collectd: What are the units of the measurements captured by the disk, interface, irq, and swap_io plugins? - collectd

Question

What are the units of the measurements captured by the disk, interface, irq, and swap_io collectd plugins?
I'm comparing collectd 5 running on one machine with collectd 4 running on another, and trying to configure them both so they capture the same metrics.
disk reports fractional floats on collectd 5 (KiB?), integers on 4 (bytes).
interface is fractional floats on collectd 5 (KiB?), integer on 4 (bytes).
irq is apparently percentages on collectd 5, cumulative jiffies(?) on 4.
swap_io is floats (with an occasional 'nan') on collectd 5, integer on 4.
Additionally, the cpu plugin is supposed, according to the collectd wiki, to capture jiffies when ValuesPercentage is false and both ReportByCpu and ReportByState are true (their defaults). The collectd 4 cpu plugin does not offer these options and measures in jiffies. I set the collectd.conf of the version 5 instance to the settings I just mentioned (even explicitly setting defaults) and...it stubbornly continues to report percentages instead of jiffies!
Update: After taking a close look at the code for collectd 5, it has become clear that the collectd wiki is lying: when ValuesPercentage is false and both ReportByCpu and ReportByState are true, the cpu plugin reports rates of change and not cumulative jiffies (as in collectd 4).
Further Update: I had misinterpreted the collectd 4 behaviour for swap_io. Turns out the swap metrics are in bytes while swap_io is in pages (and you can't configure that), so the correct collectd 5 setting is ReportBytes = false and not ReportBytes = true. That's one less discrepancy.
collectd.conf for version 4:
# We're running collectd 4.10.9
# FQDNLookup since 4.3, became true by default with 5.0
FQDNLookup false
##
## Interval (in seconds) at which to query values.
## Starting with 5.4.3, this may be overridden on a per-plugin basis.
Interval 60
##
## Client part
##
## Logging (only one allowed)
##
# syslog since 4.0
LoadPlugin syslog
<Plugin syslog>
LogLevel info
</Plugin>
##
## Inputs
##
# cpu since 1.3
LoadPlugin cpu
# df since 3.6
LoadPlugin df
<Plugin df>
FSType tmpfs
IgnoreSelected true
# ReportByDevice since 4.8
ReportByDevice true
# ReportInodes false
# ReportReserved false by default; the option became true and was removed with version 5
ReportReserved true
# ValuesAbsolute, ValuesPercentage starting with collectd 5.4
</Plugin>
# disk since 1.5
LoadPlugin disk
# interface since 1.0
LoadPlugin interface
# irq since 4.0
LoadPlugin irq
# load since 1.0 (not sure about ReportRelative)
LoadPlugin load
<Plugin load>
ReportRelative true
</Plugin>
# memory since 1.0
# No options for 4.10
LoadPlugin memory
# nfs since 3.3
LoadPlugin nfs
# processes since 3.2
LoadPlugin processes
<Plugin processes>
# ProcessMatch since 4.5
ProcessMatch "all" "(.*)"
</Plugin>
# protocols since 4.7
LoadPlugin protocols
# swap since 2.1
# No options for 4.10
LoadPlugin swap
<Plugin swap>
# 4.10 reports swap I/O in pages (and cannot be configured otherwise)
# 4.10 reports swap cached/free/used in bytes
</Plugin>
# tcpconns since 4.2
LoadPlugin tcpconns
# thermal since 4.5
LoadPlugin thermal
# uptime since 4.7
LoadPlugin uptime
##
## Server part
##
## Output (only one allowed)
##
# csv since 4.0
LoadPlugin csv
<Plugin csv>
DataDir "/var/collectd/csv"
# StoreRates since 4.3
StoreRates false
</Plugin>
# write_graphite since 5.1
# write_http since 4.8
collectd.conf for version 5:
# We are running collectd 5.4.0.git
# FQDNLookup since 4.3, became true by default with 5.0
FQDNLookup false
# Interval (in seconds) at which to query values. This may be overwritten on
# a per-plugin basis by using the 'Interval' option of the LoadPlugin block.
# This capability was announced with version 5.2 but became functional
# only with 5.4.3.
Interval 60
# Logging
# syslog since 4.0
LoadPlugin syslog
<Plugin syslog>
LogLevel info
</Plugin>
# LoadPlugin section
# cpu since 1.3
LoadPlugin cpu
<Plugin cpu>
# ReportByCpu true
# ReportByState true
# ValuesPercentage starting with collectd 5.5
# ValuesPercentage true
# ValuesPercentage false forces measurements in jiffies/second when both
# ReportByCpu and ReportByState are true (the defaults); collectd 4 does
# not do time derivation (and time derivation cannot be turned off in 5)
ValuesPercentage false
</Plugin>
# df since 3.6
LoadPlugin df
<Plugin df>
FSType rootfs
FSType sysfs
FSType proc
FSType devtmpfs
FSType devpts
FSType tmpfs
FSType fusectl
FSType cgroup
IgnoreSelected true
# ReportByDevice since 4.8
ReportByDevice true
# ValuesAbsolute, ValuesPercentage since 5.4
# ValuesAbsolute for reporting in bytes (true by default)
# ValuesAbsolute true
ValuesPercentage false
</Plugin>
# disk since 1.5
LoadPlugin disk
# irq since 4.0
LoadPlugin irq
# load since 1.0 (not sure about ReportRelative)
LoadPlugin load
<Plugin load>
ReportRelative true
</Plugin>
# memory since 1.0
# No options for 4.10
LoadPlugin memory
<Plugin memory>
ValuesAbsolute true
ValuesPercentage false
</Plugin>
# nfs since 3.3
LoadPlugin nfs
# processes since 3.2
LoadPlugin processes
<Plugin processes>
# ProcessMatch since 4.5
ProcessMatch "all" "(.*)"
</Plugin>
# protocols since 4.7
LoadPlugin protocols
# swap since 2.1
# No options for 4.10
LoadPlugin swap
<Plugin swap>
ReportByDevice true
# ReportIO true
# ReportBytes is false by default; when false, swap I/O is in pages
# 4.10 reports swap I/O in pages (and cannot be configured otherwise)
ReportBytes false
ValuesAbsolute true
ValuesPercentage false
# 4.10 reports swap cached/free/used in bytes, swap_io in/out in pages
</Plugin>
# tcpconns since 4.2
LoadPlugin tcpconns
# thermal since 4.5
LoadPlugin thermal
# thermal since 4.7
LoadPlugin uptime
# Server part
LoadPlugin write_graphite
<Plugin write_graphite>
<Node "node-graphite-1">
Host "192.168.1.170"
Port "1111"
Protocol "tcp"
EscapeCharacter "_"
AlwaysAppendDS true
SeparateInstances false
</Node>
</Plugin>

Answer 1

disk reports cumulative operations (ops) and also cumulative bytes (Io).
interface reports cumulative packet and error counts on the one hand, and cumulative bytes on the other.
swap_io reports cumulative operations.
Can you post your exact full config please, so we can troubleshoot your collectd5 CPU config?

Answer 2

Well, that turned out to be a simple, stupid mistake. The csv plugin used on one side has a default for StoreRates of false, while the graphite plugin used on the other has a default of true, and that was why some (but not all) metrics were integer (cumulative counts) while others were floats (rates of increase).