Merge branch '2.3' into 2.4

Documentation/Monitors/MariaDB-Monitor.md

@@ -36,7 +36,7 @@ Table of Contents
         * [verify_master_failure and master_failure_timeout](#verify_master_failure-and-master_failure_timeout)
         * [servers_no_promotion](#servers_no_promotion)
         * [promotion_sql_file and demotion_sql_file](#promotion_sql_file-and-demotion_sql_file)
-        * [handle_server_events](#handle_server_events)
+        * [handle_events](#handle_events)
   * [Troubleshooting](#troubleshooting)
      * [Failover/switchover fails](#failoverswitchover-fails)
      * [Slave detection shows external masters](#slave-detection-shows-external-masters)
@@ -331,7 +331,7 @@ being demoted or promoted by cluster manipulation commands. See the sections on
 `promotion_sql_file` and `demotion_sql_file` for more information.
 
 The monitor can manipulate scheduled server events when promoting or demoting a
-server. See the section on `handle_server_events` for more information.
+server. See the section on `handle_events` for more information.
 
 All cluster operations can be activated manually through MaxAdmin/MaxCtrl. See
 section [Manual activation](#manual-activation) for more details.
@@ -818,7 +818,7 @@ slave threads are stopped, breaking replication.
 promotion_sql_file=/home/root/scripts/promotion.sql
 demotion_sql_file=/home/root/scripts/demotion.sql
 ```
-#### `handle_server_events`
+#### `handle_events`
 
 This setting is on by default. If enabled, the monitor continuously queries the
 servers for enabled scheduled events and uses this information when performing

include/maxscale/backend.hh

@@ -357,6 +357,16 @@ public:
      */
     void set_close_reason(const std::string& reason);
 
+    /**
+     * Get latest close reason
+     *
+     * @return A human-readable reason why the connection was closed
+     */
+    const std::string& close_reason() const
+    {
+        return m_close_reason;
+    }
+
 private:
     /**
      * Internal state of the backend

server/core/backend.cc

@@ -197,6 +197,7 @@ bool Backend::connect(MXS_SESSION* session, SessionCommandList* sescmd)
         m_closed_at = 0;
         m_opened_at = time(NULL);
         m_state = IN_USE;
+        m_close_reason.clear();
         mxb::atomic::add(&m_backend->connections, 1, mxb::atomic::RELAXED);
         rval = true;
         m_history_size = 0;

server/modules/routing/readwritesplit/rwsplitsession.cc

@@ -678,6 +678,20 @@ void RWSplitSession::clientReply(GWBUF* writebuf, DCB* backend_dcb)
 
     if (error.is_unexpected_error())
     {
+        // The connection was killed, we can safely ignore it. When the TCP connection is
+        // closed, the router's error handling will sort it out.
+        if (error.code() == ER_CONNECTION_KILLED)
+        {
+            backend->set_close_reason("Connection was killed");
+        }
+        else
+        {
+            mxb_assert(error.code() == ER_SERVER_SHUTDOWN
+                       || error.code() == ER_NORMAL_SHUTDOWN
+                       || error.code() == ER_SHUTDOWN_COMPLETE);
+            backend->set_close_reason(std::string("Server '") + backend->name() + "' is shutting down");
+        }
+
         // The server sent an error that we didn't expect: treat it as if the connection was closed. The
         // client shouldn't see this error as we can replace the closed connection.
 
@@ -1125,9 +1139,11 @@ void RWSplitSession::handleError(GWBUF* errmsgbuf,
                 {
                     int64_t idle = mxs_clock() - backend->dcb()->last_read;
                     MXS_ERROR("Lost connection to the master server '%s', closing session.%s "
-                              "Connection has been idle for %.1f seconds. Error caused by: %s",
+                              "Connection has been idle for %.1f seconds. Error caused by: %s. "
+                              "Last close reason: %s",
                               backend->name(), errmsg.c_str(), (float)idle / 10.f,
-                              extract_error(errmsgbuf).c_str());
+                              extract_error(errmsgbuf).c_str(),
+                              backend->close_reason().empty() ? "<none>" : backend->close_reason().c_str());
                 }
 
                 // Decrement the expected response count only if we know we can continue the sesssion.