- #PXC-347: Non-PRIM during membership change

asio/asio/detail/config.hpp

@@ -654,7 +654,19 @@
 # if !defined(ASIO_HAS_TIMERFD)
 #  if defined(ASIO_HAS_EPOLL)
 #   if (__GLIBC__ > 2) || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 8)
-#    define ASIO_HAS_TIMERFD 1
+//
+// Currently timerfd is a source of unidentified problems
+// that break GCS unit-tests on some systems during testing
+// on the Jenkins. One of these problems has been found and
+// corrected: https://github.com/chriskohlhoff/asio/pull/145,
+// but there is another problem, because of which some of
+// compilations (on the systems with timerfd feature enabled)
+// occasionally lead to failures in the GCS unit-tests, that
+// may happen even after merging of the proposed correction
+// (i.e. problem, which is shown above is not unique one,
+// nevertheless after applying the correction probability
+// of failures in the GCS unit-test sharply decreased):
+//#    define ASIO_HAS_TIMERFD 1
 #   endif // (__GLIBC__ > 2) || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 8)
 #  endif // defined(ASIO_HAS_EPOLL)
 # endif // !defined(ASIO_HAS_TIMERFD)

asio/asio/detail/impl/epoll_reactor.ipp

@@ -429,10 +429,19 @@ void epoll_reactor::run(bool block, op_queue<operation>& ops)
       {
         // If timeout was expired, we should read the expiration counter:
         uint64_t count;
-        if (read(timer_fd_, &count, sizeof(count)))
+        ssize_t len = sizeof(count);
+        char * buf = reinterpret_cast<char *>(&count);
+        for(;;)
         {
-          // Just to ignore return value of the read() function
-          // without compiler warning...
+          ssize_t ret = read(timer_fd_, buf, len);
+          if(ret == len) break;
+          if (ret < 0)
+          {
+            if (EINTR != errno) break;
+            continue;
+          }
+          buf += ret;
+          len -= ret;
         }
       }
       check_timers = true;

gcomm/src/evs_proto.cpp

@@ -883,6 +883,15 @@ gu::datetime::Date gcomm::evs::Proto::handle_timers()
     {
         Timer t(TimerList::value(timers_.begin()));
         timers_.erase(timers_.begin());
+        if (state() == S_LEAVING)
+        {
+            // Due to the implementation details of the ASIO,
+            // timers can be triggered during the normal shutdown
+            // sequence, which leads to problems. We may ignore
+            // timers, if we are in the process of completing
+            // the work:
+            continue;
+        }
         switch (t)
         {
         case T_INACTIVITY:

gcomm/src/gmcast.cpp

@@ -998,8 +998,9 @@ void gcomm::GMCast::reconnect()
         {
             if (ae.retry_cnt() > ae.max_retries())
             {
-                log_info << " cleaning up " << remote_uuid << " ("
-                         << remote_addr << ")";
+                log_info << self_string()
+                         << " cleaning up " << remote_uuid
+                         << " (" << remote_addr << ")";
                 remote_addrs_.erase(i);
                 continue;//no reference to remote_addr or remote_uuid after this
             }