Instance meter returning incorect uptime samples

[Diogo Monteiro]

Instead of returning information regarding the uptime(time the instance has been running) the meter is returning the time-span since the instance has been created.

There is already a blue print documenting in more detail the issue and a possible solution: https://blueprints.launchpad.net/ceilometer/+spec/uptime-measurement

Can anybody provide some updates on this?

[gordon chung (chungg)]

@Diogo it's not being worked on currently. Anyone is free to start working on it but they would need to add a quick spec here first: https://github.com/openstack/ceilometer-specs

cheers,

[Diogo Monteiro (diogo-gmt)]

@chungg
Do you know if this issue has been resolved in the kilo release?

[gordon chung (chungg)]

@Diogo,

there has been no work on this specifically in regards to proposed blueprint. there is a cputime meter that can be used... also (a more accurate) cpu.delta meter https://review.openstack.org/#/c/221907/

[gordon chung (chungg)]

@Diogo,

we're waiting on this: http://www.redhat.com/archives/libvir-list/2015-March/msg01518.html

we will aim to address it in M* cycle

[gordon chung (chungg)]

i don't think we can target this. uptime according to libvirt thread can only be obtained via guest. marking incomplete for now but i think we might need to give up on this.

[gordon chung (chungg)]

i think the only solution outside of libvirt support would bet to compute this at storage (either panko or gnocchi or whatever storage you're using) based on nova events.

i'm closing this but someone can target those solutions if required.