Restore proto round trip for Java DirectRunner (was: Performance degradation on the direct runner)
- Dominant language
- Java
- Stars
- 8.7k
- Forks
- 4.7k
- Avg merge
- 2d 2h
- Merged PRs (30d)
- 205
Description
Running Nexmark queries with the direct runner between Beam 2.2.0 and 2.3.0 shows a performance degradation:
```
========================================
Beam 2.2.0 Beam 2.3.0
Query Runtime(sec)
Runtime(sec)
========================================
0000 6.4 10.6
0001 5.1 10.2
0002 3.0 5.8
0003 3.8
6.2
0004 0.9 1.4
0005 5.8 11.4
0006
0.8 1.4
0007 193.8 1249.1
0008 3.9 6.9
0009 0.9 1.3
0010 6.4 8.2
0011 5.0
9.4
0012 4.7 9.1
```
We can see especially Query 7 that is 10 times longer.
Imported from Jira [BEAM-3617](https://issues.apache.org/jira/browse/BEAM-3617). Original Jira may contain additional context.
Reported by: jbonofre.
Contributor guide
Research direction
Start with the Nexmark queries on the Java DirectRunner and compare Beam 2.2.0 with 2.3.0 using the reported runtime table. Investigate the proto round-trip path indicated by the title, with Query 0007 as the main regression case; done means restoring the earlier performance.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- java
- Domain
- performance
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100