Work-around for race condition in the IO layer.

Added a 100ms sleep in UpdateClient between calling
Connection::close() and destroying the Connection object.
This appears to work around the race condition described in
 https://bugzilla.redhat.com/show_bug.cgi?id=568831.
which was causing brokers to seg fault sporadically at the end of
giving an update.


git-svn-id: https://svn.apache.org/repos/asf/qpid/trunk@923414 13f79535-47bb-0310-9956-ffa450edef68

2 files changed, 8 insertions, 2 deletions

diff --git a/qpid/cpp/src/qpid/cluster/UpdateClient.cpp b/qpid/cpp/src/qpid/cluster/UpdateClient.cpp
index 17d856b79c..ab992bf8cf 100644
--- a/qpid/cpp/src/qpid/cluster/UpdateClient.cpp
+++ b/qpid/cpp/src/qpid/cluster/UpdateClient.cpp
@@ -158,6 +158,12 @@ void UpdateClient::update() {
     connection.close();
     QPID_LOG(debug,  updaterId << " update completed to " << updateeId
              << " at " << updateeUrl << ": " << membership);
+    // FIXME aconway 2010-03-15: This sleep avoids the race condition
+    // described in // https://bugzilla.redhat.com/show_bug.cgi?id=568831.
+    // It allows the connection to fully close before destroying the
+    // Connection object. Remove when the bug is fixed.
+    //
+    sys::usleep(10*1000);       // 100ms
 }
 
 namespace {
diff --git a/qpid/cpp/src/tests/ClusterFixture.cpp b/qpid/cpp/src/tests/ClusterFixture.cpp
index fd90ed170e..b7e8e88abf 100644
--- a/qpid/cpp/src/tests/ClusterFixture.cpp
+++ b/qpid/cpp/src/tests/ClusterFixture.cpp
@@ -130,11 +130,11 @@ void ClusterFixture::kill(size_t n, int sig) {
         forkedBrokers[n]->kill(sig);
 }
 
-/** Kill a broker and suppressing errors from closing connection c. */
+/** Kill a broker and suppress errors from closing connection c. */
 void ClusterFixture::killWithSilencer(size_t n, client::Connection& c, int sig) {
     ScopedSuppressLogging sl;
-    kill(n,sig);
     try { c.close(); } catch(...) {}
+    kill(n,sig);
 }
 
 /**