In Powershell, what's the most efficient way to split a large text file by record type?

Question

I am using Powershell for some ETL work, reading compressed text files in and splitting them out depending on the first three characters of each line.

If I were just filtering the input file, I could pipe the filtered stream to Out-File and be done with it. But I need to redirect the output to more than one destination, and as far as I know this can't be done with a simple pipe. I'm already using a .NET streamreader to read the compressed input files, and I'm wondering if I need to use a streamwriter to write the output files as well.

The naive version looks something like this:

while (!$reader.EndOfFile) {
  $line = $reader.ReadLine();
  switch ($line.substring(0,3) {
    "001" {Add-Content "output001.txt" $line}
    "002" {Add-Content "output002.txt" $line}
    "003" {Add-Content "output003.txt" $line}
    }
  }

That just looks like bad news: finding, opening, writing and closing a file once per row. The input files are huge 500MB+ monsters.

Is there an idiomatic way to handle this efficiently w/ Powershell constructs, or should I turn to the .NET streamwriter?

Are there methods of a (New-Item "path" -type "file") object I could use for this?

EDIT for context:

I'm using the DotNetZip library to read ZIP files as streams; thus streamreader rather than Get-Content/gc. Sample code:

[System.Reflection.Assembly]::LoadFrom("\Path\To\Ionic.Zip.dll") 
$zipfile = [Ionic.Zip.ZipFile]::Read("\Path\To\File.zip")

foreach ($entry in $zipfile) {
  $reader = new-object system.io.streamreader $entry.OpenReader();
  while (!$reader.EndOfFile) {
    $line = $reader.ReadLine();
    #do something here
  }
}

I should probably Dispose() of both the $zipfile and $reader, but that is for another question!

Answer 1

Reading

As for reading the file and parsing, I would go with switch statement:

switch -file c:\temp\stackoverflow.testfile2.txt -regex {
  "^001" {Add-Content c:\temp\stackoverflow.testfile.001.txt $_}
  "^002" {Add-Content c:\temp\stackoverflow.testfile.002.txt $_}
  "^003" {Add-Content c:\temp\stackoverflow.testfile.003.txt $_}
}

I think it is better approach because

• there is support for regex, you don't have to make substring (which might be expensive) and
• the parameter -file is quite handy ;)

Writing

As for writing the output, I'll test to use streamwriter, however if performance of Add-Content is decent for you, I would stick to it.

Added: Keith proposed to use >> operator, however, it seems that it is very slow. Besides that it writes output in Unicode which doubles the file size.

Look at my test:

[1]: (measure-command {
>>     gc c:\temp\stackoverflow.testfile2.txt  | %{$c = $_; switch ($_.Substring(0,3)) {
>>             '001'{$c >> c:\temp\stackoverflow.testfile.001.txt} `
>>             '002'{$c >> c:\temp\stackoverflow.testfile.002.txt} `
>>             '003'{$c >> c:\temp\stackoverflow.testfile.003.txt}}}
>> }).TotalSeconds
>>
159,1585874
[2]: (measure-command {
>>     gc c:\temp\stackoverflow.testfile2.txt  | %{$c = $_; switch ($_.Substring(0,3)) {
>>             '001'{$c | Add-content c:\temp\stackoverflow.testfile.001.txt} `
>>             '002'{$c | Add-content c:\temp\stackoverflow.testfile.002.txt} `
>>             '003'{$c | Add-content c:\temp\stackoverflow.testfile.003.txt}}}
>> }).TotalSeconds
>>
9,2696923

The difference is huge.

Just for comparison:

[3]: (measure-command {
>>     $reader = new-object io.streamreader c:\temp\stackoverflow.testfile2.txt
>>     while (!$reader.EndOfStream) {
>>         $line = $reader.ReadLine();
>>         switch ($line.substring(0,3)) {
>>             "001" {Add-Content c:\temp\stackoverflow.testfile.001.txt $line}
>>             "002" {Add-Content c:\temp\stackoverflow.testfile.002.txt $line}
>>             "003" {Add-Content c:\temp\stackoverflow.testfile.003.txt $line}
>>             }
>>         }
>>     $reader.close()
>> }).TotalSeconds
>>
8,2454369
[4]: (measure-command {
>>     switch -file c:\temp\stackoverflow.testfile2.txt -regex {
>>         "^001" {Add-Content c:\temp\stackoverflow.testfile.001.txt $_}
>>         "^002" {Add-Content c:\temp\stackoverflow.testfile.002.txt $_}
>>         "^003" {Add-Content c:\temp\stackoverflow.testfile.003.txt $_}
>>     }
>> }).TotalSeconds
8,6755565

Added: I was curious about the writing performance .. and I was a little bit surprised

[8]: (measure-command {
>>     $sw1 = new-object io.streamwriter c:\temp\stackoverflow.testfile.001.txt3b
>>     $sw2 = new-object io.streamwriter c:\temp\stackoverflow.testfile.002.txt3b
>>     $sw3 = new-object io.streamwriter c:\temp\stackoverflow.testfile.003.txt3b
>>     switch -file c:\temp\stackoverflow.testfile2.txt -regex {
>>         "^001" {$sw1.WriteLine($_)}
>>         "^002" {$sw2.WriteLine($_)}
>>         "^003" {$sw3.WriteLine($_)}
>>     }
>>     $sw1.Close()
>>     $sw2.Close()
>>     $sw3.Close()
>>
>> }).TotalSeconds
>>
0,1062315

It is 80 times faster. Now you you have to decide - if speed is important, use StreamWriter. If code clarity is important, use Add-Content.

---

Substring vs. Regex

According to Keith Substring is 20% faster. It depends, as always. However, in my case the results are like this:

[102]: (measure-command {
>>     gc c:\temp\stackoverflow.testfile2.txt  | %{$c = $_; switch ($_.Substring(0,3)) {
>>             '001'{$c | Add-content c:\temp\stackoverflow.testfile.001.s.txt} `
>>             '002'{$c | Add-content c:\temp\stackoverflow.testfile.002.s.txt} `
>>             '003'{$c | Add-content c:\temp\stackoverflow.testfile.003.s.txt}}}
>> }).TotalSeconds
>>
9,0654496
[103]: (measure-command {
>>     gc c:\temp\stackoverflow.testfile2.txt  | %{$c = $_; switch -regex ($_) {
>>             '^001'{$c | Add-content c:\temp\stackoverflow.testfile.001.r.txt} `
>>             '^002'{$c | Add-content c:\temp\stackoverflow.testfile.002.r.txt} `
>>             '^003'{$c | Add-content c:\temp\stackoverflow.testfile.003.r.txt}}}
>> }).TotalSeconds
>>
9,2563681

So the difference is not important and for me, regexes are more readable.

Answer 2

Given the size of input files, you definitely want to process a line at a time. I wouldn't think the re-opening/closing of the output files would be too huge a perf hit. It certainly makes the implemation possible using the pipeline even as a one-liner - really not too different from your impl. I wrapped it here to get rid of the horizontal scrollbar:

gc foo.log | %{switch ($_.Substring(0,3)) {
    '001'{$input | out-file output001.txt -enc ascii -append} `
    '002'{$input | out-file output002.txt -enc ascii -append} `
    '003'{$input | out-file output003.txt -enc ascii -append}}}