MDEV-14412 Support TCP keepalive options

Based on pull request by Oleg Obleukhov
https://github.com/MariaDB/server/pull/400

include/violite.h

@@ -51,6 +51,14 @@ enum enum_vio_io_event
   VIO_IO_EVENT_CONNECT
 };
 
+struct vio_keepalive_opts
+{
+  int interval;
+  int idle;
+  int probes;
+};
+
+
 #define VIO_LOCALHOST 1U                        /* a localhost connection */
 #define VIO_BUFFERED_READ 2U                    /* use buffered read */
 #define VIO_READ_BUFFER_SIZE 16384U             /* size of read buffer */
@@ -84,6 +92,7 @@ my_bool	vio_is_blocking(Vio *vio);
 int	vio_fastsend(Vio *vio);
 /* setsockopt SO_KEEPALIVE at SOL_SOCKET level, when possible */
 int	vio_keepalive(Vio *vio, my_bool	onoff);
+int	vio_set_keepalive_options(Vio * vio, const struct vio_keepalive_opts *opts);
 /* Whenever we should retry the last read/write operation. */
 my_bool	vio_should_retry(Vio *vio);
 /* Check that operation was timed out */
@@ -214,7 +223,6 @@ enum SSL_type
   SSL_TYPE_SPECIFIED
 };
 
-
 /* HFTODO - hide this if we don't want client in embedded server */
 /* This structure is for every connection on both sides */
 struct st_vio

mysql-test/r/mysqld--help,win.rdiff

@@ -1,6 +1,6 @@
 --- a/mysql-test/r/mysqld--help.result
 +++ b/mysql-test/r/mysqld--help.result
-@@ -334,7 +334,6 @@
+@@ -381,7 +381,6 @@
   The number of segments in a key cache
   -L, --language=name Client error messages in given language. May be given as
   a full path. Deprecated. Use --lc-messages-dir instead.
@@ -8,7 +8,7 @@
   --lc-messages=name  Set the language used for the error messages.
   -L, --lc-messages-dir=name 
   Directory where error messages are
-@@ -543,6 +542,7 @@
+@@ -602,6 +601,7 @@
   Use MySQL-5.6 (instead of MariaDB-5.3) format for TIME,
   DATETIME, TIMESTAMP columns.
   (Defaults to on; use --skip-mysql56-temporal-format to disable.)
@@ -16,7 +16,7 @@
   --net-buffer-length=# 
   Buffer length for TCP/IP and socket communication
   --net-read-timeout=# 
-@@ -957,6 +957,9 @@
+@@ -1020,6 +1020,9 @@
   characteristics (isolation level, read only/read
   write,snapshot - but not any work done / data modified
   within the transaction).
@@ -26,7 +26,7 @@
   --show-slave-auth-info 
   Show user and password in SHOW SLAVE HOSTS on this
   master.
-@@ -1069,6 +1072,10 @@
+@@ -1132,6 +1135,10 @@
   Log slow queries to given log file. Defaults logging to
   'hostname'-slow.log. Must be enabled to activate other
   slow log options
@@ -37,15 +37,15 @@
   --socket=name       Socket file to use for connection
   --sort-buffer-size=# 
   Each thread that needs to do a sort allocates a buffer of
-@@ -1087,6 +1094,7 @@
+@@ -1151,6 +1158,7 @@
-  NO_ENGINE_SUBSTITUTION, PAD_CHAR_TO_FULL_LENGTH
+  EMPTY_STRING_IS_NULL
   --stack-trace       Print a symbolic stack trace on failure
   (Defaults to on; use --skip-stack-trace to disable.)
 + --standalone        Dummy option to start as a standalone program (NT).
   --standard-compliant-cte 
   Allow only CTEs compliant to SQL standard
   (Defaults to on; use --skip-standard-compliant-cte to disable.)
-@@ -1134,6 +1142,11 @@
+@@ -1214,6 +1222,11 @@
   --thread-pool-max-threads=# 
   Maximum allowed number of worker threads in the thread
   pool
@@ -57,7 +57,7 @@
   --thread-pool-oversubscribe=# 
   How many additional active worker threads in a group are
   allowed.
-@@ -1172,8 +1185,8 @@
+@@ -1252,8 +1265,8 @@
   automatically convert it to an on-disk MyISAM or Aria
   table.
   -t, --tmpdir=name   Path for temporary files. Several paths may be specified,
@@ -68,7 +68,7 @@
   --transaction-alloc-block-size=# 
   Allocation block size for transactions to be stored in
   binary log
-@@ -1298,7 +1311,6 @@
+@@ -1387,7 +1400,6 @@
  key-cache-division-limit 100
  key-cache-file-hash-size 512
  key-cache-segments 0
@@ -76,7 +76,7 @@
  lc-messages en_US
  lc-messages-dir MYSQL_SHAREDIR/
  lc-time-names en_US
-@@ -1368,6 +1380,7 @@
+@@ -1459,6 +1471,7 @@
  myisam-stats-method NULLS_UNEQUAL
  myisam-use-mmap FALSE
  mysql56-temporal-format TRUE
@@ -84,16 +84,16 @@
  net-buffer-length 16384
  net-read-timeout 30
  net-retry-count 10
-@@ -1469,6 +1482,8 @@
+@@ -1561,6 +1574,8 @@
  session-track-state-change FALSE
- session-track-system-variables 
+ session-track-system-variables autocommit,character_set_client,character_set_connection,character_set_results,time_zone
  session-track-transaction-info OFF
 +shared-memory FALSE
 +shared-memory-base-name MYSQL
  show-slave-auth-info FALSE
  silent-startup FALSE
  skip-grant-tables TRUE
-@@ -1493,6 +1508,7 @@
+@@ -1585,6 +1600,7 @@
  slave-type-conversions 
  slow-launch-time 2
  slow-query-log FALSE
@@ -101,7 +101,7 @@
  sort-buffer-size 2097152
  sql-mode STRICT_TRANS_TABLES,ERROR_FOR_DIVISION_BY_ZERO,NO_AUTO_CREATE_USER,NO_ENGINE_SUBSTITUTION
  stack-trace TRUE
-@@ -1506,14 +1522,16 @@
+@@ -1598,9 +1614,9 @@
  sync-relay-log 10000
  sync-relay-log-info 10000
  sysdate-is-now FALSE
@@ -112,6 +112,8 @@
 +table-open-cache 2000
  table-open-cache-instances 8
  tc-heuristic-recover OFF
+ tcp-keepalive-interval 0
+@@ -1609,6 +1625,8 @@
  thread-cache-size 151
  thread-pool-idle-timeout 60
  thread-pool-max-threads 65536

mysql-test/r/mysqld--help.result

@@ -1195,6 +1195,21 @@ The following options may be given as the first argument:
  --tc-heuristic-recover=name 
  Decision to use in heuristic recover process. One of: OFF,
  COMMIT, ROLLBACK
+ --tcp-keepalive-interval=# 
+ The interval, in seconds, between when successive
+ keep-alive packets are sent if no acknowledgement is
+ received.If set to 0, system dependent default is used.
+ (Automatically configured unless set explicitly)
+ --tcp-keepalive-probes=# 
+ The number of unacknowledged probes to send before
+ considering the connection dead and notifying the
+ application layer.If set to 0, system dependent default
+ is used. (Automatically configured unless set explicitly)
+ --tcp-keepalive-time=# 
+ Timeout, in milliseconds, with no activity until the
+ first TCP keep-alive packet is sent.If set to 0, system
+ dependent default is used. (Automatically configured
+ unless set explicitly)
  --thread-cache-size=# 
  How many threads we should keep in a cache for reuse.
  These are freed after 5 minutes of idle time
@@ -1594,6 +1609,9 @@ table-definition-cache 400
 table-open-cache 431
 table-open-cache-instances 8
 tc-heuristic-recover OFF
+tcp-keepalive-interval 0
+tcp-keepalive-probes 0
+tcp-keepalive-time 0
 thread-cache-size 151
 thread-pool-idle-timeout 60
 thread-pool-max-threads 65536

mysql-test/suite/sys_vars/r/sysvars_server_embedded.result

@@ -220,6 +220,20 @@ NUMERIC_BLOCK_SIZE	NULL
 ENUM_VALUE_LIST	OFF,ON
 READ_ONLY	NO
 COMMAND_LINE_ARGUMENT	OPTIONAL
+VARIABLE_NAME	BINLOG_FILE_CACHE_SIZE
+SESSION_VALUE	NULL
+GLOBAL_VALUE	16384
+GLOBAL_VALUE_ORIGIN	COMPILE-TIME
+DEFAULT_VALUE	16384
+VARIABLE_SCOPE	GLOBAL
+VARIABLE_TYPE	BIGINT UNSIGNED
+VARIABLE_COMMENT	The size of file cache for the binary log
+NUMERIC_MIN_VALUE	8192
+NUMERIC_MAX_VALUE	18446744073709551615
+NUMERIC_BLOCK_SIZE	4096
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	REQUIRED
 VARIABLE_NAME	BINLOG_FORMAT
 SESSION_VALUE	MIXED
 GLOBAL_VALUE	MIXED
@@ -4112,6 +4126,48 @@ NUMERIC_BLOCK_SIZE	1
 ENUM_VALUE_LIST	NULL
 READ_ONLY	YES
 COMMAND_LINE_ARGUMENT	REQUIRED
+VARIABLE_NAME	TCP_KEEPALIVE_INTERVAL
+SESSION_VALUE	NULL
+GLOBAL_VALUE	0
+GLOBAL_VALUE_ORIGIN	COMPILE-TIME
+DEFAULT_VALUE	0
+VARIABLE_SCOPE	GLOBAL
+VARIABLE_TYPE	INT
+VARIABLE_COMMENT	The interval, in seconds, between when successive keep-alive packets are sent if no acknowledgement is received.If set to 0, system dependent default is used.
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	2147483
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	REQUIRED
+VARIABLE_NAME	TCP_KEEPALIVE_PROBES
+SESSION_VALUE	NULL
+GLOBAL_VALUE	0
+GLOBAL_VALUE_ORIGIN	COMPILE-TIME
+DEFAULT_VALUE	0
+VARIABLE_SCOPE	GLOBAL
+VARIABLE_TYPE	INT
+VARIABLE_COMMENT	The number of unacknowledged probes to send before considering the connection dead and notifying the application layer.If set to 0, system dependent default is used.
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	2147483
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	REQUIRED
+VARIABLE_NAME	TCP_KEEPALIVE_TIME
+SESSION_VALUE	NULL
+GLOBAL_VALUE	0
+GLOBAL_VALUE_ORIGIN	COMPILE-TIME
+DEFAULT_VALUE	0
+VARIABLE_SCOPE	GLOBAL
+VARIABLE_TYPE	INT
+VARIABLE_COMMENT	Timeout, in milliseconds, with no activity until the first TCP keep-alive packet is sent.If set to 0, system dependent default is used.
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	2147483
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	REQUIRED
 VARIABLE_NAME	THREAD_CACHE_SIZE
 SESSION_VALUE	NULL
 GLOBAL_VALUE	151

mysql-test/suite/sys_vars/r/sysvars_server_notembedded.result

@@ -4924,6 +4924,48 @@ NUMERIC_BLOCK_SIZE	1
 ENUM_VALUE_LIST	NULL
 READ_ONLY	YES
 COMMAND_LINE_ARGUMENT	REQUIRED
+VARIABLE_NAME	TCP_KEEPALIVE_INTERVAL
+SESSION_VALUE	NULL
+GLOBAL_VALUE	0
+GLOBAL_VALUE_ORIGIN	COMPILE-TIME
+DEFAULT_VALUE	0
+VARIABLE_SCOPE	GLOBAL
+VARIABLE_TYPE	INT
+VARIABLE_COMMENT	The interval, in seconds, between when successive keep-alive packets are sent if no acknowledgement is received.If set to 0, system dependent default is used.
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	2147483
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	REQUIRED
+VARIABLE_NAME	TCP_KEEPALIVE_PROBES
+SESSION_VALUE	NULL
+GLOBAL_VALUE	0
+GLOBAL_VALUE_ORIGIN	COMPILE-TIME
+DEFAULT_VALUE	0
+VARIABLE_SCOPE	GLOBAL
+VARIABLE_TYPE	INT
+VARIABLE_COMMENT	The number of unacknowledged probes to send before considering the connection dead and notifying the application layer.If set to 0, system dependent default is used.
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	2147483
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	REQUIRED
+VARIABLE_NAME	TCP_KEEPALIVE_TIME
+SESSION_VALUE	NULL
+GLOBAL_VALUE	0
+GLOBAL_VALUE_ORIGIN	COMPILE-TIME
+DEFAULT_VALUE	0
+VARIABLE_SCOPE	GLOBAL
+VARIABLE_TYPE	INT
+VARIABLE_COMMENT	Timeout, in milliseconds, with no activity until the first TCP keep-alive packet is sent.If set to 0, system dependent default is used.
+NUMERIC_MIN_VALUE	0
+NUMERIC_MAX_VALUE	2147483
+NUMERIC_BLOCK_SIZE	1
+ENUM_VALUE_LIST	NULL
+READ_ONLY	NO
+COMMAND_LINE_ARGUMENT	REQUIRED
 VARIABLE_NAME	THREAD_CACHE_SIZE
 SESSION_VALUE	NULL
 GLOBAL_VALUE	151

mysql-test/suite/sys_vars/r/tcp_keepalive_time.result

@@ -0,0 +1,9 @@
+set global tcp_keepalive_time = 1;
+set global tcp_keepalive_interval =1;
+set global tcp_keepalive_probes=1;
+connect con1,"127.0.0.1",root,,test,$MASTER_MYPORT,;
+disconnect con1;
+connection default;
+set global tcp_keepalive_time= default;
+set global tcp_keepalive_interval= default;
+set global tcp_keepalive_probes=default;

mysql-test/suite/sys_vars/t/tcp_keepalive_time.test

@@ -0,0 +1,11 @@
+--source include/not_embedded.inc
+# A smoke test for TCP keepalive options
+set global tcp_keepalive_time = 1;
+set global tcp_keepalive_interval =1;
+set global tcp_keepalive_probes=1;
+connect(con1,"127.0.0.1",root,,test,$MASTER_MYPORT,);
+disconnect con1;
+connection default;
+set global tcp_keepalive_time= default;
+set global tcp_keepalive_interval= default;
+set global tcp_keepalive_probes=default;

sql/sql_connect.cc

@@ -45,6 +45,7 @@ HASH global_index_stats;
 extern mysql_mutex_t LOCK_global_user_client_stats;
 extern mysql_mutex_t LOCK_global_table_stats;
 extern mysql_mutex_t LOCK_global_index_stats;
+extern vio_keepalive_opts opt_vio_keepalive;
 
 /*
   Get structure for logging connection data for the current user
@@ -1064,6 +1065,7 @@ static int check_connection(THD *thd)
     bzero((char*) &net->vio->remote, sizeof(net->vio->remote));
   }
   vio_keepalive(net->vio, TRUE);
+  vio_set_keepalive_options(net->vio, &opt_vio_keepalive);
 
   if (thd->packet.alloc(thd->variables.net_buffer_length))
   {

sql/sys_vars.cc

@@ -5311,6 +5311,38 @@ static Sys_var_ulong Sys_host_cache_size(
        NO_MUTEX_GUARD, NOT_IN_BINLOG, ON_CHECK(NULL),
        ON_UPDATE(fix_host_cache_size));
 
+vio_keepalive_opts opt_vio_keepalive;
+
+static Sys_var_int Sys_keepalive_time(
+       "tcp_keepalive_time",
+       "Timeout, in milliseconds, with no activity until the first TCP keep-alive packet is sent."
+       "If set to 0, system dependent default is used.",
+       AUTO_SET GLOBAL_VAR(opt_vio_keepalive.idle),
+       CMD_LINE(REQUIRED_ARG), VALID_RANGE(0, INT_MAX32/1000),
+       DEFAULT(0),
+       BLOCK_SIZE(1),
+       NO_MUTEX_GUARD, NOT_IN_BINLOG, ON_CHECK(NULL));
+
+static Sys_var_int Sys_keepalive_interval(
+       "tcp_keepalive_interval",
+       "The interval, in seconds, between when successive keep-alive packets are sent if no acknowledgement is received."
+       "If set to 0, system dependent default is used.",
+       AUTO_SET GLOBAL_VAR(opt_vio_keepalive.interval),
+       CMD_LINE(REQUIRED_ARG), VALID_RANGE(0, INT_MAX32/1000),
+       DEFAULT(0),
+       BLOCK_SIZE(1),
+       NO_MUTEX_GUARD, NOT_IN_BINLOG, ON_CHECK(NULL));
+
+static Sys_var_int Sys_keepalive_probes(
+       "tcp_keepalive_probes",
+       "The number of unacknowledged probes to send before considering the connection dead and notifying the application layer."
+       "If set to 0, system dependent default is used.",
+       AUTO_SET GLOBAL_VAR(opt_vio_keepalive.probes),
+       CMD_LINE(REQUIRED_ARG), VALID_RANGE(0, INT_MAX32/1000),
+       DEFAULT(0),
+       BLOCK_SIZE(1),
+       NO_MUTEX_GUARD, NOT_IN_BINLOG, ON_CHECK(NULL));
+
 static Sys_var_charptr Sys_ignore_db_dirs(
        "ignore_db_dirs",
        "Specifies a directory to add to the ignore list when collecting "

vio/viosocket.c

@@ -28,6 +28,7 @@
 #ifdef __WIN__
   #include <winsock2.h>
   #include <MSWSock.h>
+  #include <mstcpip.h>
   #pragma comment(lib, "ws2_32.lib")
 #endif
 #include "my_context.h"
@@ -522,6 +523,63 @@ int vio_keepalive(Vio* vio, my_bool set_keep_alive)
   DBUG_RETURN(r);
 }
 
+/*
+  Set socket options for keepalive e.g., TCP_KEEPCNT, TCP_KEEPIDLE/TCP_KEEPALIVE, TCP_KEEPINTVL
+*/
+int vio_set_keepalive_options(Vio* vio, const struct vio_keepalive_opts *opts)
+{
+#if defined _WIN32
+  struct tcp_keepalive s;
+  DWORD  nbytes;
+
+  if (vio->type == VIO_TYPE_NAMEDPIPE || vio->type == VIO_TYPE_SHARED_MEMORY)
+    return 0;
+
+  if (!opts->idle && !opts->interval)
+    return 0;
+
+  s.onoff= 1;
+  s.keepalivetime= opts->idle? opts->idle * 1000 : 7200;
+  s.keepaliveinterval= opts->interval?opts->interval * 1000 : 1;
+
+  return WSAIoctl(vio->mysql_socket.fd, SIO_KEEPALIVE_VALS, (LPVOID) &s, sizeof(s),
+           NULL, 0, &nbytes, NULL, NULL);
+
+#elif defined (TCP_KEEPIDLE) || defined (TCP_KEEPALIVE)
+
+  int ret= 0;
+  if (opts->idle)
+  {
+#ifdef TCP_KEEPIDLE // Linux only
+    ret= mysql_socket_setsockopt(vio->mysql_socket, IPPROTO_TCP, TCP_KEEPIDLE, (char *)&opts->idle, sizeof(opts->idle));
+#elif defined (TCP_KEEPALIVE)
+    ret= mysql_socket_setsockopt(vio->mysql_socket, IPPROTO_TCP, TCP_KEEPALIVE, (char *)&opts->idle, sizeof(opts->idle));
+#endif
+    if(ret)
+      return ret;
+  }
+
+#ifdef TCP_KEEPCNT // Linux only
+  if(opts->probes)
+  {
+    ret= mysql_socket_setsockopt(vio->mysql_socket, IPPROTO_TCP, TCP_KEEPCNT, (char *)&opts->probes, sizeof(opts->probes));
+    if(ret)
+      return ret;
+  }
+#endif
+
+#ifdef TCP_KEEPINTVL  // Linux only
+  if(opts->interval)
+  {
+    ret= mysql_socket_setsockopt(vio->mysql_socket, IPPROTO_TCP, TCP_KEEPINTVL, (char *)&opts->interval, sizeof(opts->interval));
+  }
+#endif
+  return ret;
+#else /*TCP_KEEPIDLE || TCP_KEEPALIVE */
+  return -1; 
+#endif
+}
+
 
 /**
   Indicate whether a I/O operation must be retried later.