MemUsed metric includes Buffered/Cached memory

[BestFriendChris]

In the Memory check, physUsed is calculated by MemTotal - MemFree as shown below

physTotal = int(meminfo['MemTotal'])
physFree = int(meminfo['MemFree'])
physUsed = physTotal - physFree

(found at checks/system.py)

This makes it awkward (in Datadog graphs) to differentiate between memory being used by the cache and memory being used by normal processes.

I'd like to change the physUsed definition to be

physTotal = int(meminfo['MemTotal'])
physFree = int(meminfo['MemFree'])
physBuffers = int(meminfo['Buffers'])
physCached = int(meminfo['Cached'])
physUsed = physTotal - physFree - physBuffers - physCached

Alternatively, we could add another data point that represents physUsed - physBuffers - physCached. I'm at a loss on what to call it however.

One side point; free deals with this problem by showing both:

             total       used       free     shared    buffers     cached
Mem:           512        494         17          0         76        326
-/+ buffers/cache:         91        420

The -/+ buffers/cache line represents used - buffers - cached and free + buffers + cached

Chris

[alq666]

@BestFriendChris I'm going to cut a new metric for "usable" memory: free + buffers + cached. I'm cleaning up that code too as it's a bit rusty.

[BestFriendChris]

@alq666 Cool. My ultimate goal is to have a stacked line graph that contains (my definition of) used, buffers, cached, free

[alq666]

We released v2.2.9 last week. It supports additional memory metrics.