Fix CHEF-1835 -- child chef-chef now spawned using

fork/exec instead of popen4, so we don't use grep
shenanigans to determine its pid.

2 files changed, 55 insertions, 28 deletions

diff --git a/features/data/cookbooks/run_interval/recipes/default.rb b/features/data/cookbooks/run_interval/recipes/default.rb
index 4b7f432de2..d0e6c30d15 100644
--- a/features/data/cookbooks/run_interval/recipes/default.rb
+++ b/features/data/cookbooks/run_interval/recipes/default.rb
@@ -2,7 +2,7 @@
 # Cookbook Name:: run_interval
 # Recipe:: default
 #
-# Copyright 2009, Opscode
+# Copyright 2009, 2010, Opscode
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -16,3 +16,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
+
+# Force chef-client to exit once this cookbook has been applied twice.
+# The test depends on chef having run twice, so this number is tied to
+# run_interval.feature!
+
+$run_interval_global ||= 2
+
+$run_interval_global -= 1
+exit(2) if $run_interval_global == 0
diff --git a/features/steps/run_client_steps.rb b/features/steps/run_client_steps.rb
index 34e6ce2180..c8361ba773 100644
--- a/features/steps/run_client_steps.rb
+++ b/features/steps/run_client_steps.rb
@@ -23,16 +23,19 @@ include Chef::Mixin::ShellOut
 
 CHEF_CLIENT = File.join(CHEF_PROJECT_ROOT, "chef", "bin", "chef-client")
 
+def chef_client_command_string
+  @log_level ||= ENV["LOG_LEVEL"] ? ENV["LOG_LEVEL"] : "error"
+  @chef_args ||= ""
+  @config_file ||= File.expand_path(File.join(configdir, 'client.rb'))
+
+  "#{File.join(File.dirname(__FILE__), "..", "..", "chef", "bin", "chef-client")} -l #{@log_level} -c #{@config_file} #{@chef_args}"
+end
 
 ###
 # When
 ###
 When /^I run the chef\-client$/ do
-  @log_level ||= ENV["LOG_LEVEL"] ? ENV["LOG_LEVEL"] : "error"
-  @chef_args ||= ""
-  @config_file ||= File.expand_path(File.join(configdir, 'client.rb'))
-  status = Chef::Mixin::Command.popen4(
-    "#{File.join(File.dirname(__FILE__), "..", "..", "chef", "bin", "chef-client")} -l #{@log_level} -c #{@config_file} #{@chef_args}") do |p, i, o, e|
+  status = Chef::Mixin::Command.popen4(chef_client_command_string()) do |p, i, o, e|
     @stdout = o.gets(nil)
     @stderr = e.gets(nil)
   end
@@ -65,29 +68,44 @@ When /^I run the chef\-client with '(.+)' for '(.+)' seconds$/ do |args, run_for
 end
 
 When /^I run the chef\-client for '(.+)' seconds$/ do |run_for|
-  cid = Process.fork {
-    sleep run_for.to_i
-
-    client_pid = `ps ax | grep chef-client | grep -v grep | grep -v rake | grep -v cucumber | awk '{ print $1 }'`.to_i
-
-    # Send a SIGINT to the child process so it has a chance to cleanly exit,
-    # including flushing its stdout.
-    Process.kill("INT", client_pid)
-
-    sleep 1
-
-    # Send KILL to the child chef-client. Due to OHAI-223, where ohai sometimes
-    # ignores/doesn't exit correctly on receipt of SIGINT, brutally kill the
-    # subprocess.
-    begin
-      Process.kill("KILL", client_pid)
-    rescue Errno::ESRCH
-      # SIGINT worked above, so the KILL failed. Ignore it as this means things
-      # are working the way they're supposed to.
+  # Normal behavior depends on the run_interval/recipes/default.rb to count down
+  # and exit subordinate chef-client after two runs. However, we will forcably
+  # kill the client if that didn't work.
+  begin
+    stdout_filename = "/tmp/chef.run_interval.stdout.#{$$}.txt"
+    stderr_filename = "/tmp/chef.run_interval.stderr.#{$$}.txt"
+    client_pid = Process.fork do
+      STDOUT.reopen(File.open(stdout_filename, "w"))
+      STDERR.reopen(File.open(stderr_filename, "w"))
+      exec "#{chef_client_command_string}"
+      exit 2
     end
-  }
-  When 'I run the chef-client'
-  Process.waitpid2(cid)
+  
+    killer_pid = Process.fork {
+      sleep run_for.to_i
+
+      # Send KILL to the child chef-client. Due to OHAI-223, where ohai sometimes
+      # ignores/doesn't exit correctly on receipt of SIGINT, brutally kill the
+      # subprocess.
+      begin
+        Process.kill("KILL", client_pid)
+      rescue Errno::ESRCH
+        # Kill didn't work; the process exited while we were waiting, like
+        # it's supposed to.
+      end
+    }
+
+    Process.waitpid2(killer_pid)
+    @status = Process.waitpid2(client_pid).last
+    
+    # Read these in so they can be used in later steps.
+    @stdout = IO.read(stdout_filename)
+    @stderr = IO.read(stderr_filename)
+  ensure
+    # clean up after ourselves.
+    File.delete(stdout_filename)
+    File.delete(stderr_filename)
+  end
 end
 
 When /^I run the chef\-client at log level '(.+)'$/ do |log_level|