voonhous commented on PR #19906: URL: https://github.com/apache/hudi/pull/19906#issuecomment-5639223708
Same analysis as the fta table, extended to every job: test-execution minutes per job, summed from the surefire class times and the scalatest "Run completed in" lines of each job log, for the baseline and the three runs of this PR. This excludes queueing, runner setup and the build step; it still includes whatever CPU the shared VM had that hour. | job | #19892 run (min) | run 3 (min) | run 4 (min) | run 5 (min) | run 3 | run 4 | run 5 | direction | |---|---|---|---|---|---|---|---|---| | test-spark-java17-ut-fat-and-ftc | 49.4 | 44.2 | 36.8 | 58.1 | -11% | -26% | +18% | mixed | | test-utilities-streamer | 46.7 | 44.8 | 43.2 | 45.8 | -4% | -7% | -2% | faster in all three | | test-spark-java17-fta | 46.1 | 54.9 | 57.8 | 52.0 | +19% | +25% | +13% | slower in all three | | test-spark-java17-ut-other | 45.3 | 42.9 | 33.5 | 45.5 | -5% | -26% | 0% | mixed | | test-spark-client-and-hadoop-common | 44.6 | 39.5 | 44.9 | 44.4 | -12% | +1% | 0% | mixed | | test-spark-java17-ut-functional | 43.1 | 28.9 | 40.2 | 40.1 | -33% | -7% | -7% | faster in all three | | test-spark-java17-ftb | 40.4 | 41.8 | 40.4 | 38.6 | +4% | 0% | -5% | mixed | | test-common-and-other-modules | 33.9 | 35.3 | 31.1 | 29.4 | +4% | -8% | -13% | mixed | | test-spark-java17-scala-other-1 | 32.2 | 20.4 | 32.5 | 25.5 | -37% | +1% | -21% | mixed | | test-spark-java17-scala-dml-1 | 30.5 | 23.6 | 30.0 | 30.6 | -22% | -2% | 0% | mixed | | test-utilities-rest | 30.2 | 30.0 | 23.1 | 22.2 | -1% | -24% | -27% | faster in all three | | test-spark-java17-scala-other-2 | 24.3 | 28.1 | 22.4 | 23.1 | +15% | -8% | -5% | mixed | | test-spark-java17-scala-dml-2 | 23.0 | 22.6 | 23.5 | 23.2 | -2% | +2% | +1% | mixed | | test-hudi-hadoop-mr-and-hudi-java-client | 9.8 | 15.0 | 14.2 | 14.7 | +54% | +45% | +51% | slower in all three | | total | 499.7 | 472.0 | 473.5 | 493.1 | -6% | -5% | -1% | | What it says: three jobs are faster in every run (utilities-streamer, utilities-rest, ut-functional), two are slower in every run (fta, and hadoop-mr-and-java-client by half), and the other nine swing by up to 37% between runs of the same code, which is the size of the noise on this pool. The total is -1% to -6%. A fourth run would add one sample to a spread that wide and would not change the conclusion, so this PR stays closed. -- 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]
