Change Details

People have used qacct for various reasons, see eg. T181742. However, qacct is slowwwww: {P6445} ... taking around 30 seconds to read 14 MiB from NFS, and the file is 3.6G: ``` $ ls -lh /var/lib/gridengine/default/common/accounting /data/project/.system/accounting -rw-r--r-- 1 sgeadmin sgeadmin 3.6G Dec 8 19:14 /data/project/.system/accounting lrwxrwxrwx 1 root sgeadmin 32 Feb 22 2014 /var/lib/gridengine/default/common/accounting -> /data/project/.system/accounting ``` ... which means a complete scan, just to find a single job exit status, might take two hours, assuming nobody notice the resource hog and ask to kill it. We could: * Rotate the accounting logs, so qacct won't have to go through the whole file to find information about a job that recently exited, or * Use an alternative qacct parser such as that from tool [[https://phabricator.wikimedia.org/source/tool-grid-jobs/browse/master/grid_jobs/__init__.py|grid-jobs]], which will not, by default, scan through the file accounting logs.

People have used qacct for various reasons, see eg. T181742. However, qacct is slowwwww: {P6445} ... taking around 30 seconds to read 14 MiB from NFS, and the file is 3.6G: ``` $ ls -lh /var/lib/gridengine/default/common/accounting /data/project/.system/accounting -rw-r--r-- 1 sgeadmin sgeadmin 3.6G Dec 8 19:14 /data/project/.system/accounting lrwxrwxrwx 1 root sgeadmin 32 Feb 22 2014 /var/lib/gridengine/default/common/accounting -> /data/project/.system/accounting ``` ... which means a complete scan, just to find a single job exit status, might take two hours, assuming nobody notice the resource hog and ask to kill it. We could: * Rotate the accounting logs (T168701), so qacct won't have to go through the whole file to find information about a job that recently exited, or * Use an alternative qacct parser such as that from tool [[https://phabricator.wikimedia.org/source/tool-grid-jobs/browse/master/grid_jobs/__init__.py|grid-jobs]], which will not, by default, scan through the file accounting logs.

People have used qacct for various reasons, see eg. T181742. However, qacct is slowwwww: {P6445} ... taking around 30 seconds to read 14 MiB from NFS, and the file is 3.6G: ``` $ ls -lh /var/lib/gridengine/default/common/accounting /data/project/.system/accounting -rw-r--r-- 1 sgeadmin sgeadmin 3.6G Dec 8 19:14 /data/project/.system/accounting lrwxrwxrwx 1 root sgeadmin 32 Feb 22 2014 /var/lib/gridengine/default/common/accounting -> /data/project/.system/accounting ``` ... which means a complete scan, just to find a single job exit status, might take two hours, assuming nobody notice the resource hog and ask to kill it. We could: * Rotate the accounting logs (T168701), so qacct won't have to go through the whole file to find information about a job that recently exited, or * Use an alternative qacct parser such as that from tool [[https://phabricator.wikimedia.org/source/tool-grid-jobs/browse/master/grid_jobs/__init__.py|grid-jobs]], which will not, by default, scan through the file accounting logs.