int
NdbRestarter::connect(){
disconnect();
handle = ndb_mgm_create_handle();
if (handle == NULL){
g_err << "handle == NULL" << endl;
return -1;
}
g_info << "Connecting to mgmsrv at " << addr.c_str() << endl;
if (ndb_mgm_set_connectstring(handle,addr.c_str()))
{
MGMERR(handle);
g_err << "Connection to " << addr.c_str() << " failed" << endl;
return -1;
}
if (ndb_mgm_connect(handle, 0, 0, 0) == -1)
{
MGMERR(handle);
g_err << "Connection to " << addr.c_str() << " failed" << endl;
return -1;
}
connected = true;
return 0;
}
int NdbRestarter::restartAll(bool initial,
bool nostart,
bool abort){
if (!isConnected())
return -1;
if (ndb_mgm_restart2(handle, 0, NULL, initial, 1, abort) == -1) {
MGMERR(handle);
g_err << "Could not restart(stop) all nodes " << endl;
// return -1; Continue anyway - Magnus
}
if (waitClusterNoStart(60) != 0){
g_err << "Cluster didnt enter STATUS_NOT_STARTED within 60s" << endl;
return -1;
}
if(nostart){
g_debug << "restartAll: nostart == true" << endl;
return 0;
}
if (ndb_mgm_start(handle, 0, NULL) == -1) {
MGMERR(handle);
g_err << "Could not restart(start) all nodes " << endl;
return -1;
}
return 0;
}
int
NdbRestarter::restartNodes(int * nodes, int cnt,
Uint32 flags)
{
if (!isConnected())
return -1;
int ret = 0;
int unused;
if ((ret = ndb_mgm_restart4(handle, cnt, nodes,
(flags & NRRF_INITIAL),
(flags & NRRF_NOSTART),
(flags & NRRF_ABORT),
(flags & NRRF_FORCE),
&unused)) <= 0)
{
/**
* ndb_mgm_restart4 returned error, one reason could
* be that the node have not stopped fast enough!
* Check status of the node to see if it's on the
* way down. If that's the case ignore the error
*/
if (getStatus() != 0)
return -1;
g_info << "ndb_mgm_restart4 returned with error, checking node state"
<< endl;
for (int j = 0; j<cnt; j++)
{
int _nodeId = nodes[j];
for(unsigned i = 0; i < ndbNodes.size(); i++)
{
if(ndbNodes[i].node_id == _nodeId)
{
g_info <<_nodeId<<": status="<<ndbNodes[i].node_status<<endl;
/* Node found check state */
switch(ndbNodes[i].node_status){
case NDB_MGM_NODE_STATUS_RESTARTING:
case NDB_MGM_NODE_STATUS_SHUTTING_DOWN:
break;
default:
MGMERR(handle);
g_err << "Could not stop node with id = "<< _nodeId << endl;
return -1;
}
}
}
}
}
if ((flags & NRRF_NOSTART) == 0)
{
wait_until_ready(nodes, cnt);
}
return 0;
}
int NdbRestarter::startNodes(const int * nodes, int num_nodes){
if (!isConnected())
return -1;
if (ndb_mgm_start(handle, num_nodes, nodes) != num_nodes) {
MGMERR(handle);
g_err << "Could not start all nodes " << endl;
return -1;
}
return 0;
}
int NdbRestarter::startAll(){
if (!isConnected())
return -1;
if (ndb_mgm_start(handle, 0, NULL) == -1) {
MGMERR(handle);
g_err << "Could not start all nodes " << endl;
return -1;
}
return 0;
}
int NdbRestarter::insertErrorInNode(int _nodeId, int _error){
if (!isConnected())
return -1;
ndb_mgm_reply reply;
reply.return_code = 0;
if (ndb_mgm_insert_error(handle, _nodeId, _error, &reply) == -1){
MGMERR(handle);
g_err << "Could not insert error in node with id = "<< _nodeId << endl;
}
if(reply.return_code != 0){
g_err << "Error: " << reply.message << endl;
}
return 0;
}
int NdbRestarter::exitSingleUserMode(){
if (!isConnected())
return -1;
ndb_mgm_reply reply;
reply.return_code = 0;
if (ndb_mgm_exit_single_user(handle, &reply) == -1){
MGMERR(handle);
g_err << "Could not exit single user mode " << endl;
}
if(reply.return_code != 0){
g_err << "Error: " << reply.message << endl;
}
return reply.return_code;
}
int NdbRestarter::dumpStateOneNode(int _nodeId, const int * _args, int _num_args){
if (!isConnected())
return -1;
ndb_mgm_reply reply;
reply.return_code = 0;
if (ndb_mgm_dump_state(handle, _nodeId, _args, _num_args, &reply) == -1){
MGMERR(handle);
g_err << "Could not dump state in node with id = "<< _nodeId << endl;
}
if(reply.return_code != 0){
g_err << "Error: " << reply.message << endl;
}
return reply.return_code;
}
int
NdbRestarter::restartOneDbNode(int _nodeId,
bool inital,
bool nostart,
bool abort){
if (!isConnected())
return -1;
int ret = 0;
if ((ret = ndb_mgm_restart2(handle, 1, &_nodeId,
inital, nostart, abort)) <= 0) {
/**
* ndb_mgm_restart2 returned error, one reason could
* be that the node have not stopped fast enough!
* Check status of the node to see if it's on the
* way down. If that's the case ignore the error
*/
if (getStatus() != 0)
return -1;
g_info << "ndb_mgm_restart2 returned with error, checking node state" << endl;
for(size_t i = 0; i < ndbNodes.size(); i++){
if(ndbNodes[i].node_id == _nodeId){
g_info <<_nodeId<<": status="<<ndbNodes[i].node_status<<endl;
/* Node found check state */
switch(ndbNodes[i].node_status){
case NDB_MGM_NODE_STATUS_RESTARTING:
case NDB_MGM_NODE_STATUS_SHUTTING_DOWN:
return 0;
default:
break;
}
}
}
MGMERR(handle);
g_err << "Could not stop node with id = "<< _nodeId << endl;
return -1;
}
return 0;
}
int NdbRestarter::enterSingleUserMode(int _nodeId){
if (!isConnected())
return -1;
ndb_mgm_reply reply;
reply.return_code = 0;
if (ndb_mgm_enter_single_user(handle, _nodeId, &reply) == -1){
MGMERR(handle);
g_err << "Could not enter single user mode api node = "<< _nodeId << endl;
}
if(reply.return_code != 0){
g_err << "Error: " << reply.message << endl;
}
return reply.return_code;
}
int
NdbRestarter::getStatus(){
int retries = 0;
struct ndb_mgm_cluster_state * status;
struct ndb_mgm_node_state * node;
ndbNodes.clear();
mgmNodes.clear();
apiNodes.clear();
if (!isConnected())
return -1;
while(retries < 10){
status = ndb_mgm_get_status(handle);
if (status == NULL){
ndbout << "status==NULL, retries="<<retries<<endl;
MGMERR(handle);
retries++;
continue;
}
for (int i = 0; i < status->no_of_nodes; i++){
node = &status->node_states[i];
switch(node->node_type){
case NDB_MGM_NODE_TYPE_NDB:
ndbNodes.push_back(*node);
break;
case NDB_MGM_NODE_TYPE_MGM:
mgmNodes.push_back(*node);
break;
case NDB_MGM_NODE_TYPE_API:
apiNodes.push_back(*node);
break;
default:
if(node->node_status == NDB_MGM_NODE_STATUS_UNKNOWN ||
node->node_status == NDB_MGM_NODE_STATUS_NO_CONTACT){
retries++;
ndbNodes.clear();
mgmNodes.clear();
apiNodes.clear();
free(status);
status = NULL;
i = status->no_of_nodes;
ndbout << "kalle"<< endl;
break;
}
abort();
break;
}
}
if(status == 0){
ndbout << "status == 0" << endl;
continue;
}
free(status);
return 0;
}
g_err << "getStatus failed" << endl;
return -1;
}
static int
waitClusterStatus(const char* _addr,
ndb_mgm_node_status _status)
{
int _startphase = -1;
/* Ignore SIGPIPE */
signal(SIGPIPE, SIG_IGN);
handle = ndb_mgm_create_handle();
if (handle == NULL){
g_err << "Could not create ndb_mgm handle" << endl;
return -1;
}
g_info << "Connecting to mgmsrv at " << _addr << endl;
if (ndb_mgm_set_connectstring(handle, _addr))
{
MGMERR(handle);
g_err << "Connectstring " << _addr << " invalid" << endl;
return -1;
}
if (ndb_mgm_connect(handle,0,0,1)) {
MGMERR(handle);
g_err << "Connection to " << _addr << " failed" << endl;
return -1;
}
int attempts = 0;
int resetAttempts = 0;
const int MAX_RESET_ATTEMPTS = 10;
bool allInState = false;
int timeout_ms= _timeout * 10; /* In number of 100 milliseconds */
while (allInState == false){
if (_timeout > 0 && attempts > _timeout){
/**
* Timeout has expired waiting for the nodes to enter
* the state we want
*/
bool waitMore = false;
/**
* Make special check if we are waiting for
* cluster to become started
*/
if(_status == NDB_MGM_NODE_STATUS_STARTED){
waitMore = true;
/**
* First check if any node is not starting
* then it's no idea to wait anymore
*/
for (size_t n = 0; n < ndbNodes.size(); n++){
if (ndbNodes[n].node_status != NDB_MGM_NODE_STATUS_STARTED &&
ndbNodes[n].node_status != NDB_MGM_NODE_STATUS_STARTING)
waitMore = false;
}
}
if (!waitMore || resetAttempts > MAX_RESET_ATTEMPTS){
g_err << "waitNodeState("
<< ndb_mgm_get_node_status_string(_status)
<<", "<<_startphase<<")"
<< " timeout after " << attempts <<" attemps" << endl;
return -1;
}
g_err << "waitNodeState("
<< ndb_mgm_get_node_status_string(_status)
<<", "<<_startphase<<")"
<< " resetting number of attempts "
<< resetAttempts << endl;
attempts = 0;
resetAttempts++;
}
if (getStatus() != 0){
return -1;
}
/* Assume all nodes are in state(if there is any) */
allInState = (ndbNodes.size() > 0);
/* Loop through all nodes and check their state */
for (size_t n = 0; n < ndbNodes.size(); n++) {
ndb_mgm_node_state* ndbNode = &ndbNodes[n];
assert(ndbNode != NULL);
g_info << "Node " << ndbNode->node_id << ": "
<< ndb_mgm_get_node_status_string(ndbNode->node_status)<< endl;
if (ndbNode->node_status != _status)
allInState = false;
}
if (!allInState) {
g_info << "Waiting for cluster enter state "
<< ndb_mgm_get_node_status_string(_status)<< endl;
NdbSleep_MilliSleep(100);
}
attempts++;
}
return 0;
}
int
getStatus(){
int retries = 0;
struct ndb_mgm_cluster_state * status;
struct ndb_mgm_node_state * node;
ndbNodes.clear();
while(retries < 10){
status = ndb_mgm_get_status(handle);
if (status == NULL){
ndbout << "status==NULL, retries="<<retries<<endl;
MGMERR(handle);
retries++;
ndb_mgm_disconnect(handle);
if (ndb_mgm_connect(handle,0,0,1)) {
MGMERR(handle);
g_err << "Reconnect failed" << endl;
break;
}
continue;
}
int count = status->no_of_nodes;
for (int i = 0; i < count; i++){
node = &status->node_states[i];
switch(node->node_type){
case NDB_MGM_NODE_TYPE_NDB:
ndbNodes.push_back(*node);
break;
case NDB_MGM_NODE_TYPE_MGM:
/* Don't care about MGM nodes */
break;
case NDB_MGM_NODE_TYPE_API:
/* Don't care about API nodes */
break;
default:
if(node->node_status == NDB_MGM_NODE_STATUS_UNKNOWN ||
node->node_status == NDB_MGM_NODE_STATUS_NO_CONTACT){
retries++;
ndbNodes.clear();
free(status);
status = NULL;
count = 0;
ndbout << "kalle"<< endl;
break;
}
abort();
break;
}
}
if(status == 0){
ndbout << "status == 0" << endl;
continue;
}
free(status);
return 0;
}
return -1;
}
int
NdbRestarter::getStatus(){
int retries = 0;
struct ndb_mgm_cluster_state * status;
struct ndb_mgm_node_state * node;
ndbNodes.clear();
mgmNodes.clear();
apiNodes.clear();
if (!isConnected())
return -1;
while(retries < 10){
status = ndb_mgm_get_status(handle);
if (status == NULL){
if (m_reconnect){
if (connect() == 0){
g_err << "Reconnected..." << endl;
continue;
}
const int err = ndb_mgm_get_latest_error(handle);
if (err == NDB_MGM_COULD_NOT_CONNECT_TO_SOCKET){
g_err << "Could not connect to socket, sleep and retry" << endl;
retries= 0;
NdbSleep_SecSleep(1);
continue;
}
}
const int err = ndb_mgm_get_latest_error(handle);
ndbout << "status==NULL, retries="<<retries<< " err=" << err << endl;
MGMERR(handle);
retries++;
continue;
}
for (int i = 0; i < status->no_of_nodes; i++){
node = &status->node_states[i];
switch(node->node_type){
case NDB_MGM_NODE_TYPE_NDB:
ndbNodes.push_back(*node);
break;
case NDB_MGM_NODE_TYPE_MGM:
mgmNodes.push_back(*node);
break;
case NDB_MGM_NODE_TYPE_API:
apiNodes.push_back(*node);
break;
default:
if(node->node_status == NDB_MGM_NODE_STATUS_UNKNOWN ||
node->node_status == NDB_MGM_NODE_STATUS_NO_CONTACT){
retries++;
ndbNodes.clear();
mgmNodes.clear();
apiNodes.clear();
free(status);
status = NULL;
i = status->no_of_nodes;
ndbout << "kalle"<< endl;
break;
}
abort();
break;
}
}
if(status == 0){
ndbout << "status == 0" << endl;
continue;
}
free(status);
return 0;
}
g_err << "getStatus failed" << endl;
return -1;
}
