[ https://ovirt-jira.atlassian.net/browse/OVIRT-636?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18608#comment-18608 ]
Barak Korren commented on OVIRT-636: ------------------------------------ We can also create the zram drive and do the mounts in the job code... I think its best to let the job create and destroy things because that will ensure things get cleaned. But - we need to ensure we keep cache directories around, so maybe we need both - move caches to RAM on host startup and write them back to disk periodically and on shut-down. And have the job workspaces be put on ram and destroyed after every run. IIRC /var/lib/lago is the local cache directory that is bind-mounted into the Mock chroot, so that need to me moved to RAM on boot and synced back to disk. The rest of the job I/O seems to indeed happen in /var/lib/mock and /home/jenkins so we can consider placing a ram drive there each time the job starts. But /home/jenkins might be tricky because it may be used for other stuff (the slave binary for example), so we should probably just dynamically create a ram-drive for $WORKSPACE at the job level. One way to move things into RAM at boot time is to use '/etc/sysconfig/readonly-root' and '/etc/rwtab', this has the benefit of making the hosts mostly stateless, but may also make changing configuration a little tricky. > Add disks in memory for lago slaves > ------------------------------------ > > Key: OVIRT-636 > URL: https://ovirt-jira.atlassian.net/browse/OVIRT-636 > Project: oVirt - virtualization made easy > Issue Type: Bug > Reporter: eyal edri [Administrator] > Assignee: infra > > We need to try moving the Lago slaves in CI to work with memory instead of > local disks, > It might speed up the tests significantly. > We have 2 choices: > 1. add code to the job and mount the dir on /dev/shm/ > 2. create zram drive and mount the following dirs on it: > # /var/lib/mock > # /var/lib/lago > # /home/jenkins > Each host should have enough memory to run current tests on them, > however we'll need to make sure to clean that drive after each run. > Lets try it on one of the slaves to see what is the best solution before > implementing for all slaves. -- This message was sent by Atlassian JIRA (v1000.148.3#100005) _______________________________________________ Infra mailing list Infra@ovirt.org http://lists.ovirt.org/mailman/listinfo/infra