Manageiq: VM Uptime

[ahrechushkin]

Hello!
I've a problem with calculate uptime and chargeback
We use rate allocated per hour. I think it will be calculated by this formula:

Memory allocation per hour (in MB) * Hourly Allocation cost per megabyte * Number of Memory Allocation metrics available for the day

^ https://access.redhat.com/documentation/en-us/red_hat_cloudforms/4.7/html/monitoring_alerts_and_reporting/sect_chargeback

But where i can find allocation metrics ?
In MetricRollup i see only used metric

So, and the last question: how i can calculate vm uptime? (if it possible)

[Fryguy]

Allocation is based on what is defined for the object during that hour. So, if a VM has 4 vcpus and 16GB memory, those are the allocation amounts. They would be derived from the values in vim_performance_states, which are snapshots of the state of the VM every hour. Additionally, there are columns in the metrics table, such as derived_memory_available, which is used for allocation calculations.

@gtanzillo @lpichler Can you keep me honest here ^?

how i can calculate vm uptime? (if it possible)

That one I'm not sure. I don't believe we store uptime from the provider, so collecting that would be an enhancement. We do have boot_time, but that's not exactly the same thing. We do also have when the power state changed (state_changed_on), so you could technically derive it from VMs that are powered on and seeing when the power on state last changed. @agrare do you know?

[agrare]

We are able to get boot_time from some providers (amazon, rhev, and vsphere) which will be more accurate than state_changed_on (state_changed_on is when we saved it, not when it actually started) but that will only give you the most recent uptime.

If you're trying to calculate e.g. how long has a vm been up over a period of time like a day, week, etc... you're best is probably to look at power state ems_events and sum the time between start and stop events within that time period.

[ahrechushkin]

So, I found a way. A fetch from consumption info about cpu_usage_average. If usage more than 0, I think that vm was powered on during 1 hour.
And use this info I can collect uptime per consumption period.
Also, I saw that consumption_rollup have a field power_status, but it always nil...

[ahrechushkin]

Also i wrote a method which calculate costs for dynamic allocated resources by uptime
If it make sense i can create pr

    def dynamic_allocated_value(metric, sub_metric = nil)
      uptime_index = ChargeableField.col_index('cpu_usage_rate_average') 
      case metric
      when 'derived_vm_numvcpus', 'derived_memory_available'
        @rollup_array.map do |rollup|
          rollup[uptime_index].nil? ? 0.0 : rollup[ChargeableField.col_index(metric)]
        end.compact.sum / consumed_hours_in_interval
      else
        avg(metric, sub_metric)
      end
    end

[agrare]

Depending on how accurate you need the uptime calculation to be that should work. A VM could have been on for 1 minute of that 1 hour rollup and it will be counted as on for that whole hour.

Not sure if you have access to the realtime metric datapoints from there, but those are a 20s resolution and thus cut this error down dramatically.

[ahrechushkin]

A VM could have been on for 1 minute of that 1 hour rollup and it will be counted as on for that whole hour.

I think it's an acceptable accuracy (other cloud providers have the same way for calculating)

[ahrechushkin]

@agrare @Fryguy guys, it make sense to pr or no?
https://github.com/ManageIQ/manageiq/issues/21144#issuecomment-817131787

[agrare]

@ahrechushkin definitely create a PR, we can discuss there.

I'd still like to investigate other approaches for more accurate uptime calculations but if +/- 1hr is close enough that's better than not having as long as we document that.