steveloughran commented on pull request #2971: URL: https://github.com/apache/hadoop/pull/2971#issuecomment-1057227746
just tested this through google and abfs. a key finding is that the dir creation stage is pretty slow, esp on gcs. I'm going to add a getfilestatus call before each mkdir, rather than always call mkdir and handle failures; while nominally less io on the good path, it doesn't seem to offer any real benefits. # gcs ``` means=((commit_file_rename.mean=(samples=1, sum=168, mean=168.0000)) (committer_commit_job.mean=(samples=1, sum=881, mean=881.0000)) (committer_task_directory_count=(samples=1, sum=2, mean=2.0000)) (committer_task_directory_depth=(samples=1, sum=3, mean=3.0000)) (committer_task_file_count=(samples=1, sum=1, mean=1.0000)) (committer_task_file_size=(samples=1, sum=6358, mean=6358.0000)) (committer_task_manifest_file_size=(samples=1, sum=19526, mean=19526.0000)) (job_stage_cleanup.mean=(samples=1, sum=0, mean=0.0000)) (job_stage_create_target_dirs.mean=(samples=1, sum=193, mean=193.0000)) (job_stage_load_manifests.mean=(samples=1, sum=76, mean=76.0000)) (job_stage_optional_validate_output.mean=(samples=1, sum=35, mean=35.0000)) (job_stage_rename_files.mean=(samples=1, sum=176, mean=176.0000)) (job_stage_save_success_marker.mean=(samples=1, sum=394, mean=394.0000)) (job_stage_setup.mean=(samples=1, sum=302, mean=302.0000)) (op_create_directories.mean=(samples=1, sum=193, mean=193.0000)) (op_delete.mean=(samples=2, sum=44, mean=22.0000)) (op_get_file_status.failures.mean=(samples=3, sum=60, mean=20.0000)) (op_get_file_status.mean=(samples=1, sum=27, mean=27.0000)) (op_list_status.mean=(samples=4, sum=111, mean=27.7500)) (op_load_all_manifests.mean=(samples=1, sum=51, mean=51.0000)) (op_load_manifest.mean=(samples=1, sum=41, mean=41.0000)) (op_mkdirs.mean=(samples=5, sum=761, mean=152.2000)) (op_msync.mean=(samples=1, sum=0, mean=0.0000)) (op_rename.mean=(samples=1, sum=121, mean=121.0000)) (task_stage_commit.mean=(samples=1, sum=87, mean=87.0000)) (task_stage_save_task_manifest.mean=(samples=1, sum=252, mean=252.0000)) (task_stage_scan_directory.mean=(samples=1, sum=87, mean=87.0000)) (task_stage_setup.mean=(samples=1, sum=169, mean=169.0000))); ``` # abfs ``` means=((commit_file_rename.mean=(samples=1, sum=97, mean=97.0000)) (committer_commit_job.mean=(samples=1, sum=460, mean=460.0000)) (committer_task_directory_count=(samples=1, sum=2, mean=2.0000)) (committer_task_directory_depth=(samples=1, sum=3, mean=3.0000)) (committer_task_file_count=(samples=1, sum=1, mean=1.0000)) (committer_task_file_size=(samples=1, sum=6373, mean=6373.0000)) (committer_task_manifest_file_size=(samples=1, sum=19784, mean=19784.0000)) (job_stage_cleanup.mean=(samples=1, sum=0, mean=0.0000)) (job_stage_create_target_dirs.mean=(samples=1, sum=54, mean=54.0000)) (job_stage_load_manifests.mean=(samples=1, sum=74, mean=74.0000)) (job_stage_optional_validate_output.mean=(samples=1, sum=23, mean=23.0000)) (job_stage_rename_files.mean=(samples=1, sum=110, mean=110.0000)) (job_stage_save_success_marker.mean=(samples=1, sum=194, mean=194.0000)) (job_stage_setup.mean=(samples=1, sum=237, mean=237.0000)) (op_create_directories.mean=(samples=1, sum=49, mean=49.0000)) (op_delete.mean=(samples=2, sum=40, mean=20.0000)) (op_get_file_status.failures.mean=(samples=3, sum=80, mean=26.6667)) (op_get_file_status.mean=(samples=1, sum=17, mean=17.0000)) (op_list_status.mean=(samples=4, sum=75, mean=18.7500)) (op_load_all_manifests.mean=(samples=1, sum=57, mean=57.0000)) (op_load_manifest.mean=(samples=1, sum=32, mean=32.0000)) (op_mkdirs.mean=(samples=5, sum=282, mean=56.4000)) (op_msync.mean=(samples=1, sum=0, mean=0.0000)) (op_rename.mean=(samples=1, sum=75, mean=75.0000)) (task_stage_commit.mean=(samples=1, sum=129, mean=129.0000)) (task_stage_save_task_manifest.mean=(samples=1, sum=99, mean=99.0000)) (task_stage_scan_directory.mean=(samples=1, sum=129, mean=129.0000)) (task_stage_setup.mean=(samples=1, sum=64, mean=64.0000))); ``` -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
