovery: JobManager (and its components), Checkpointing, Yarn/Mesos, ZooKeeper: **no**
- The S3 file system connector: **no**
## Documentation
- Does this pull request introduce a new feature? **no**
- If yes, how is the feature documented? **JavaDocs**
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
With regards,
Apache Git Services