Why is Ruby's popen3 crashing because "Too many files open"?

Question

I am using Popen3 to run some Perl scrips and then dump their output into a text file. Once in the text file, I search for the result of the Perl script. I get the error after running for about 40 minutes, which is about 220 files.

ruby/1.8/open3.rb:49:in `pipe': Too many open files (Errno::EMFILE)
    from /ruby/1.8/open3.rb:49:in `popen3'
    from ./RunAtfs.rb:9
    from ./RunAtfs.rb:8:in `glob'
    from ./RunAtfs.rb:8

The script is below.

require 'logger'
require 'open3'
atfFolder = ARGV[0]
testResult = ARGV[1]
res = "result.txt"
open('result.txt', 'w') { }
Dir.glob(atfFolder+'/*.pl') do |atfTest|
 Open3.popen3("atf.pl -c run-config.pl -t #{atfTest}") do |i, o, e, t|
   while line = e.gets
     $testFile = testResult + line[/^[0-9]+$/].to_s + "testOutput.txt"
      log = Logger.new($testFile)
      log.info(line)
      end
    log.close
end
 lastLine = `tail +1 #{$testFile}`
 file = File.open(res, 'a')
 if(lastLine.include? "(PASSED)")
    file.puts("Test #{atfTest} --> Passed")
    file.close
    File.delete($testFile)
 else
    file.puts("Test #{atfTest} --> Failed!")
    file.close
 end
end

This script is processing 4900 Perl files so I don't know if that is just too many files for popen3 or I am not using it correctly.

Thanks for helping me!

I refactored my script after some very helpful pointers! The code is working great!

require 'open3'

atf_folder, test_result = ARGV[0, 2]
File.open('result.txt', 'w') do |file| end

Dir.glob("#{ atf_folder }/*.pl") do |atf_test|

test_file = atf_test[/\/\w+.\./][1..-2].to_s + ".txt"
comp_test_path = test_result + test_file
File.open(comp_test_path, 'w') do |file| end

Open3.popen3("atf.pl -c run-config.pl -t #{ atf_test }") do |i, o, e, t|

    while line = e.gets

      File.open(comp_test_path, 'a') do |file|
        file.puts(line)
      end
    end
end

last_line = `tail +1 #{comp_test_path}`
File.open('result.txt', 'a') do |file|

    output_str = if (last_line.include? "(PASSED)")

    File.delete(comp_test_path)

    "Passed"

    else

    "Failed!"

end

file.puts "Test #{ atf_test } --> #{ output_str }"

end
end

Answer 1

Consider this:

require 'logger'
require 'open3'

atf_folder, test_result = ARGV[0, 2]

Dir.glob("#{ atf_folder }/*.pl") do |atf_test|

  Open3.popen3("atf.pl -c run-config.pl -t #{ atf_test }") do |i, o, e, t|

    while line = e.gets
      $testFile = test_result + line[/^[0-9]+$/].to_s + "testOutput.txt"
      log = Logger.new($testFile)
      log.info(line)
      log.close
    end

  end

  lastLine = `tail +1 #{ $testFile }`
  File.open('result.txt', 'a') do |file|

    output_str = if (lastLine.include? "(PASSED)")

                  File.delete($testFile)

                  "Passed"

                else

                  "Failed!"

                end

    file.puts "Test #{ atf_test } --> #{ output_str }"

  end

end

It's untested, of course, since there's no sample data, but it's written more idomatically for Ruby.

Things to note:

• atf_folder, test_result = ARGV[0, 2] slices the ARGV array and uses parallel assignment to retrieve both parameters at once. You should test to see that you got values for them. And, as you move to more complex scripts, take advantage of the OptionParser class that comes in Ruby's STDLIB.

• Ruby lets us pass a block to File.open, which automatically closes the file when the block exits. This is a major strength of Ruby and helps reduce errors like you're seeing. Logger doesn't do that, so extra care has to be take to avoid leaving hanging file-handles, like you're doing. Instead, use:

    log = Logger.new($testFile)
    log.info(line)
    log.close
  

  to immediately close the handle. You are doing it outside the loop, not inside it, so you had a bunch of open handles.

  Also consider whether you need Logger, or if a regular File.open would suffice. Logger has additional overhead.

• Your use of $testFile is questionable. $variables are globals, and their use is generally an indicator you're doing something wrong, at least until you understand why and when you should use them. I'd refactor the code using that.
• In Ruby, variables and methods are in snake_case, not CamelCase, which is used for classes and modules. That doesn't seem like much until_you_run_into CodeDoingTheWrongThing and_have_to_read_it. (Notice how your brain bogged down deciphering the camel-case?)

In general, I question whether this is the fastest way to do what you want. I suspect you could write a shell script using grep or tail that would at least keep up, and maybe run faster. You might sit down with your sysadmin and do some brain-pickin'.