Speed up builds of freshly checked out branches

[nineinchnick]

Add a quick profile, which allows to quickly build the project without any checks or tests, assuming it has just been checked out from a clean branch (master) which has been tested by the CI workflow and is known to be correct. The new profile is supposed to make it easier, instead of having to remember all the correct properties (compare the commands in the next section).

I'm not disabling any modules (like docs) on purpose, they still have to be disabled explicitly. Note that building the tarball with Provisio takes up ~25% of wall-time, and the GC and JIT take ~50%.

Timings

I ran all the commands on my macbook with the profiler enabled:

export MAVEN_OPTS="-agentpath:$HOME/Downloads/async-profiler-2.5.1-macos/build/libasyncProfiler.dylib=start,event=cpu,file=mvn-profile.html"

Base

Running on master (only skip tests and checks):

./mvnw clean install --strict-checksums -V -DskipTests -Dair.check.skip-all -pl '!:trino-server-rpm,!docs' 

yields Total time: 04:41 min.

When also skipping test compilation, building source JARs, and others:

./mvnw clean install --strict-checksums -V -DskipTests -Dmaven.test.skip=true -Dmaven.source.skip=true -Dmaven.site.skip=true -Dmaven.javadoc.skip=true -Dair.check.skip-all -pl '!:trino-server-rpm,!docs'

yields Total time: 03:48 min.

Running the same command using Maven Daemon, the 3rd run after starting the daemon yields Total time: 01:28 min (Wall Clock).

With the profile

Running:

./mvnw clean install --strict-checksums -V -Dquick -pl '!:trino-server-rpm,!docs'

yields Total time: 03:08 min, 40 seconds quicker.

Running this 3rd time after starting the daemon:

mvnd clean install --strict-checksums -V -Dquick -pl '!:trino-server-rpm,!docs'

yields Total time: 01:19 min (Wall Clock), 10 seconds quicker.

All the work here is based solely on https://peter.palaga.org/presentations/211021-baselone-maven-my-life-is-short/

[hashhar]

Who is the expected user of the quick profile?

Can you extract "Run git commit id Maven plugin only once" to separate PR. It's non-controversial and always a gain so it can be merged right now.

[findepi]

cc @losipiuk

[nineinchnick]

I'm closing this. The gain of adding options other than -DskipTests -Dair.check.skip-all is tiny, about 4 seconds, and we can't reliably disable test compilation. Other changes are in airlift/airbase#293

[nineinchnick]

Benchmark results:

Benchmark 1: ./mvnw clean install -Pquick -pl "!:trino-server-rpm,!docs"
  Time (mean ± σ):     382.767 s ± 28.953 s    [User: 509.683 s, System: 24.475 s]
  Range (min … max):   362.434 s … 415.916 s    3 runs
 
Benchmark 2: ./mvnw clean install -DskipTests -Dair.check.skip-all=true -pl "!:trino-server-rpm,!docs"
  Time (mean ± σ):     368.216 s ±  1.681 s    [User: 512.733 s, System: 25.812 s]
  Range (min … max):   366.666 s … 370.003 s    3 runs

There's some variance for -Pquick but notice how small is difference in the minimum time.

I also ran benchmarks for the noTests profile, that only works when ~/.m2/repository is populated (any previous mvn install completed):

Benchmark 1: ./mvnw clean install -Pquick,noTests -pl "!:trino-server-rpm,!docs"
  Time (mean ± σ):     276.097 s ±  0.659 s    [User: 366.680 s, System: 21.953 s]
  Range (min … max):   275.593 s … 276.842 s    3 runs

[ksobolew]

Is there some value in the first commit alone? Seems like a nice cleanup at least.

[nineinchnick]

It'll allow to avoid loading unused plugins in some cases, but the actual time difference is 1-2 second. I can open up another PR with only that if airlift/airbase#293 gets merged, there won't be any difference without it.