When working with metrics, the problem comes up that they may use different unit name but have the same unit in fact. There are a lot of real world examples like 'kB' and 'Kbyte'. In cc-metric-collector, the collectors read data from different sources which may use different units or the programmer specifies a unit for a metric by hand. The cc-units system is not comparable with the SI unit system. If you are looking for a package for the SI units, see here.
In order to enable unit comparison and conversion, the ccUnits package provides some helpers:
NewUnit(unit string) Unit // create a new unit from some string like 'GHz', 'Mbyte' or 'kevents/s'
func GetUnitUnitFactor(in Unit, out Unit) (func(value float64) float64, error) // Get conversion function between two units
func GetPrefixFactor(in Prefix, out Prefix) func(value float64) float64 // Get conversion function between two prefixes
func GetUnitPrefixFactor(in Unit, out Prefix) (func(value float64) float64, Unit) // Get conversion function for prefix changes and the new unit for further use
type Unit interface {
Valid() bool
String() string
Short() string
AddUnitDenominator(div Measure)
}
In order to get the "normalized" string unit back or test for validity, you can use:
u := NewUnit("MB")
fmt.Println(u.Valid()) // true
fmt.Printf("Long string %q", u.String()) // MegaBytes
fmt.Printf("Short string %q", u.Short()) // MBytes
v := NewUnit("foo")
fmt.Println(v.Valid()) // false
If you have two units or other components and need the conversion function:
// Get conversion functions for 'kB' to 'MBytes'
u1 := NewUnit("kB")
u2 := NewUnit("MBytes")
convFunc, err := GetUnitUnitFactor(u1, u2) // Returns an error if the units have different measures
if err == nil {
v2 := convFunc(v1)
fmt.Printf("%f %s\n", v2, u2.Short())
}
// Get conversion function for 'kB' -> 'G' prefix.
// Returns the function and the new unit 'GBytes'
p1 := NewPrefix("G")
convFunc, u_p1 := GetUnitPrefixFactor(u1, p1)
// or
// convFunc, u_p1 := GetUnitPrefixStringFactor(u1, "G")
if convFunc != nil {
v2 := convFunc(v1)
fmt.Printf("%f %s\n", v2, u_p1.Short())
}
// Get conversion function for two prefixes: 'G' -> 'T'
p2 := NewPrefix("T")
convFunc = GetPrefixPrefixFactor(p1, p2)
if convFunc != nil {
v2 := convFunc(v1)
fmt.Printf("%f %s -> %f %s\n", v1, p1.Prefix(), v2, p2.Prefix())
}
(In the ClusterCockpit ecosystem the separation between values and units if useful since they are commonly not stored as a single entity but the value is a field in the CCMetric while unit is a tag or a meta information).
If you have a metric and want the derivation to a bandwidth or events per second, you can use the original unit:
in_unit, err := metric.GetMeta("unit")
if err == nil {
value, ok := metric.GetField("value")
if ok {
out_unit = NewUnit(in_unit)
out_unit.AddUnitDenominator("seconds")
seconds := timeDiff.Seconds()
y, err := lp.New(metric.Name()+"_bw",
metric.Tags(),
metric.Meta(),
map[string]interface{"value": value/seconds},
metric.Time())
if err == nil {
y.AddMeta("unit", out_unit.Short())
}
}
}
Some used measures like Bytes and Flops are non-dividable. Consequently there prefixes like Milli, Micro and Nano are not useful. This is quite handy since a unit mb
for MBytes
is not uncommon but would by default be parsed as "MilliBytes".
Special parsing rules for the following measures: iff prefix==Milli
, use prefix==Mega
Bytes
Flops
Packets
Events
Cycles
Requests
This means the prefixes Micro
(like ubytes
) and Nano
like (nflops/sec
) are not allowed and return an invalid unit. But you can specify mflops
and mb
.
Prefixes for %
or percent
are ignored.
const (
Base Prefix = 1
Exa = 1e18
Peta = 1e15
Tera = 1e12
Giga = 1e9
Mega = 1e6
Kilo = 1e3
Milli = 1e-3
Micro = 1e-6
Nano = 1e-9
Kibi = 1024
Mebi = 1024 * 1024
Gibi = 1024 * 1024 * 1024
Tebi = 1024 * 1024 * 1024 * 1024
)
The prefixes are detected using a regular expression ^([kKmMgGtTpP]?[i]?)(.*)
that splits the prefix from the measure. You probably don't need to deal with the prefixes in the code.
const (
None Measure = iota
Bytes
Flops
Percentage
TemperatureC
TemperatureF
Rotation
Hertz
Time
Watt
Joule
Cycles
Requests
Packets
Events
)
There a regular expression for each of the measures like ^([bB][yY]?[tT]?[eE]?[sS]?)
for the Bytes
measure.
If the selected units are not suitable for your metric, feel free to send a PR.
For a new prefix, add it to the big const
in ccUnitPrefix.go
and adjust the prefix-unit-splitting regular expression. Afterwards, you have to add cases to the three functions String()
, Prefix()
and NewPrefix()
. NewPrefix()
contains the parser (k
or K
-> Kilo
). The other one are used for output. String()
outputs a longer version of the prefix (Kilo
), while Prefix()
returns only the short notation (K
).
Adding new prefixes is probably rare but adding a new measure is a more common task. At first, add it to the big const
in ccUnitMeasure.go
. Moreover, create a regular expression matching the measure (and pre-compile it like the others). Add the expression matching to NewMeasure()
. The String()
and Short()
functions return descriptive strings for the measure in long form (like Hertz
) and short form (like Hz
).
If there are special conversation rules between measures and you want to convert one measure to another, like temperatures in Celsius to Fahrenheit, a special case in GetUnitPrefixFactor()
is required.
The two parsers for prefix and measure are called under the hood by NewUnit()
and there might some special rules apply. Like in the above section about 'special unit detection', special rules for your new measure might be required. Currently there are two special cases:
- Measures that are non-dividable like Flops, Bytes, Events, ... cannot use
Milli
,Micro
andNano
. The prefixm
is forced toM
for these measures - If the prefix is
p
/P
(Peta
) ore
/E
(Exa
) and the measure is not detectable, it retries detection with the prefix. So first round it tries, for example, prefixp
and measureackets
which fails, so it retries the detection with measurepackets
and<empty>
prefix (resolves toBase
prefix).
The ccUnits
package is a simple implemtation of a unit system and comes with some limitations:
- The unit denominator (like
s
inMbyte/s
) can only have theBase
prefix, you cannot specifyByte/ms
for "Bytes per milli second".