jihoonson opened a new pull request #11597:
URL: https://github.com/apache/druid/pull/11597
### Description
This PR improves the performance of the segments table by 1) exploiting
pre-sorted data structure and 2) caching results to avoid repeated execution of
expensive operations. For 1), when the published segments caching is enabled,
the cached published segments in `MetadataSegmentView` is sorted in the same
order used to sort the available segments in `DruidSchema`. When a query scans
segments, the query engine now can merge published and available segments using
the merge sorted algorithm as they are sorted in the same order. The previous
hash-based merge algorithm is still used when the caching is disabled. Also,
`druid.sql.planner.forceHashBasedMergeForSegmentsTable` is added to force the
hash-based merge even when the caching is enabled, which is off by default
(merge-sorted is on by default).
For 2), scanning the segments table requires to convert timestamps to
strings and serialize other non-primitive types to JSON strings. Since string
conversion and JSON serialization are expensive operations, a cache is added to
avoid calling those operations for the same object repeatedly. The cache can be
GCed once the scan is finished.
The benchmark results are in the below. This might be a bit biased since all
segments had the same metadata, where the technique used for 2) can show the
best performance. In reality, since segments can have different segment
granularity, dimensions, and metrics, the performance gain could be lower than
what is shown here.
```
master
Benchmark (availableSegmentsInterval)
(forceHashBasedMerge) (numSegmentsPerInterval) (publishedSegmentsInterval)
(segmentGranularity) (sql) Mode Cnt Score Error Units
SystemSchemaBenchmark.segmentsTable 2021-01-01/P3Y
true 10 2021-01-02/P3Y
DAY 0 avgt 10 30.596 ± 0.399 ms/op
SystemSchemaBenchmark.segmentsTable 2021-01-01/P3Y
true 10 2021-01-02/P3Y
DAY 1 avgt 10 39.703 ± 0.238 ms/op
SystemSchemaBenchmark.segmentsTable 2021-01-01/P3Y
true 100 2021-01-02/P3Y
DAY 0 avgt 10 382.867 ± 6.577 ms/op
SystemSchemaBenchmark.segmentsTable 2021-01-01/P3Y
true 100 2021-01-02/P3Y
DAY 1 avgt 10 387.911 ± 3.867 ms/op
SystemSchemaBenchmark.segmentsTable 2021-01-01/P3Y
true 1000 2021-01-02/P3Y
DAY 0 avgt 10 3971.577 ± 70.304 ms/op
SystemSchemaBenchmark.segmentsTable 2021-01-01/P3Y
true 1000 2021-01-02/P3Y
DAY 1 avgt 10 5334.702 ± 44.773 ms/op
```
```
PR
* Benchmark (availableSegmentsInterval)
(forceHashBasedMerge) (numSegmentsPerInterval) (publishedSegmentsInterval)
(segmentGranularity) (sql) Mode Cnt Score Error Units
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
true 10 2021-01-02/P3Y
DAY 0 avgt 10 14.914 ± 0.305 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
true 10 2021-01-02/P3Y
DAY 1 avgt 10 21.591 ± 0.139 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
true 100 2021-01-02/P3Y
DAY 0 avgt 10 199.416 ± 2.208 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
true 100 2021-01-02/P3Y
DAY 1 avgt 10 203.366 ± 4.706 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
true 1000 2021-01-02/P3Y
DAY 0 avgt 10 2110.735 ± 65.326 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
true 1000 2021-01-02/P3Y
DAY 1 avgt 10 2615.910 ± 69.914 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
false 10 2021-01-02/P3Y
DAY 0 avgt 10 11.967 ± 0.308 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
false 10 2021-01-02/P3Y
DAY 1 avgt 10 18.917 ± 0.220 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
false 100 2021-01-02/P3Y
DAY 0 avgt 10 123.630 ± 0.839 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
false 100 2021-01-02/P3Y
DAY 1 avgt 10 132.013 ± 0.664 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
false 1000 2021-01-02/P3Y
DAY 0 avgt 10 1303.304 ± 40.063 ms/op
* SegmentsTableBenchmark.segmentsTable 2021-01-01/P3Y
false 1000 2021-01-02/P3Y
DAY 1 avgt 10 1799.447 ± 23.659 ms/op
```
<hr>
##### Key changed/added classes in this PR
* `MetadataSegmentView`
* `SegmentsTableRow`
* `SystemSchema`
<hr>
<!-- Check the items by putting "x" in the brackets for the done things. Not
all of these items apply to every PR. Remove the items which are not done or
not relevant to the PR. None of the items from the checklist below are strictly
necessary, but it would be very helpful if you at least self-review the PR. -->
This PR has:
- [x] been self-reviewed.
- [ ] using the [concurrency
checklist](https://github.com/apache/druid/blob/master/dev/code-review/concurrency.md)
(Remove this item if the PR doesn't have any relation to concurrency.)
- [x] added documentation for new or modified features or behaviors.
- [x] added Javadocs for most classes and all non-trivial methods. Linked
related entities via Javadoc links.
- [ ] added or updated version, license, or notice information in
[licenses.yaml](https://github.com/apache/druid/blob/master/dev/license.md)
- [x] added comments explaining the "why" and the intent of the code
wherever would not be obvious for an unfamiliar reader.
- [x] added unit tests or modified existing tests to cover new code paths,
ensuring the threshold for [code
coverage](https://github.com/apache/druid/blob/master/dev/code-review/code-coverage.md)
is met.
- [ ] added integration tests.
- [ ] been tested in a test Druid cluster.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
