Uint32
TransporterRegistry::poll_TCP(Uint32 timeOutMillis)
{
bool hasdata = false;
if (false && nTCPTransporters == 0)
{
tcpReadSelectReply = 0;
return 0;
}
NDB_SOCKET_TYPE maxSocketValue = -1;
// Needed for TCP/IP connections
// The read- and writeset are used by select
FD_ZERO(&tcpReadset);
// Prepare for sending and receiving
for (int i = 0; i < nTCPTransporters; i++) {
TCP_Transporter * t = theTCPTransporters[i];
// If the transporter is connected
NodeId nodeId = t->getRemoteNodeId();
if (is_connected(nodeId) && t->isConnected()) {
const NDB_SOCKET_TYPE socket = t->getSocket();
// Find the highest socket value. It will be used by select
if (socket > maxSocketValue)
maxSocketValue = socket;
// Put the connected transporters in the socket read-set
FD_SET(socket, &tcpReadset);
}
hasdata |= t->hasReceiveData();
}
timeOutMillis = hasdata ? 0 : timeOutMillis;
struct timeval timeout;
timeout.tv_sec = timeOutMillis / 1000;
timeout.tv_usec = (timeOutMillis % 1000) * 1000;
// The highest socket value plus one
maxSocketValue++;
tcpReadSelectReply = select(maxSocketValue, &tcpReadset, 0, 0, &timeout);
if(false && tcpReadSelectReply == -1 && errno == EINTR)
g_eventLogger.info("woke-up by signal");
#ifdef NDB_WIN32
if(tcpReadSelectReply == SOCKET_ERROR)
{
NdbSleep_MilliSleep(timeOutMillis);
}
#endif
return tcpReadSelectReply || hasdata;
}
void
ErrorReporter::handleError(int messageID,
const char* problemData,
const char* objRef,
NdbShutdownType nst)
{
WriteMessage(messageID, problemData,
objRef, theEmulatedJamIndex, theEmulatedJam);
g_eventLogger.info(problemData);
g_eventLogger.info(objRef);
childReportError(messageID);
if(messageID == NDBD_EXIT_ERROR_INSERT){
NdbShutdown(NST_ErrorInsert);
} else {
if (nst == NST_ErrorHandler)
nst = s_errorHandlerShutdownType;
NdbShutdown(nst);
}
}
/*
* MAIN
*/
int main(int argc, char** argv)
{
NDB_INIT(argv[0]);
load_defaults("my",load_default_groups,&argc,&argv);
int ho_error;
#ifndef DBUG_OFF
opt_debug= "d:t:O,/tmp/ndb_mgmd.trace";
#endif
if ((ho_error=handle_options(&argc, &argv, my_long_options,
ndb_std_get_one_option)))
exit(ho_error);
start:
glob= new MgmGlobals;
/**
* OSE specific. Enable shared ownership of file system resources.
* This is needed in order to use the cluster log since the events
* from the cluster is written from the 'ndb_receive'(NDBAPI) thread/process.
*/
#if defined NDB_OSE || defined NDB_SOFTOSE
efs_segment_share();
#endif
global_mgmt_server_check = 1;
if (opt_interactive ||
opt_non_interactive ||
g_print_full_config) {
opt_daemon= 0;
}
if (opt_mycnf && opt_config_filename)
{
ndbout_c("Both --mycnf and -f is not supported");
return 0;
}
if (opt_mycnf == 0 && opt_config_filename == 0)
{
struct stat buf;
if (stat("config.ini", &buf) != -1)
opt_config_filename = "config.ini";
}
glob->socketServer = new SocketServer();
MgmApiService * mapi = new MgmApiService();
glob->mgmObject = new MgmtSrvr(glob->socketServer,
opt_config_filename,
opt_connect_str);
if (g_print_full_config)
goto the_end;
if (glob->mgmObject->init())
goto error_end;
my_setwd(NdbConfig_get_path(0), MYF(0));
glob->localNodeId= glob->mgmObject->getOwnNodeId();
if (glob->localNodeId == 0) {
goto error_end;
}
glob->port= glob->mgmObject->getPort();
if (glob->port == 0)
goto error_end;
glob->interface_name = 0;
glob->use_specific_ip = false;
if(!glob->use_specific_ip){
int count= 5; // no of retries for tryBind
while(!glob->socketServer->tryBind(glob->port, glob->interface_name)){
if (--count > 0) {
NdbSleep_MilliSleep(1000);
continue;
}
ndbout_c("Unable to setup port: %s:%d!\n"
"Please check if the port is already used,\n"
"(perhaps a ndb_mgmd is already running),\n"
"and if you are executing on the correct computer",
(glob->interface_name ? glob->interface_name : "*"), glob->port);
goto error_end;
}
free(glob->interface_name);
glob->interface_name = 0;
}
if(!glob->socketServer->setup(mapi, &glob->port, glob->interface_name))
{
ndbout_c("Unable to setup management port: %d!\n"
"Please check if the port is already used,\n"
"(perhaps a ndb_mgmd is already running),\n"
"and if you are executing on the correct computer",
glob->port);
delete mapi;
goto error_end;
}
if(!glob->mgmObject->check_start()){
ndbout_c("Unable to check start management server.");
ndbout_c("Probably caused by illegal initial configuration file.");
goto error_end;
}
if (opt_daemon) {
// Become a daemon
char *lockfile= NdbConfig_PidFileName(glob->localNodeId);
char *logfile= NdbConfig_StdoutFileName(glob->localNodeId);
NdbAutoPtr<char> tmp_aptr1(lockfile), tmp_aptr2(logfile);
if (NdbDaemon_Make(lockfile, logfile, 0) == -1) {
ndbout << "Cannot become daemon: " << NdbDaemon_ErrorText << endl;
return 1;
}
}
#ifndef NDB_WIN32
signal(SIGPIPE, SIG_IGN);
#endif
{
BaseString error_string;
if(!glob->mgmObject->start(error_string)){
ndbout_c("Unable to start management server.");
ndbout_c("Probably caused by illegal initial configuration file.");
ndbout_c(error_string.c_str());
goto error_end;
}
}
//glob->mgmObject->saveConfig();
mapi->setMgm(glob->mgmObject);
char msg[256];
BaseString::snprintf(msg, sizeof(msg),
"NDB Cluster Management Server. %s", NDB_VERSION_STRING);
ndbout_c(msg);
g_eventLogger.info(msg);
BaseString::snprintf(msg, 256, "Id: %d, Command port: %d",
glob->localNodeId, glob->port);
ndbout_c(msg);
g_eventLogger.info(msg);
g_StopServer = false;
g_RestartServer= false;
glob->socketServer->startServer();
#if ! defined NDB_OSE && ! defined NDB_SOFTOSE
if(opt_interactive) {
BaseString con_str;
if(glob->interface_name)
con_str.appfmt("host=%s:%d", glob->interface_name, glob->port);
else
con_str.appfmt("localhost:%d", glob->port);
Ndb_mgmclient com(con_str.c_str(), 1);
while(g_StopServer != true && read_and_execute(&com, "ndb_mgm> ", 1));
} else
#endif
{
while(g_StopServer != true)
NdbSleep_MilliSleep(500);
}
if(g_RestartServer)
g_eventLogger.info("Restarting server...");
else
g_eventLogger.info("Shutting down server...");
glob->socketServer->stopServer();
// We disconnect from the ConfigRetreiver mgmd when we delete glob below
glob->socketServer->stopSessions(true);
g_eventLogger.info("Shutdown complete");
the_end:
delete glob;
if(g_RestartServer)
goto start;
ndb_end(opt_endinfo ? MY_CHECK_ERROR | MY_GIVE_INFO : 0);
return 0;
error_end:
delete glob;
ndb_end(opt_endinfo ? MY_CHECK_ERROR | MY_GIVE_INFO : 0);
return 1;
}
// run as own thread
void
TransporterRegistry::start_clients_thread()
{
int persist_mgm_count= 0;
DBUG_ENTER("TransporterRegistry::start_clients_thread");
while (m_run_start_clients_thread) {
NdbSleep_MilliSleep(100);
persist_mgm_count++;
if(persist_mgm_count==50)
{
ndb_mgm_check_connection(m_mgm_handle);
persist_mgm_count= 0;
}
for (int i= 0, n= 0; n < nTransporters && m_run_start_clients_thread; i++){
Transporter * t = theTransporters[i];
if (!t)
continue;
n++;
const NodeId nodeId = t->getRemoteNodeId();
switch(performStates[nodeId]){
case CONNECTING:
if(!t->isConnected() && !t->isServer) {
bool connected= false;
/**
* First, we try to connect (if we have a port number).
*/
if (t->get_s_port())
connected= t->connect_client();
/**
* If dynamic, get the port for connecting from the management server
*/
if( !connected && t->get_s_port() <= 0) { // Port is dynamic
int server_port= 0;
struct ndb_mgm_reply mgm_reply;
if(!ndb_mgm_is_connected(m_mgm_handle))
ndb_mgm_connect(m_mgm_handle, 0, 0, 0);
if(ndb_mgm_is_connected(m_mgm_handle))
{
int res=
ndb_mgm_get_connection_int_parameter(m_mgm_handle,
t->getRemoteNodeId(),
t->getLocalNodeId(),
CFG_CONNECTION_SERVER_PORT,
&server_port,
&mgm_reply);
DBUG_PRINT("info",("Got dynamic port %d for %d -> %d (ret: %d)",
server_port,t->getRemoteNodeId(),
t->getLocalNodeId(),res));
if( res >= 0 )
{
/**
* Server_port == 0 just means that that a mgmt server
* has not received a new port yet. Keep the old.
*/
if (server_port)
t->set_s_port(server_port);
}
else if(ndb_mgm_is_connected(m_mgm_handle))
{
g_eventLogger.info("Failed to get dynamic port to connect to: %d", res);
ndb_mgm_disconnect(m_mgm_handle);
}
else
{
g_eventLogger.info("Management server closed connection early. "
"It is probably being shut down (or has problems). "
"We will retry the connection. %d %s %s line: %d",
ndb_mgm_get_latest_error(m_mgm_handle),
ndb_mgm_get_latest_error_desc(m_mgm_handle),
ndb_mgm_get_latest_error_msg(m_mgm_handle),
ndb_mgm_get_latest_error_line(m_mgm_handle)
);
}
}
/** else
* We will not be able to get a new port unless
* the m_mgm_handle is connected. Note that not
* being connected is an ok state, just continue
* until it is able to connect. Continue using the
* old port until we can connect again and get a
* new port.
*/
}
}
break;
case DISCONNECTING:
if(t->isConnected())
t->doDisconnect();
break;
default:
break;
}
}
}
DBUG_VOID_RETURN;
}
void
MgmApiSession::get_nodeid(Parser_t::Context &,
const class Properties &args)
{
const char *cmd= "get nodeid reply";
Uint32 version, nodeid= 0, nodetype= 0xff;
Uint32 timeout= 20; // default seconds timeout
const char * transporter;
const char * user;
const char * password;
const char * public_key;
const char * endian= NULL;
const char * name= NULL;
Uint32 log_event= 1;
bool log_event_version;
union { long l; char c[sizeof(long)]; } endian_check;
args.get("version", &version);
args.get("nodetype", &nodetype);
args.get("transporter", &transporter);
args.get("nodeid", &nodeid);
args.get("user", &user);
args.get("password", &password);
args.get("public key", &public_key);
args.get("endian", &endian);
args.get("name", &name);
args.get("timeout", &timeout);
/* for backwards compatability keep track if client uses new protocol */
log_event_version= args.get("log_event", &log_event);
endian_check.l = 1;
if(endian
&& strcmp(endian,(endian_check.c[sizeof(long)-1])?"big":"little")!=0) {
m_output->println(cmd);
m_output->println("result: Node does not have the same endianness as the management server.");
m_output->println("");
return;
}
bool compatible;
switch (nodetype) {
case NODE_TYPE_MGM:
case NODE_TYPE_API:
compatible = ndbCompatible_mgmt_api(NDB_VERSION, version);
break;
case NODE_TYPE_DB:
compatible = ndbCompatible_mgmt_ndb(NDB_VERSION, version);
break;
default:
m_output->println(cmd);
m_output->println("result: unknown nodetype %d", nodetype);
m_output->println("");
return;
}
struct sockaddr_in addr;
SOCKET_SIZE_TYPE addrlen= sizeof(addr);
int r = getpeername(m_socket, (struct sockaddr*)&addr, &addrlen);
if (r != 0 ) {
m_output->println(cmd);
m_output->println("result: getpeername(%d) failed, err= %d", m_socket, r);
m_output->println("");
return;
}
NodeId tmp= nodeid;
if(tmp == 0 || !m_allocated_resources->is_reserved(tmp)){
BaseString error_string;
int error_code;
NDB_TICKS tick= 0;
/* only report error on second attempt as not to clog the cluster log */
while (!m_mgmsrv.alloc_node_id(&tmp, (enum ndb_mgm_node_type)nodetype,
(struct sockaddr*)&addr, &addrlen,
error_code, error_string,
tick == 0 ? 0 : log_event))
{
/* NDB_MGM_ALLOCID_CONFIG_MISMATCH is a non retriable error */
if (tick == 0 && error_code != NDB_MGM_ALLOCID_CONFIG_MISMATCH)
{
// attempt to free any timed out reservations
tick= NdbTick_CurrentMillisecond();
struct PurgeStruct ps;
m_mgmsrv.get_connected_nodes(ps.free_nodes);
// invert connected_nodes to get free nodes
ps.free_nodes.bitXORC(NodeBitmask());
ps.str= 0;
ps.tick= tick;
m_mgmsrv.get_socket_server()->
foreachSession(stop_session_if_timed_out,&ps);
m_mgmsrv.get_socket_server()->checkSessions();
error_string = "";
continue;
}
const char *alias;
const char *str;
alias= ndb_mgm_get_node_type_alias_string((enum ndb_mgm_node_type)
nodetype, &str);
m_output->println(cmd);
m_output->println("result: %s", error_string.c_str());
/* only use error_code protocol if client knows about it */
if (log_event_version)
m_output->println("error_code: %d", error_code);
m_output->println("");
return;
}
}
#if 0
if (!compatible){
m_output->println(cmd);
m_output->println("result: incompatible version mgmt 0x%x and node 0x%x",
NDB_VERSION, version);
m_output->println("");
return;
}
#endif
m_output->println(cmd);
m_output->println("nodeid: %u", tmp);
m_output->println("result: Ok");
m_output->println("");
m_allocated_resources->reserve_node(tmp, timeout*1000);
if (name)
g_eventLogger.info("Node %d: %s", tmp, name);
return;
}
void
WatchDog::run()
{
unsigned int anIPValue, sleep_time;
unsigned int oldIPValue = 0;
unsigned int theIntervalCheck = theInterval;
struct MicroSecondTimer start_time, last_time, now;
NdbTick_getMicroTimer(&start_time);
last_time = start_time;
// WatchDog for the single threaded NDB
while (!theStop)
{
sleep_time= 100;
NdbSleep_MilliSleep(sleep_time);
if(theStop)
break;
NdbTick_getMicroTimer(&now);
if (NdbTick_getMicrosPassed(last_time, now)/1000 > sleep_time*2)
{
struct tms my_tms;
times(&my_tms);
g_eventLogger.info("Watchdog: User time: %llu System time: %llu",
(Uint64)my_tms.tms_utime,
(Uint64)my_tms.tms_stime);
g_eventLogger.warning("Watchdog: Warning overslept %u ms, expected %u ms.",
NdbTick_getMicrosPassed(last_time, now)/1000,
sleep_time);
}
last_time = now;
// Verify that the IP thread is not stuck in a loop
anIPValue = *theIPValue;
if (anIPValue != 0)
{
oldIPValue = anIPValue;
globalData.incrementWatchDogCounter(0);
NdbTick_getMicroTimer(&start_time);
theIntervalCheck = theInterval;
}
else
{
int warn = 1;
Uint32 elapsed = NdbTick_getMicrosPassed(start_time, now)/1000;
/*
oldIPValue == 9 indicates malloc going on, this can take some time
so only warn if we pass the watchdog interval
*/
if (oldIPValue == 9)
if (elapsed < theIntervalCheck)
warn = 0;
else
theIntervalCheck += theInterval;
if (warn)
{
const char *last_stuck_action = get_action(oldIPValue);
g_eventLogger.warning("Ndb kernel is stuck in: %s", last_stuck_action);
{
struct tms my_tms;
times(&my_tms);
g_eventLogger.info("Watchdog: User time: %llu System time: %llu",
(Uint64)my_tms.tms_utime,
(Uint64)my_tms.tms_stime);
}
if (elapsed > 3 * theInterval)
{
shutdownSystem(last_stuck_action);
}
}
}
}
return;
}