getting usedReal greater than memTotalReal

[nuno-silva]

usedReal is getting bigger than memTotalReal, even with little swap usage:

Reverting #6 seems to fix it, so there may be a problem with it:

Here's some example data from my RPi Zero device:

.1.3.6.1.2.1.25.2.2.0  HOST-RESOURCES-MIB::hrMemorySize.0 = INTEGER: 443076 KBytes
.1.3.6.1.4.1.2021.4.5.0  UCD-SNMP-MIB::memTotalReal.0 = INTEGER: 443076 kB
.1.3.6.1.4.1.2021.4.6.0  UCD-SNMP-MIB::memAvailReal.0 = INTEGER: 29188 kB
.1.3.6.1.4.1.2021.4.3.0  UCD-SNMP-MIB::memTotalSwap.0 = INTEGER: 102396 kB
.1.3.6.1.4.1.2021.4.4.0  UCD-SNMP-MIB::memAvailSwap.0 = INTEGER: 98812 kB
.1.3.6.1.4.1.2021.4.14.0  UCD-SNMP-MIB::memBuffer.0 = INTEGER: 42620 kB
.1.3.6.1.4.1.2021.4.15.0  UCD-SNMP-MIB::memCached.0 = INTEGER: 302084 kB

root@zero:~# cat /proc/meminfo | grep -E "Mem|Swap[TF]|^Cache|Buff"
MemTotal:         443076 kB
MemFree:           29140 kB
MemAvailable:     318788 kB
Buffers:           42620 kB
Cached:           302084 kB
SwapTotal:        102396 kB
SwapFree:          98812 kB

Note that the new OID for totalReal used in #6 gives me exactly the same result, so the issue must be in the new computation of usedReal:

https://github.com/netniV/cacti-netsnmp-memory/blob/f00431317331ee3bb03a961b1050baca311f4716/scripts/ss_netsnmp_memory.php#L116-L117

Indeed, (443 076 - (29188 - 42620 - 302084)) = 758592, which is greater than my 443076 KB of RAM.

Contrary to https://github.com/netniV/cacti-netsnmp-memory/pull/6#issue-385059158, my availReal is not the result of /proc/meminfo MemAvailable + buffer + cache: 29188 != 318788+42620+302084 = 663492. Rather, memAvailReal seems to be meminfo's MemFree, so the old computation seems just fine.

[nuno-silva]

@gjimenezf can you perhaps try increasing the memory usage on one of your servers and see if this happens to you?

[nuno-silva]

Here's an example from another cacti instance:

• At ~02:37 I allocated 20GB of RAM.

• At ~02:48 I allocated 20 more GB of RAM.

• At ~3:00 I started increasing the amount of buffers by reading a really big file. You can see it makes the graph go further above memTotalReal, so it's clear that at least "buffers" are being stacked/summed twice in the graph: once by memBuffer and then again in the (wrong) computation of usedReal. Cache too, even though I did not test that.

• At ~3:18, I reverted #6. After this, the graph starts to look more like the memory usage reported by htop (green is used, blue is buffers, yellow is cache, gray is free):

[netniV]

So does this mean #6 needs reverting?

[nuno-silva]

Yes, at least partially. I would wait for @gjimenezf's input though. Maybe he's running some shiny new kernel or snmpd that changed way the values are reported and if that's the case more changes are needed.

[netniV]

We have had no feedback from @gjimenezf so, I guess its our call @nuno-silva?

[nuno-silva]

Yes, please revert #6.

Both UCD-SNMP-MIB::memTotalReal.0 (.1.3.6.1.4.1.2021.4.5.0) and HOST-RESOURCES-MIB::hrMemorySize.0 (.1.3.6.1.2.1.25.2.2.0) OIDs give the same value on all my host, so I think you can keep using memTotalReal for consistency with the other ones and only revert the bad calculations.

[netniV]

I have now reverted c3feb030e865c3da1ec0f76bdf5b3f84d1a2a3ec so that 29a27a0 should show as expected. Unless you also want memCached deleting not adding? But that seems to be the bug with his change.

[nuno-silva]

29a27a0 looks good. Thanks!