2. # whoami
● Jan Horacek aka jahor, @jhrcz
● sysadmin at Etnetera a.s. (http://etnetera.cz/en)
● like to automate things, have the workflow given and documented
● 150empl development company, java and oss based
● 10 sysadmins, 80% developers/coders
● we make webs (but not only...;o)
● jNetPublish, CMS (http://jnetpublish.cz), customizing and running for our clients
● soyka (http://soyka.etnetera.cz/en) - personalisation tool
● smartmeter (http://smartmeter.etnetera.cz/en) - stress testing tool, jmeter on steroids
● ewa (http://ewa.etnetera.cz) - etnetera web accelerator
| Název prezentace | Jméno Příjmení
3. OpenNebula @ Etnetera
● ~3 year in production - a bit cloudy VDC
● 8 nodes (192cores, 768gigs ram), 11TB ds, 260 images, 150 vms
● clvm, gfs2, mgmt node disconnected from cluster
● lvm as system datastore
● contextualization independent - dhcp in sync with onevnets
● now we are on latest stable (after testing)
● 3.2.1 (2012)
● 4.2 (2014-04) - switch to “almost vanilla” lvm driver
● 4.6.2 (2014-08), 4.8 (2014-11), 4.10.1 (2014-11)
● started with rewritten lvm storage driver - aka gfs2clvm
● now lvm driver with ~5 small patches
| Název prezentace | Jméno Příjmení
4. there is allways something... small to hack on
● lvm as a system datastore - nonpersistent vols in lvm
● io capping / boosting (with cgroups), cpu.shares ~ vcpu (gh)
● per vm io boost, low io base + BLKIO_BOOST vm param
● reveals wrong behaving vms
● global cpu / memory overcommit (gh)
● vm owner (gh) / vm name fixing (gh)
● per vm dns names ($VMID.vms.one1.etn) (gh)
● template owner vs image owner checker (gh)
● template backup and history (gh)
● reinstantiate command (new vmid, uptodate tpl) (gh)
● kvm clock resync after suspend (gh)
| Název prezentace | Jméno Příjmení
6. sometime you do bad choices - they could have sollution
ONE could start small and then it could grow in time
ONE is realy simple, transparent and customizable
Thank YOU!
| Název prezentace | Jméno Příjmení