zookeeper 1.0.2-java → 1.0.3-java

data/CHANGELOG

@@ -1,3 +1,9 @@
+v1.0.3 Linux: Fix hang on reopen after fork
+
+  * Linux users relying on fork() should upgrade to 1.0.3, as it
+    fixes a bug in the reopen code that would cause a hang in the child.
+    If you're running on linux, you *really* should upgrade
+
 v1.0.0 Single Zookeeper namespace
 
   * The top level Zookeeper class is now a module, the former Zookeeper

data/Rakefile

@@ -72,7 +72,7 @@ task :clobber do
 end
 
 # cargo culted from http://blog.flavorjon.es/2009/06/easily-valgrind-gdb-your-ruby-c.html
-VALGRIND_BASIC_OPTS = '--num-callers=50 --error-limit=no --partial-loads-ok=yes --undef-value-errors=no'
+VALGRIND_BASIC_OPTS = '--num-callers=50 --error-limit=no --partial-loads-ok=yes --undef-value-errors=no --trace-children=yes'
 
 task 'valgrind' do
   cd 'ext' do

data/ext/zookeeper_base.rb

@@ -72,6 +72,22 @@ class ZookeeperBase
   threadsafe_inquisitor :connected?, :connecting?, :associating?, :running?
 
   attr_reader :event_queue
+  
+  # this method may be called in either the fork case, or from the constructor
+  # to set up this state initially (so all of this is in one location). we rely
+  # on the forked? method to determine which it is
+  def reopen_after_fork!
+    logger.debug { "#{self.class}##{__method__} fork detected!" } if forked?
+
+    @mutex = Monitor.new
+    @dispatch_shutdown_cond = @mutex.new_cond
+    @event_queue = QueueWithPipe.new
+
+    @dispatcher = nil if @dispatcher and not @dispatcher.alive?
+
+    update_pid!  # from Forked
+  end
+  private :reopen_after_fork!
  
   def reopen(timeout = 10, watcher=nil)
     if watcher and (watcher != @default_watcher)
@@ -81,13 +97,16 @@ class ZookeeperBase
     reopen_after_fork! if forked?
 
     @mutex.synchronize do
+      if @czk 
+        @czk.close
+        @czk = nil
+      end
+
+      @czk = CZookeeper.new(@host, @event_queue)
+
       # flushes all outstanding watcher reqs.
       @watcher_reqs.clear
       set_default_global_watcher
-
-      orig_czk, @czk = @czk, CZookeeper.new(@host, @event_queue)
-
-      orig_czk.close if orig_czk
       
       @czk.wait_until_connected(timeout)
     end
@@ -100,15 +119,12 @@ class ZookeeperBase
     @watcher_reqs = {}
     @completion_reqs = {}
 
-    update_pid!  # from Forked
-
     @current_req_id = 0
 
-    # set up state that also needs to be re-setup after a fork()
+    @dispatcher = @czk = nil
+
     reopen_after_fork!
     
-    @czk = nil
-    
     # approximate the java behavior of raising java.lang.IllegalArgumentException if the host
     # argument ends with '/'
     raise ArgumentError, "Host argument #{host.inspect} may not end with /" if host.end_with?('/')
@@ -202,24 +218,6 @@ class ZookeeperBase
   end
  
 protected
-  # this method may be called in either the fork case, or from the constructor
-  # to set up this state initially (so all of this is in one location). we rely
-  # on the forked? method to determine which it is
-  def reopen_after_fork!
-    logger.debug { "#{self.class}##{__method__}" }
-    @mutex = Monitor.new
-    @dispatch_shutdown_cond = @mutex.new_cond
-    @event_queue = QueueWithPipe.new
-
-    if @dispatcher and not @dispatcher.alive?
-      logger.debug { "#{self.class}##{__method__} re-starting dispatch thread" }
-      @dispatcher = nil
-      setup_dispatch_thread!
-    end
-
-    update_pid!
-  end
-
   # this is a hack: to provide consistency between the C and Java drivers when
   # using a chrooted connection, we wrap the callback in a block that will
   # strip the chroot path from the returned path (important in an async create

data/ext/zookeeper_c.c

@@ -55,6 +55,8 @@ typedef enum {
 static int destroy_zkrb_instance(struct zkrb_instance_data* ptr) {
   int rv = ZOK;
 
+  zkrb_debug("destroy_zkrb_instance, zk_local_ctx: %p, zh: %p, queue: %p", ptr, ptr->zh, ptr->queue);
+
   if (ptr->zh) {
     const void *ctx = zoo_get_context(ptr->zh);
     /* Note that after zookeeper_close() returns, ZK handle is invalid */
@@ -163,6 +165,7 @@ static VALUE method_zkrb_init(int argc, VALUE* argv, VALUE self) {
   VALUE data;
   struct zkrb_instance_data *zk_local_ctx;
   data = Data_Make_Struct(CZookeeper, struct zkrb_instance_data, 0, free_zkrb_instance_data, zk_local_ctx);
+
   zk_local_ctx->queue = zkrb_queue_alloc();
 
   if (zk_local_ctx->queue == NULL)
@@ -184,6 +187,9 @@ static VALUE method_zkrb_init(int argc, VALUE* argv, VALUE self) {
           ctx,
           0);
 
+  zkrb_debug("method_zkrb_init, zk_local_ctx: %p, zh: %p, queue: %p, calling_ctx: %p",
+      zk_local_ctx, zk_local_ctx->zh, zk_local_ctx->queue, ctx);
+
 #warning [wickman] TODO handle this properly on the Ruby side rather than C side
   if (!zk_local_ctx->zh) {
     rb_raise(rb_eRuntimeError, "error connecting to zookeeper: %d", errno);

data/lib/zookeeper/version.rb

@@ -1,4 +1,4 @@
 module Zookeeper
-  VERSION = '1.0.2'
+  VERSION = '1.0.3'
   DRIVER_VERSION = '3.3.5'
 end

data/spec/forked_connection_spec.rb

@@ -17,8 +17,8 @@ unless defined?(::JRUBY_VERSION)
     before do
       if defined?(::Rubinius)
         pending("this test is currently broken in rbx")
-      elsif ENV['TRAVIS']
-        pending("this test is currently hanging in travis")
+#       elsif ENV['TRAVIS']
+#         pending("this test is currently hanging in travis")
       else
         @zk = Zookeeper.new(connection_string)
         rm_rf(@zk, path)
@@ -54,6 +54,8 @@ unless defined?(::JRUBY_VERSION)
 
       @zk.stat(:path => "#{pids_root}/child", :watcher => cb)
 
+      logger.debug { "------------------->   FORK   <---------------------------" }
+
       @pid = fork do
         logger.debug { "reopening connection in child: #{$$}" }
         @zk.reopen
@@ -69,7 +71,6 @@ unless defined?(::JRUBY_VERSION)
       logger.debug { "waiting on child #{@pid}" }
 
       @latch.await until @event 
-      p @event
 
       _, status = Process.wait2(@pid)
 

metadata

@@ -1,13 +1,13 @@
 --- !ruby/object:Gem::Specification 
 name: zookeeper
 version: !ruby/object:Gem::Version 
-  hash: 19
+  hash: 17
   prerelease: 
   segments: 
   - 1
   - 0
-  - 2
-  version: 1.0.2
+  - 3
+  version: 1.0.3
 platform: java
 authors: 
 - Phillip Pearson