switchman 1.5.15 → 1.5.16

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: bd0d33eb675d2b10b4c3f4c58cd70664fffc02d9
-  data.tar.gz: ceb9f08769064d7c046b79c2d4570680d835302a
+  metadata.gz: 211d4cd37f40dcadb51a8b83fb6d3fd85462c664
+  data.tar.gz: 70f7d6ec5c7b2f201398f22b7f0bcaf55805b368
 SHA512:
-  metadata.gz: bb26c88cfc2df081bdfc3d3ca9e2f3fb0c683f5a1ea5f9174229de2f0f4a6b2c23397f061711db2c6a02a5a718f1d91c1ffda9c627f3264cd443d7d4697587ed
-  data.tar.gz: d8d0e4ce64c6e80d4f8df2c47f770947eaeb2f0e10b06e34ca265ee2642d5b87d5a03ec9a900a1d75f3c3b45c1f4285cd0e8074d8ccfbc050d1b1d10ddc0b4d5
+  metadata.gz: 9736326b9ee5c92bf46ac6fdecf789ab969bcfa697f1f42ec758f243b97801ad3e075dafeef8fbe89c377567f427b4f56d1b612d0accc3a66f1377970dc14708
+  data.tar.gz: c40ac77605fb1d865c73fbf885ee438b1d1d66372eacdd4e107dff4bf99a099f519667f7b33d4ee546a5a1cdd18020544ff009c2a301ee495e61d53a7468a05a

data/app/models/switchman/shard_internal.rb

@@ -1,6 +1,7 @@
 require 'switchman/database_server'
 require 'switchman/default_shard'
 require 'switchman/environment'
+require 'switchman/errors'
 
 module Switchman
   class Shard < ::ActiveRecord::Base
@@ -151,6 +152,7 @@ module Switchman
       #                sub-processes per database server. Note that parallel
       #                invocation currently uses forking, so should be used sparingly
       #                because errors are not raised, and you cannot get results back
+      #    :max_procs - only run this many parallel processes at a time
       #    :exception - :ignore, :raise, :defer (wait until the end and raise the first
       #                error), or a proc
       def with_each_shard(*args)
@@ -225,10 +227,13 @@ module Switchman
             end
           end
 
-          fd_to_name_map = {}
+          exception_pipes = []
+          pids = []
           out_fds = []
           err_fds = []
-          pids = []
+          pid_to_name_map = {}
+          fd_to_name_map = {}
+          errors = []
 
           wait_for_output = lambda do |out_fds, err_fds, fd_to_name_map|
             ready, _ = IO.select(out_fds + err_fds)
@@ -244,7 +249,6 @@ module Switchman
             end
           end
 
-          exception_pipe = IO.pipe
           scopes.each do |server, subscopes|
             if !(::ActiveRecord::Relation === subscopes.first) && subscopes.first.class != Array
               subscopes = [subscopes]
@@ -263,55 +267,69 @@ module Switchman
                 name = server.id
               end
 
-              details = Open4.pfork4(lambda do
+              exception_pipe = IO.pipe
+              exception_pipes << exception_pipe
+              pid, io_in, io_out, io_err = Open4.pfork4(lambda do
                 begin
                   ::ActiveRecord::Base.clear_all_connections!
                   Switchman.config[:on_fork_proc].try(:call)
                   $0 = [$0, ARGV, name].flatten.join(' ')
                   with_each_shard(subscope, categories, options) { yield }
+                  exception_pipe.last.close
                 rescue Exception => e
-                  exception_pipe.last.write(Marshal.dump(e))
+                  Marshal.dump(e, exception_pipe.last)
                   exception_pipe.last.flush
-                  exit 1
+                  exception_pipe.last.close
+                  exit! 1
                 end
               end)
-              # don't care about writing to stdin
-              details[1].close
-              out_fds << details[2]
-              err_fds << details[3]
-              pids << details[0]
-              fd_to_name_map[details[2]] = name
-              fd_to_name_map[details[3]] = name
+              exception_pipe.last.close
+              pids << pid
+              io_in.close # don't care about writing to stdin
+              out_fds << io_out
+              err_fds << io_err
+              pid_to_name_map[pid] = name
+              fd_to_name_map[io_out] = name
+              fd_to_name_map[io_err] = name
 
               while max_procs && pids.count >= max_procs
                 while max_procs && out_fds.count >= max_procs
                   # wait for output if we've hit the max_procs limit
                   wait_for_output.call(out_fds, err_fds, fd_to_name_map)
                 end
-                pids.delete(Process.wait) # we've gotten all the output from one fd so wait for its child process to exit
+                # we've gotten all the output from one fd so wait for its child process to exit
+                found_pid, status = Process.wait2
+                pids.delete(found_pid)
+                errors << pid_to_name_map[found_pid] if status.exitstatus != 0
               end
             end
           end
 
-          exception_pipe.last.close
-
           while out_fds.any? || err_fds.any?
             wait_for_output.call(out_fds, err_fds, fd_to_name_map)
           end
-          pids.each { |pid| Process.waitpid2(pid) }
+          pids.each do |pid|
+            _, status = Process.waitpid2(pid)
+            errors << pid_to_name_map[pid] if status.exitstatus != 0
+          end
 
           # I'm not sure why, but we have to do this
           ::ActiveRecord::Base.clear_all_connections!
+
           # check for an exception; we only re-raise the first one
-          # (all the sub-processes shared the same pipe, so we only
-          # have to check the one)
-          begin
-            exception = Marshal.load exception_pipe.first
-            raise exception
-          rescue EOFError
-            # No exceptions
-          ensure
-            exception_pipe.first.close
+          exception_pipes.each do |exception_pipe|
+            begin
+              exception = Marshal.load exception_pipe.first
+              raise exception
+            rescue EOFError
+              # No exceptions
+            ensure
+              exception_pipe.first.close
+            end
+          end
+
+          unless errors.empty?
+            raise ParallelShardExecError.new("The following subprocesses did not exit cleanly: #{errors.sort.join(", ")}")
           end
           return
         end

data/lib/switchman.rb

@@ -1,5 +1,5 @@
 require "shackles"
-require "open4"
+require "switchman/open4"
 require "switchman/engine"
 
 module Switchman

data/lib/switchman/errors.rb

@@ -1,3 +1,4 @@
 module Switchman
   class NonExistentShardError < RuntimeError; end
+  class ParallelShardExecError < RuntimeError; end
 end

data/lib/switchman/open4.rb

@@ -0,0 +1,78 @@
+require 'open4'
+
+# This fixes a bug with exception handling,
+# see https://github.com/ahoward/open4/pull/30
+module Open4
+  def self.do_popen(b = nil, exception_propagation_at = nil, closefds=false, &cmd)
+    pw, pr, pe, ps = IO.pipe, IO.pipe, IO.pipe, IO.pipe
+
+    verbose = $VERBOSE
+    begin
+      $VERBOSE = nil
+
+      cid = fork {
+        if closefds
+          exlist = [0, 1, 2] | [pw,pr,pe,ps].map{|p| [p.first.fileno, p.last.fileno] }.flatten
+          ObjectSpace.each_object(IO){|io|
+            io.close if (not io.closed?) and (not exlist.include? io.fileno) rescue nil
+          }
+        end
+
+        pw.last.close
+        STDIN.reopen pw.first
+        pw.first.close
+
+        pr.first.close
+        STDOUT.reopen pr.last
+        pr.last.close
+
+        pe.first.close
+        STDERR.reopen pe.last
+        pe.last.close
+
+        STDOUT.sync = STDERR.sync = true
+
+        begin
+          cmd.call(ps)
+        rescue Exception => e
+          begin
+            Marshal.dump(e, ps.last)
+            ps.last.flush
+          rescue Errno::EPIPE
+            raise e
+          end
+        ensure
+          ps.last.close unless ps.last.closed?
+        end
+
+        exit!
+      }
+    ensure
+      $VERBOSE = verbose
+    end
+
+    [ pw.first, pr.last, pe.last, ps.last ].each { |fd| fd.close }
+
+    Open4.propagate_exception cid, ps.first if exception_propagation_at == :init
+
+    pw.last.sync = true
+
+    pi = [ pw.last, pr.first, pe.first ]
+
+    begin
+      return [cid, *pi] unless b
+
+      begin
+        b.call(cid, *pi)
+      ensure
+        pi.each { |fd| fd.close unless fd.closed? }
+      end
+
+      Open4.propagate_exception cid, ps.first if exception_propagation_at == :block
+
+      Process.waitpid2(cid).last
+    ensure
+      ps.first.close unless ps.first.closed?
+    end
+  end
+end

data/lib/switchman/version.rb

@@ -1,3 +1,3 @@
 module Switchman
-  VERSION = "1.5.15"
+  VERSION = "1.5.16"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: switchman
 version: !ruby/object:Gem::Version
-  version: 1.5.15
+  version: 1.5.16
 platform: ruby
 authors:
 - Cody Cutrer
@@ -10,7 +10,7 @@ authors:
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2016-08-17 00:00:00.000000000 Z
+date: 2016-08-22 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: railties
@@ -195,6 +195,7 @@ files:
 - lib/switchman/engine.rb
 - lib/switchman/environment.rb
 - lib/switchman/errors.rb
+- lib/switchman/open4.rb
 - lib/switchman/r_spec_helper.rb
 - lib/switchman/rails.rb
 - lib/switchman/schema_cache.rb