childprocess
This gem aims at being a simple and reliable solution for controlling external programs running in the background on any Ruby / OS combination.
The code originated in the selenium-webdriver gem, but should prove useful as a standalone library.
Requirements
- Ruby 2.4+, JRuby 9+
Windows users must ensure the ffi
gem (>= 1.0.11
) is installed in order to use ChildProcess.
Usage
The object returned from ChildProcess.build
will implement ChildProcess::AbstractProcess
.
Basic examples
process = ChildProcess.build("ruby", "-e", "sleep")
# inherit stdout/stderr from parent...
process.io.inherit!
# ...or pass an IO
process.io.stdout = Tempfile.new("child-output")
# modify the environment for the child
process.environment["a"] = "b"
process.environment["c"] = nil
# set the child's working directory
process.cwd = '/some/path'
# start the process
process.start
# check process status
process.alive? #=> true
process.exited? #=> false
# wait indefinitely for process to exit...
process.wait
process.exited? #=> true
# get the exit code
process.exit_code #=> 0
# ...or poll for exit + force quit
begin
process.poll_for_exit(10)
rescue ChildProcess::TimeoutError
process.stop # tries increasingly harsher methods to kill the process.
end
Advanced examples
Output to pipe
r, w = IO.pipe
begin
process = ChildProcess.build("sh" , "-c",
"for i in {1..3}; do echo $i; sleep 1; done")
process.io.stdout = w
process.start # This results in a fork, inheriting the write end of the pipe.
# Close parent's copy of the write end of the pipe so when the (forked) child
# process closes its write end of the pipe the parent receives EOF when
# attempting to read from it. If the parent leaves its write end open, it
# will not detect EOF.
w.close
thread = Thread.new do
begin
loop do
print r.readpartial(16384)
end
rescue EOFError
# Child has closed the write end of the pipe
end
end
process.wait
thread.join
ensure
r.close
end
Note that if you just want to get the output of a command, the backtick method on Kernel may be a better fit.
Write to stdin
process = ChildProcess.build("cat")
out = Tempfile.new("duplex")
out.sync = true
process.io.stdout = process.io.stderr = out
process.duplex = true # sets up pipe so process.io.stdin will be available after .start
process.start
process.io.stdin.puts "hello world"
process.io.stdin.close
process.poll_for_exit(exit_timeout_in_seconds)
out.rewind
out.read #=> "hello world\n"
Pipe output to another ChildProcess
search = ChildProcess.build("grep", '-E', %w(redis memcached).join('|'))
search.duplex = true # sets up pipe so search.io.stdin will be available after .start
search.io.stdout = $stdout
search.start
listing = ChildProcess.build("ps", "aux")
listing.io.stdout = search.io.stdin
listing.start
listing.wait
search.io.stdin.close
search.wait
Prefer posix_spawn on *nix
If the parent process is using a lot of memory, fork+exec
can be very expensive. The posix_spawn()
API removes this overhead.
ChildProcess.posix_spawn = true
process = ChildProcess.build(*args)
To be able to use this, please make sure that you have the ffi
gem installed.
Ensure entire process tree dies
By default, the child process does not create a new process group. This means there's no guarantee that the entire process tree will die when the child process is killed. To solve this:
process = ChildProcess.build(*args)
process.leader = true
process.start
Detach from parent
process = ChildProcess.build("sleep", "10")
process.detach = true
process.start
Invoking a shell
As opposed to Kernel#system
, Kernel#exec
et al., ChildProcess will not automatically execute your command in a shell (like /bin/sh
or cmd.exe
) depending on the arguments.
This means that if you try to execute e.g. gem executables (like bundle
or gem
) or Windows executables (with .com
or .bat
extensions) you may see a ChildProcess::LaunchError
.
You can work around this by being explicit about what interpreter to invoke:
ChildProcess.build("cmd.exe", "/c", "bundle")
ChildProcess.build("ruby", "-S", "bundle")
Log to file
Errors and debugging information are logged to $stderr
by default but a custom logger can be used instead.
logger = Logger.new('logfile.log')
logger.level = Logger::DEBUG
ChildProcess.logger = logger
Caveats
- With JRuby on Unix, modifying
ENV["PATH"]
before using childprocess could lead to 'Command not found' errors, since JRuby is unable to modify the environment used for PATH searches injava.lang.ProcessBuilder
. This can be avoided by settingChildProcess.posix_spawn = true
. - With JRuby on Java >= 9, the JVM may need to be configured to allow JRuby to access neccessary implementations; this can be done by adding
--add-opens java.base/java.io=org.jruby.dist
and--add-opens java.base/sun.nio.ch=org.jruby.dist
to theJAVA_OPTS
environment variable that is used by JRuby when launching the JVM.
Implementation
How the process is launched and killed depends on the platform:
- Unix :
fork + exec
(orposix_spawn
if enabled) - Windows :
CreateProcess()
and friends - JRuby :
java.lang.{Process,ProcessBuilder}
Note on Patches/Pull Requests
- Fork it
- Create your feature branch (off of the development branch)
git checkout -b my-new-feature dev
- Commit your changes
git commit -am 'Add some feature'
- Push to the branch
git push origin my-new-feature
- Create new Pull Request
Publishing a New Release
When publishing a new gem release:
- Ensure latest build is green on the
dev
branch - Ensure CHANGELOG is updated
- Ensure version is bumped following Semantic Versioning
- Merge the
dev
branch intomaster
:git checkout master && git merge dev
- Ensure latest build is green on the
master
branch - Build gem from the green
master
branch:git checkout master && gem build childprocess.gemspec
- Push gem to RubyGems:
gem push childprocess-<VERSION>.gem
- Tag commit with version, annotated with release notes:
git tag -a <VERSION>
Copyright
Copyright (c) 2010-2015 Jari Bakken. See LICENSE for details.