WL#12999 Make the client get a better error message on wait_timeout

When the server terminates a client connection because of timeout, it
tries to send an error message to the client. This message is later
caught by the client on a best effort as the client discover that
the write socket is terminated.
Patch contributed by Mattias Jonsson @ booking.

RB: 25049

Author: Kristofer Älvring

client/mysql.cc

@@ -3042,11 +3042,19 @@ static int reconnect(void) {
   return 0;
 }
 
-static void get_current_db() {
+/**
+  Checks the current DB and updates the global variable current_db
+  If the command fails hen he current_db is set to nullptr.
+
+  @return Error state
+    @retval true An error occurred
+    @retval false Success; current_db is updated
+*/
+static bool get_current_db() {
   MYSQL_RES *res;
 
   /* If one_database is set, current_db is not supposed to change. */
-  if (one_database) return;
+  if (one_database) return false;
 
   my_free(current_db);
   current_db = nullptr;
@@ -3057,7 +3065,11 @@ static void get_current_db() {
     if (row && row[0])
       current_db = my_strdup(PSI_NOT_INSTRUMENTED, row[0], MYF(MY_WME));
     mysql_free_result(res);
+  } else {
+    /* We failed to issue the command and we likely lost connection */
+    return true;
   }
+  return false;
 }
 
 /***************************************************************************
@@ -3073,8 +3085,10 @@ static int mysql_real_query_for_lazy(const char *buf, size_t length,
     if (set_params && global_attrs->set_params(&mysql)) break;
     if (!mysql_real_query(&mysql, buf, (ulong)length)) break;
     error = put_error(&mysql);
-    if (mysql_errno(&mysql) != CR_SERVER_GONE_ERROR || retry > 1 ||
-        !opt_reconnect)
+    if ((mysql_errno(&mysql) != CR_SERVER_GONE_ERROR &&
+         mysql_errno(&mysql) != CR_SERVER_LOST &&
+         mysql.net.error != NET_ERROR_SOCKET_UNUSABLE) ||
+        retry > 1 || !opt_reconnect)
       break;
     if (reconnect()) break;
   }
@@ -4270,8 +4284,9 @@ static int com_use(String *buffer MY_ATTRIBUTE((unused)), char *line) {
     We need to recheck the current database, because it may change
     under our feet, for example if DROP DATABASE or RENAME DATABASE
     (latter one not yet available by the time the comment was written)
+    If this command fails we assume we lost connection.
   */
-  get_current_db();
+  if (get_current_db()) connected = false;
 
   if (!current_db || cmp_database(charset_info, current_db, tmp)) {
     if (one_database) {

include/mysql_com.h

@@ -876,6 +876,12 @@ struct Vio;
 /// Default width for blob in bytes @todo - align this with sizes from field.h
 #define MAX_BLOB_WIDTH 16777216
 
+#define NET_ERROR_UNSET 0               /**< No error has occurred yet */
+#define NET_ERROR_SOCKET_RECOVERABLE 1  /**< Socket still usable */
+#define NET_ERROR_SOCKET_UNUSABLE 2     /**< Do not use the socket */
+#define NET_ERROR_SOCKET_NOT_READABLE 3 /**< Try write and close socket */
+#define NET_ERROR_SOCKET_NOT_WRITABLE 4 /**< Try read and close socket */
+
 typedef struct NET {
   MYSQL_VIO vio;
   unsigned char *buff, *buff_end, *write_pos, *read_pos;

mysql-test/include/wait_until_connected_again.inc

@@ -40,7 +40,7 @@ while ($mysql_errno)
   # Strangely enough, the server might return "Too many connections"
   # while being shutdown, thus 1040 is an "allowed" error
   # See BUG#36228
-  --error 0,1040,1053,2002,2003,2006,2013,1045,ER_SECURE_TRANSPORT_REQUIRED,2016,2017
+  --error 0,1040,1053,2002,2003,2006,2013,1045,ER_SECURE_TRANSPORT_REQUIRED,2016,2017,ER_CLIENT_INTERACTION_TIMEOUT
   show session status;
   if ($mysql_errno == 1045){
     --let mysql_errno=0

mysql-test/include/wait_until_disconnected.inc

@@ -11,7 +11,7 @@ if ($disconnect_timeout)
 let $mysql_errno= 0;
 while (!$mysql_errno)
 {
-  --error 0,1040,1053,2002,2003,2006,2013,2016,2017
+  --error 0,1040,1053,2002,2003,2006,2013,2016,2017,ER_CLIENT_INTERACTION_TIMEOUT
   show session status;
 
   dec $counter;

mysql-test/r/mysqltest.result

@@ -1315,7 +1315,7 @@ SELECT 1 AS res;
 res
 1
 SELECT 1 AS res;
-ERROR HY000: MySQL server has gone away
+Got one of the listed errors
 #
 # Bug#23743035: ADD A COPY_FILES_WILDCARD COMMAND
 #

mysql-test/r/ssl.result

@@ -2275,6 +2275,6 @@ Ssl_cipher	SSL_CIPHER
 SET @@SESSION.wait_timeout = 2;
 # Wait for ssl_con to be disconnected.
 # Check that ssl_con has been disconnected.
-# CR_SERVER_LOST, CR_SERVER_GONE_ERROR
+# Different behavior depending on how the plattform implements the 
 SELECT 1;
 Got one of the listed errors

mysql-test/r/wait_timeout.result

@@ -42,7 +42,6 @@ disconnection con1;
 SET @@SESSION.wait_timeout = 2;
 # Wait for con1 to be disconnected.
 # Check that con1 has been disconnected.
-# CR_SERVER_LOST, CR_SERVER_GONE_ERROR
 SELECT 1;
 Got one of the listed errors
 #
@@ -52,8 +51,27 @@ Got one of the listed errors
 SET @@SESSION.wait_timeout = 2;
 # Wait for con1 to be disconnected.
 # Check that con1 has been disconnected.
-# CR_SERVER_LOST, CR_SERVER_GONE_ERROR
+# Client interaction timeout
 SELECT 1;
 Got one of the listed errors
+SELECT "Check that we don't reconnect with reconnection disabled.";
+Got one of the listed errors
+#
+# Test UNIX domain sockets timeout with reconnect.
+#
+# Open con2 and set a timeout.
+--enable_reconnect;
+SET @is_old_connection = 1;
+SELECT @is_old_connection;
+@is_old_connection
+1
+SET @@SESSION.wait_timeout = 2;
+# Wait for con2 to be disconnected.
+# Check that con2 has been reconnected.
+SELECT "Unix domain socket will hit wait_timeout with reconnect";
+Got one of the listed errors
+SELECT @is_old_connection;
+@is_old_connection
+NULL
 # must find the pattern
 Pattern "The wait_timeout period was exceeded, the idle time since last command was too long." found

mysql-test/t/mysqltest.test

@@ -3287,7 +3287,7 @@ SELECT 1 AS res;
 disconnect con1;
 --source include/wait_until_disconnected.inc
 
---error CR_SERVER_GONE_ERROR
+--error CR_SERVER_GONE_ERROR,CR_SERVER_LOST
 SELECT 1 AS res;
 
 connection default;

mysql-test/t/ssl.test

@@ -48,8 +48,8 @@ let $wait_condition=
 
 --echo # Check that ssl_con has been disconnected.
 connection ssl_con;
---echo # CR_SERVER_LOST, CR_SERVER_GONE_ERROR
---error 2006,2013
+--echo # Different behavior depending on how the plattform implements the 
+--error ER_CLIENT_INTERACTION_TIMEOUT,CR_SERVER_LOST
 SELECT 1;
 
 connection default;

mysql-test/t/wait_timeout.test

@@ -66,7 +66,7 @@ connection default;
 # When the connection is closed in this way, the error code should
 # be consistent see Bug#2845 for an explanation
 # depending on platform/client, either errno 2006 or 2013 can occur below
---error 2006,2013
+--error ER_CLIENT_INTERACTION_TIMEOUT,CR_SERVER_LOST
 SELECT 2;
 --echo --enable_reconnect;
 --enable_reconnect
@@ -123,7 +123,7 @@ connection con1;
 # When the connection is closed in this way, the error code should
 # be consistent see Bug#2845 for an explanation
 # depending on platform/client, either errno 2006 or 2013 can occur below
---error 2006,2013
+--error ER_CLIENT_INTERACTION_TIMEOUT,CR_SERVER_LOST
 SELECT 2;
 --echo --enable_reconnect;
 --enable_reconnect
@@ -148,6 +148,7 @@ connect (default,localhost,root,,test,,);
 --echo # Open con1 and set a timeout.
 connect(con1,localhost,root,,);
 
+--disable_reconnect
 LET $ID= `SELECT connection_id()`;
 SET @@SESSION.wait_timeout = 2;
 
@@ -160,8 +161,7 @@ let $wait_condition=
 
 --echo # Check that con1 has been disconnected.
 connection con1;
---echo # CR_SERVER_LOST, CR_SERVER_GONE_ERROR
---error 2006,2013
+--error ER_CLIENT_INTERACTION_TIMEOUT,CR_SERVER_LOST
 SELECT 1;
 
 disconnect con1;
@@ -172,7 +172,7 @@ connection default;
 --echo #
 
 --echo # Open con1 and set a timeout.
-connect(con1,127.0.0.1,root,,);
+connect(con1,127.0.0.1,root,,,,,TCP,,);
 
 LET $ID= `SELECT connection_id()`;
 SET @@SESSION.wait_timeout = 2;
@@ -186,16 +186,47 @@ let $wait_condition=
 
 --echo # Check that con1 has been disconnected.
 connection con1;
---echo # CR_SERVER_LOST, CR_SERVER_GONE_ERROR
---error 2006,2013
+--disable_reconnect
+--echo # Client interaction timeout
+--error ER_CLIENT_INTERACTION_TIMEOUT,CR_SERVER_LOST
 SELECT 1;
 
-disconnect con1;
+--error ER_CLIENT_INTERACTION_TIMEOUT,CR_SERVER_LOST
+SELECT "Check that we don't reconnect with reconnection disabled.";
+
+--echo #
+--echo # Test UNIX domain sockets timeout with reconnect.
+--echo #
+
+--echo # Open con2 and set a timeout.
+connect(con2,localhost,root,,);
+
+--echo --enable_reconnect;
+--enable_reconnect
+SET @is_old_connection = 1;
+SELECT @is_old_connection;
+
+LET $ID= `SELECT connection_id()`;
+SET @@SESSION.wait_timeout = 2;
+
+--echo # Wait for con2 to be disconnected.
 connection default;
+let $wait_condition=
+  SELECT COUNT(*) = 0 FROM INFORMATION_SCHEMA.PROCESSLIST
+  WHERE ID = $ID;
+--source include/wait_condition.inc
 
+--echo # Check that con2 has been reconnected.
+connection con2;
+--error ER_CLIENT_INTERACTION_TIMEOUT,CR_SERVER_LOST
+SELECT "Unix domain socket will hit wait_timeout with reconnect";
+SELECT @is_old_connection;
+connection default;
+--disable_reconnect
+disconnect con2;
+disconnect con1;
 # Wait till all disconnects are completed
 --source include/wait_until_count_sessions.inc
-
 #
 # Bug #28940167 WAIT_TIMEOUT ERROR NOT CLEAR AND NOT SENT TO CLIENT BEFORE CLOSING CONNECTION
 #

share/messages_to_clients.txt

@@ -9502,6 +9502,9 @@ ER_TABLE_MUST_HAVE_A_VISIBLE_COLUMN
 ER_WARN_IMPORT_COMPRESSION_FAIL
   eng "Compression failed while doing import with the following error : %s"
 
+ER_CLIENT_INTERACTION_TIMEOUT
+    eng "The client was disconnected by the server because of inactivity. See wait_timeout and interactive_timeout for configuring this behavior."
+
 #
 #  End of 8.0 error messages (server-to-client).
 #  Do NOT add messages intended for the error log above!