Adding support for MemAvailable

[truthbk]

Based on: #1826 - bringing in @jraede changes and making sure it's only reported if the kernel actually has it in /proc/meminfo

[remh]

This will need some changes in the backend as well.
@olivielpeau i think you've done similar things in the past when you add guest cpu metrics. Can you work with @truthbk on that please ?

[truthbk]

@remh I was adding the MemAvailable stuff to the backend, but it kind of feels like redundant. Maybe we should just report MemAvailable as MemUsable when its available, and compute usable when it's not available (aka. like it's done now). Thoughts?

[truthbk]

It doesn't make sense to report usable and available. We just use available if the kernel has it, or compute usable if it's not.

[talwai]

doesn't look like physBuffers and physCached are populated until L374-375 . Maybe move this block of code below that?

[truthbk]

Thanks for spotting that! Careless.

[talwai]

👍

[remh]

@talwai @truthbk Did you check if those metrics were roughly equivalent ? We need to make sure that they'll be concistent with the previous value.

[truthbk]

@remh the thing is MemAvailable was introduced precisely because it's not always so straight-forward to estimate the amount of memory available to the system: https://access.redhat.com/solutions/776393. As it turns out the typical way of doing free + cached + buffers didn't always reflect the actual memory available.

[JohnLZeller]

@remh can +1 what @truthbk is saying here. Looks like the MemAvailable number is lower than our calculation (ex memavail: 529 VS calculated: 663), meaning that our manual calculation is actually off a bit.