Partly thanks to the fact that PowerShell is able to invoke .Net methods, it offers quiet some possibilities to filter objects. At stackoverflow you will find a lot of (PowerShell) questions and answers measuring the performance of a specific extricated command or cmdlet. This usually leaves wrong impression as the performance of a complete (PowerShell) solution is supposed to be better than the sum of its parts. Each command is depended on the expected input and - output. Especially when using the PowerShell pipeline, commands (cmdlets) interact with prior commands and commands that follow. Therefore it is important to look at the bigger picture and understand how and where each command gains its performance.
This means that I can't tell which command you should choose, but with a better understanding of the commands and concepts listed below, I hope you are better able to find "fastest way" for your specific solution.
Language Integrated Query (LINQ) is often (dis)qualified as the fasted solution to filter objects in PowerShell (see also High Performance PowerShell with LINQ):
(Measure-Command {
$Result = [Linq.Enumerable]::Where($array, [Func[object,bool]] { param($Item); return $Item.Index -eq 43122 })
}).totalmilliseconds
4.0715
Just over 4ms
!, none of the other methods can ever beat that...
But before jumping into any conclusions that LINQ beats any other method by a factor 100 or more you should keep the following in mind. There are two pitfalls in measuring the performance of a LINQ query when you just look at the performance of the activity itself:
- LINQ has a big cache, meaning that you should restart a new PowerShell session to measure the actual results (or just not, if you often want to reuse the query). After restarting the PowerShell session, you will find that it will take about 6 times longer to initiate the LINQ query.
- But more importantly, LINQ performs lazy evaluation (also called deferred execution). This means that actually nothing has been done yet other than defining what should be done. This actually shows if you want to access one of the properties of the
$Result
:
(Measure-Command {
$Result.Property1
}).totalmilliseconds
532.366
Where it usually takes about 15ms
to retrieve a property of a single object:
$Item = [PSCustomObject]@{Index=1; Property1='Hello!'; Property2=(Get-Random)}
(Measure-Command {
$Item.Property1
}).totalmilliseconds
15.3708
Bottom line, you need to instantiate the results to correctly measure the performance of a LINQ query (for this, let's just retrieve one of the properties of the returned object within the measurement):
(Measure-Command {
$Result = ([Linq.Enumerable]::Where($array, [Func[object,bool]] { param($Item); return $Item.Index -eq 43122 })).Property1
}).totalmilliseconds
570.5087
(which is still fast.)
Hash tables are generally fast because they are based on a binary search algorithm, this means that you maximal have to guess ln 50000 / ln 2 = 16 times
to find your object. Nevertheless, building a HashTabe
for a single lookup is a little over done. But if you control the contruction of the object list, you might construct the hash table on the go:
(Measure-Command {
$ht = @{}
$array = @(1..50000).foreach{$ht[$PSItem] = [PSCustomObject]@{Index=$PSItem;Property1='Hello!';Property2=(Get-Random)}}
$ht.43122
}).totalmilliseconds
3415.1196
vs:
(Measure-Command {
$array = @(1..50000).foreach{[PSCustomObject]@{Index=$PSItem;Property1='Hello!';Property2=(Get-Random)}}
$ht = @{}; $array.foreach{$ht[$PSItem.index] = $psitem}
$ht.43122
}).totalmilliseconds
3969.6451
As you might already have concluded yourself the Where
method appears about twice as fast then the Where-Object
cmdlet:
Where-Object
cmdlet:
(Measure-Command {
$Result = $Array | Where-Object index -eq 43122
}).totalmilliseconds
721.545
Where
method:
(Measure-Command {
$Result = $Array.Where{$_ -eq 43122}
}).totalmilliseconds
319.0967
The reason for that is because the Where
command requires you load the whole array into memory which is actually not required for the Where-Object
cmdlet. If the data is already in memory (e.g. by assigning it to a variable $array = ...
) it is not be a big deal but this might actually a disadvantage by itself: except that it consumes memory, you have to wait until all objects are received before you can start filtering...
Don't underestimate the power of the PowerShell cmdlets like Where-Object
especially look to the solution as a whole in combination with the pipeline. As shown above, if you just measure on the specific action you might find these cmdlets slow but if you measure your whole end-to-end solution you might find that there isn't much difference and that cmdlets might even outperform methods other techniques. Where LINQ queries are extremely reactive, PowerShell cmdlets are extremely proactive.
In general, if your input is not yet in memory and supplied via the pipeline, you should try to continue to build on that pipeline and avoid stalling it in any way by avoiding variables assignments ($array = ...
) and the use of brackets ((...)
) :
Presume that your objects come from a slower input, in that case all the other solutions need to wait for the very last object to be able start filtering where the Where-Object
has already filtered most of the objects on the fly and as soon it has found it, is indeterminately passed to the next cmdlet...
For example let's presume that the data comes from a csv
file rather then memory...
$Array | Export-Csv .Test.csv
Where-Object
cmdlet:
(Measure-Command {
Import-Csv -Path .Test.csv | Where-Object index -eq 43122 | Export-Csv -Path .Result.csv
}).totalmilliseconds
717.8306
Where
method:
(Measure-Command {
$Array = Import-Csv -Path .Test.csv
Export-Csv -Path .Result.csv -InputObject $Array.Where{$_ -eq 43122}
}).totalmilliseconds
747.3657
This is just a single test example, but in most cases where the data isn't instantly available in memory, Where-Object streaming appears to be a often faster then using the Where method.
Besides, the Where
method uses a lot more memory which might make performance even worse if your file (list of objects) size exceeds the available physical memory. (See also: Can the following Nested foreach loop be simplified in PowerShell?).
Instead of using the Where-Object
cmdlet or the Where
method, you might consider to iterate through all the objects and just compare them with an If
statement. Before going into depth of this approach it is worth mentioning that comparison operators already iterate through the left argument by itself, quote:
When the input to an operator is a scalar value, comparison operators
return a Boolean value. When the input is a collection of values, the
comparison operators return any matching values. If there are no
matches in a collection, comparison operators return an empty array.
This means that if you just want to know whether the object with the specific property exists and don't care about the object itself, you might just simply compare the specific property collection:
(Measure-Command {
If ($Array.Index -eq 43122) {'Found object with the specific property value'}
}).totalmilliseconds
55.3483
For the ForEach-Object
cmdlet and the ForEach
method, you will see that the approach just takes a little longer then using their counterparts (Where-Object
cmdlet and the Where
method) as there is a little more overhead for the embedded comparison:
Directly from memory:
ForEach-Object
cmdlet:
(Measure-Command {
$Result = $Array | ForEach-Object {If ($_.index -eq 43122) {$_}}
}).totalmilliseconds
1031.1599
ForEach
method:
(Measure-Command {
$Result = $Array.ForEach{If ($_.index -eq 43122) {$_}}
}).totalmilliseconds
781.6769
Streaming from disk:
ForEach-Object
cmdlet:
(Measure-Command {
Import-Csv -Path .Test.csv |
ForEach-Object {If ($_.index -eq 43122)