Replace toJsonMap function with a lazy value

[chuwy]

Given it's used by ordered many times - we can significantly improve performance by computing it only once:

snowplow-scala-analytics-sdk/src/main/scala/com.snowplowanalytics.snowplow.analytics.scalasdk/Event.scala

Lines 202 to 203 in 5dc24d9

	def ordered: List[(String, Option[Json])] =
	Event.parser.knownKeys.map(key => (key.name, toJsonMap.get(key.name)))

[benjben]

Before the rewrite, here are the results of benchmarking (with jmh:run -i 10 -wi 3 -f2 -t3) :

[info] # JMH version: 1.25
[info] # VM version: JDK 11.0.6, OpenJDK 64-Bit Server VM, 11.0.6+10-post-Debian-1bpo91
[info] # VM invoker: /usr/lib/jvm/java-11-openjdk-amd64/bin/java
[info] # VM options: <none>
[info] # Warmup: 3 iterations, 10 s each
[info] # Measurement: 10 iterations, 10 s each
[info] # Timeout: 10 min per iteration
[info] # Threads: 3 threads, will synchronize iterations
[info] # Benchmark mode: Throughput, ops/time
[info] # Benchmark: com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered
[info] # Run progress: 0.00% complete, ETA 00:08:40
[info] # Fork: 1 of 2
[info] # Warmup Iteration   1: ≈ 10⁻³ ops/us
[info] # Warmup Iteration   2: ≈ 10⁻³ ops/us
[info] # Warmup Iteration   3: ≈ 10⁻³ ops/us
[info] Iteration   1: ≈ 10⁻³ ops/us
[info] Iteration   2: ≈ 10⁻³ ops/us
[info] Iteration   3: ≈ 10⁻³ ops/us
[info] Iteration   4: ≈ 10⁻³ ops/us
[info] Iteration   5: ≈ 10⁻³ ops/us
[info] Iteration   6: ≈ 10⁻³ ops/us
[info] Iteration   7: ≈ 10⁻³ ops/us
[info] Iteration   8: 0.001 ops/us
[info] Iteration   9: 0.001 ops/us
[info] Iteration  10: 0.001 ops/us
[info] # Run progress: 25.00% complete, ETA 00:06:32
[info] # Fork: 2 of 2
[info] # Warmup Iteration   1: ≈ 10⁻³ ops/us
[info] # Warmup Iteration   2: 0.001 ops/us
[info] # Warmup Iteration   3: ≈ 10⁻³ ops/us
[info] Iteration   1: 0.001 ops/us
[info] Iteration   2: 0.001 ops/us
[info] Iteration   3: 0.001 ops/us
[info] Iteration   4: 0.001 ops/us
[info] Iteration   5: 0.001 ops/us
[info] Iteration   6: 0.001 ops/us
[info] Iteration   7: 0.001 ops/us
[info] Iteration   8: 0.001 ops/us
[info] Iteration   9: 0.001 ops/us
[info] Iteration  10: 0.001 ops/us
[info] Result "com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered":
[info]   0.001 ±(99.9%) 0.001 ops/us [Average]
[info]   (min, avg, max) = (≈ 10⁻³, 0.001, 0.001), stdev = 0.001
[info]   CI (99.9%): [≈ 10⁻³, 0.001] (assumes normal distribution)
[info] # JMH version: 1.25
[info] # VM version: JDK 11.0.6, OpenJDK 64-Bit Server VM, 11.0.6+10-post-Debian-1bpo91
[info] # VM invoker: /usr/lib/jvm/java-11-openjdk-amd64/bin/java
[info] # VM options: <none>
[info] # Warmup: 3 iterations, 10 s each
[info] # Measurement: 10 iterations, 10 s each
[info] # Timeout: 10 min per iteration
[info] # Threads: 3 threads, will synchronize iterations
[info] # Benchmark mode: Average time, time/op
[info] # Benchmark: com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered
[info] # Run progress: 50.00% complete, ETA 00:04:21
[info] # Fork: 1 of 2
[info] # Warmup Iteration   1: 6524.157 ±(99.9%) 3151.615 us/op
[info] # Warmup Iteration   2: 5658.176 ±(99.9%) 675.118 us/op
[info] # Warmup Iteration   3: 5859.169 ±(99.9%) 189.335 us/op
[info] Iteration   1: 5685.592 ±(99.9%) 774.390 us/op
[info] Iteration   2: 5579.917 ±(99.9%) 306.011 us/op
[info] Iteration   3: 5647.106 ±(99.9%) 392.849 us/op
[info] Iteration   4: 5763.273 ±(99.9%) 650.658 us/op
[info] Iteration   5: 5775.713 ±(99.9%) 494.140 us/op
[info] Iteration   6: 5732.225 ±(99.9%) 2641.175 us/op
[info] Iteration   7: 5249.234 ±(99.9%) 344.314 us/op
[info] Iteration   8: 4537.562 ±(99.9%) 64.047 us/op
[info] Iteration   9: 4482.543 ±(99.9%) 200.290 us/op
[info] Iteration  10: 4403.862 ±(99.9%) 185.006 us/op
[info] # Run progress: 75.00% complete, ETA 00:02:10
[info] # Fork: 2 of 2
[info] # Warmup Iteration   1: 7038.108 ±(99.9%) 2290.543 us/op
[info] # Warmup Iteration   2: 6060.794 ±(99.9%) 282.454 us/op
[info] # Warmup Iteration   3: 6143.986 ±(99.9%) 110.664 us/op
[info] Iteration   1: 6346.754 ±(99.9%) 246.708 us/op
[info] Iteration   2: 6491.202 ±(99.9%) 2625.066 us/op
[info] Iteration   3: 6381.568 ±(99.9%) 1226.350 us/op
[info] Iteration   4: 6570.449 ±(99.9%) 2968.226 us/op
[info] Iteration   5: 6372.206 ±(99.9%) 1110.280 us/op
[info] Iteration   6: 6325.100 ±(99.9%) 684.072 us/op
[info] Iteration   7: 6022.833 ±(99.9%) 373.478 us/op
[info] Iteration   8: 5453.750 ±(99.9%) 477.195 us/op
[info] Iteration   9: 5346.006 ±(99.9%) 527.296 us/op
[info] Iteration  10: 5256.194 ±(99.9%) 158.473 us/op
[info] Result "com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered":
[info]   5671.154 ±(99.9%) 574.776 us/op [Average]
[info]   (min, avg, max) = (4403.862, 5671.154, 6570.449), stdev = 661.913
[info]   CI (99.9%): [5096.379, 6245.930] (assumes normal distribution)
[info] # Run complete. Total time: 00:08:43
[info] REMEMBER: The numbers below are just data. To gain reusable insights, you need to follow up on
[info] why the numbers are the way they are. Use profilers (see -prof, -lprof), design factorial
[info] experiments, perform baseline and negative tests that provide experimental control, make sure
[info] the benchmarking environment is safe on JVM/OS/HW level, ask for reviews from the domain experts.
[info] Do not assume the numbers tell you what you want them to tell.
[info] Benchmark                       Mode  Cnt     Score     Error   Units
[info] BenchmarkAnalyticsSDK.ordered  thrpt   20     0.001 ±   0.001  ops/us
[info] BenchmarkAnalyticsSDK.ordered   avgt   20  5671.154 ± 574.776   us/op

[benjben]

And now with the lazy val :

[info] # JMH version: 1.25
[info] # VM version: JDK 11.0.6, OpenJDK 64-Bit Server VM, 11.0.6+10-post-Debian-1bpo91
[info] # VM invoker: /usr/lib/jvm/java-11-openjdk-amd64/bin/java
[info] # VM options: <none>
[info] # Warmup: 3 iterations, 10 s each
[info] # Measurement: 10 iterations, 10 s each
[info] # Timeout: 10 min per iteration
[info] # Threads: 3 threads, will synchronize iterations
[info] # Benchmark mode: Throughput, ops/time
[info] # Benchmark: com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered
[info] # Run progress: 0.00% complete, ETA 00:08:40
[info] # Fork: 1 of 2
[info] # Warmup Iteration   1: 0.558 ops/us
[info] # Warmup Iteration   2: 0.593 ops/us
[info] # Warmup Iteration   3: 0.628 ops/us
[info] Iteration   1: 0.718 ops/us
[info] Iteration   2: 0.747 ops/us
[info] Iteration   3: 0.750 ops/us
[info] Iteration   4: 0.764 ops/us
[info] Iteration   5: 0.764 ops/us
[info] Iteration   6: 0.758 ops/us
[info] Iteration   7: 0.748 ops/us
[info] Iteration   8: 0.762 ops/us
[info] Iteration   9: 0.766 ops/us
[info] Iteration  10: 0.764 ops/us
[info] # Run progress: 25.00% complete, ETA 00:06:32
[info] # Fork: 2 of 2
[info] # Warmup Iteration   1: 0.649 ops/us
[info] # Warmup Iteration   2: 0.638 ops/us
[info] # Warmup Iteration   3: 0.687 ops/us
[info] Iteration   1: 0.691 ops/us
[info] Iteration   2: 0.714 ops/us
[info] Iteration   3: 0.712 ops/us
[info] Iteration   4: 0.706 ops/us
[info] Iteration   5: 0.680 ops/us
[info] Iteration   6: 0.707 ops/us
[info] Iteration   7: 0.710 ops/us
[info] Iteration   8: 0.705 ops/us
[info] Iteration   9: 0.705 ops/us
[info] Iteration  10: 0.692 ops/us
[info] Result "com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered":
[info]   0.728 ±(99.9%) 0.026 ops/us [Average]
[info]   (min, avg, max) = (0.680, 0.728, 0.766), stdev = 0.029
[info]   CI (99.9%): [0.703, 0.754] (assumes normal distribution)
[info] # JMH version: 1.25
[info] # VM version: JDK 11.0.6, OpenJDK 64-Bit Server VM, 11.0.6+10-post-Debian-1bpo91
[info] # VM invoker: /usr/lib/jvm/java-11-openjdk-amd64/bin/java
[info] # VM options: <none>
[info] # Warmup: 3 iterations, 10 s each
[info] # Measurement: 10 iterations, 10 s each
[info] # Timeout: 10 min per iteration
[info] # Threads: 3 threads, will synchronize iterations
[info] # Benchmark mode: Average time, time/op
[info] # Benchmark: com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered
[info] # Run progress: 50.00% complete, ETA 00:04:21
[info] # Fork: 1 of 2
[info] # Warmup Iteration   1: 4.553 ±(99.9%) 1.211 us/op
[info] # Warmup Iteration   2: 4.377 ±(99.9%) 0.218 us/op
[info] # Warmup Iteration   3: 4.350 ±(99.9%) 0.800 us/op
[info] Iteration   1: 4.201 ±(99.9%) 0.283 us/op
[info] Iteration   2: 4.368 ±(99.9%) 0.614 us/op
[info] Iteration   3: 4.219 ±(99.9%) 0.291 us/op
[info] Iteration   4: 4.215 ±(99.9%) 0.260 us/op
[info] Iteration   5: 4.286 ±(99.9%) 0.292 us/op
[info] Iteration   6: 4.184 ±(99.9%) 0.144 us/op
[info] Iteration   7: 4.187 ±(99.9%) 0.155 us/op
[info] Iteration   8: 4.192 ±(99.9%) 0.386 us/op
[info] Iteration   9: 4.280 ±(99.9%) 0.369 us/op
[info] Iteration  10: 4.201 ±(99.9%) 0.267 us/op
[info] # Run progress: 75.00% complete, ETA 00:02:10
[info] # Fork: 2 of 2
[info] # Warmup Iteration   1: 4.593 ±(99.9%) 0.942 us/op
[info] # Warmup Iteration   2: 4.392 ±(99.9%) 0.511 us/op
[info] # Warmup Iteration   3: 4.233 ±(99.9%) 0.559 us/op
[info] Iteration   1: 4.328 ±(99.9%) 0.505 us/op
[info] Iteration   2: 4.229 ±(99.9%) 0.450 us/op
[info] Iteration   3: 4.233 ±(99.9%) 0.531 us/op
[info] Iteration   4: 4.472 ±(99.9%) 0.417 us/op
[info] Iteration   5: 4.290 ±(99.9%) 0.687 us/op
[info] Iteration   6: 4.229 ±(99.9%) 0.213 us/op
[info] Iteration   7: 4.243 ±(99.9%) 0.209 us/op
[info] Iteration   8: 4.259 ±(99.9%) 0.660 us/op
[info] Iteration   9: 4.356 ±(99.9%) 0.354 us/op
[info] Iteration  10: 4.452 ±(99.9%) 0.305 us/op
[info] Result "com.snowplowanalytics.snowplow.analytics.scalasdk.benchmark.BenchmarkAnalyticsSDK.ordered":
[info]   4.271 ±(99.9%) 0.073 us/op [Average]
[info]   (min, avg, max) = (4.184, 4.271, 4.472), stdev = 0.085
[info]   CI (99.9%): [4.198, 4.345] (assumes normal distribution)
[info] # Run complete. Total time: 00:08:42
[info] REMEMBER: The numbers below are just data. To gain reusable insights, you need to follow up on
[info] why the numbers are the way they are. Use profilers (see -prof, -lprof), design factorial
[info] experiments, perform baseline and negative tests that provide experimental control, make sure
[info] the benchmarking environment is safe on JVM/OS/HW level, ask for reviews from the domain experts.
[info] Do not assume the numbers tell you what you want them to tell.
[info] Benchmark                       Mode  Cnt  Score   Error   Units
[info] BenchmarkAnalyticsSDK.ordered  thrpt   20  0.728 ± 0.026  ops/us
[info] BenchmarkAnalyticsSDK.ordered   avgt   20  4.271 ± 0.073   us/op

[benjben]

Even better than a 1000x improvement !

[benjben]

With

def ordered: List[(String, Option[Json])] =
    this.asJsonObject.toIterable.map { case (k, v) => (k, Some(v))}.toList

we have

[info]   27.385 ±(99.9%) 3.354 us/op [Average]
[info]   (min, avg, max) = (21.310, 27.385, 31.401), stdev = 3.863
[info]   CI (99.9%): [24.031, 30.739] (assumes normal distribution

A bit slower so will stick to the existing one

[chuwy]

My idea was more to introduce another/replace later the function as Option[Json] is useless an in fact always Json.Null. Without Some(x) and toList I think it will be slightly faster, but it just doesn't matter much and probably a part of the other ticket (but worth to leave the benchmark itself!)