[Profiling] Map stack frames more efficiently

[danielmitterdorfer]

The mapping code for stack frames uses the utility method ObjectPath#eval to read nested properties. Callers need to pass a dot-separated path which is then split internally via a regex. This is quite slow: in a typical deployment we saw overheads of 50ms just for mapping stack frames (total response time is ~ 1 second).

With this commit we pass the property path as an array to avoid this overhead. In a microbenchmark the new implementation was 23 times faster than the current one.

[elasticsearchmachine]

Hi @danielmitterdorfer, I've created a changelog YAML for you.

[elasticsearchmachine]

Pinging @elastic/es-search (Team:Search)

[danielmitterdorfer]

Below are the results of a microbenchmark with JMH's gc profiler enabled. We can see that the new implementation pathFromArray is ~ 23 times faster than pathFromString. Additionally, it does not cause any allocations:

Benchmark                                                             Mode  Cnt     Score     Error   Units
ObjectPathBenchmark.pathFromArray                                    thrpt   30   385.025 ±   0.888  ops/us
ObjectPathBenchmark.pathFromArray:·gc.alloc.rate                     thrpt   30    ≈ 10⁻⁴            MB/sec
ObjectPathBenchmark.pathFromArray:·gc.alloc.rate.norm                thrpt   30    ≈ 10⁻⁷              B/op
ObjectPathBenchmark.pathFromArray:·gc.count                          thrpt   30       ≈ 0            counts
ObjectPathBenchmark.pathFromString                                   thrpt   30    16.230 ±   1.733  ops/us
ObjectPathBenchmark.pathFromString:·gc.alloc.rate                    thrpt   30  3537.559 ± 377.722  MB/sec
ObjectPathBenchmark.pathFromString:·gc.alloc.rate.norm               thrpt   30   240.010 ±   0.001    B/op
ObjectPathBenchmark.pathFromString:·gc.churn.G1_Eden_Space           thrpt   30  3540.005 ± 379.118  MB/sec
ObjectPathBenchmark.pathFromString:·gc.churn.G1_Eden_Space.norm      thrpt   30   240.156 ±   0.687    B/op
ObjectPathBenchmark.pathFromString:·gc.churn.G1_Survivor_Space       thrpt   30     0.014 ±   0.003  MB/sec
ObjectPathBenchmark.pathFromString:·gc.churn.G1_Survivor_Space.norm  thrpt   30     0.001 ±   0.001    B/op
ObjectPathBenchmark.pathFromString:·gc.count                         thrpt   30  2310.000            counts
ObjectPathBenchmark.pathFromString:·gc.time                          thrpt   30  1785.000                ms

[benwtrent]

Makes sense to me! Creating a string only to split it again is a ton of useless work.

[danielmitterdorfer]

Thanks for the quick review!