diff options
Diffstat (limited to 'ndb')
-rw-r--r-- | ndb/include/mgmapi/mgmapi.h | 57 | ||||
-rw-r--r-- | ndb/src/mgmapi/mgmapi.cpp | 240 | ||||
-rw-r--r-- | ndb/src/mgmclient/CommandInterpreter.cpp | 239 | ||||
-rw-r--r-- | ndb/src/mgmsrv/MgmtSrvr.cpp | 94 | ||||
-rw-r--r-- | ndb/src/mgmsrv/MgmtSrvr.hpp | 22 | ||||
-rw-r--r-- | ndb/src/mgmsrv/Services.cpp | 86 | ||||
-rw-r--r-- | ndb/src/mgmsrv/Services.hpp | 9 |
7 files changed, 541 insertions, 206 deletions
diff --git a/ndb/include/mgmapi/mgmapi.h b/ndb/include/mgmapi/mgmapi.h index 0f3d75d50fb..4e015cc3d7e 100644 --- a/ndb/include/mgmapi/mgmapi.h +++ b/ndb/include/mgmapi/mgmapi.h @@ -701,6 +701,28 @@ extern "C" { const int * node_list, int abort); /** + * Stops cluster nodes + * + * @param handle Management handle. + * @param no_of_nodes Number of database nodes to stop<br> + * -1: All database and management nodes<br> + * 0: All database nodes in cluster<br> + * n: Stop the <var>n</var> node(s) specified in + * the array node_list + * @param node_list List of node IDs of database nodes to be stopped + * @param abort Don't perform graceful stop, + * but rather stop immediately + * @param disconnect Returns true if you need to disconnect to apply + * the stop command (e.g. stopping the mgm server + * that handle is connected to) + * + * @return Number of nodes stopped (-1 on error). + */ + int ndb_mgm_stop3(NdbMgmHandle handle, int no_of_nodes, + const int * node_list, int abort, int *disconnect); + + + /** * Restart database nodes * * @param handle Management handle. @@ -740,6 +762,31 @@ extern "C" { int nostart, int abort); /** + * Restart nodes + * + * @param handle Management handle. + * @param no_of_nodes Number of database nodes to be restarted:<br> + * 0: Restart all database nodes in the cluster<br> + * n: Restart the <var>n</var> node(s) specified in the + * array node_list + * @param node_list List of node IDs of database nodes to be restarted + * @param initial Remove filesystem from restarting node(s) + * @param nostart Don't actually start node(s) but leave them + * waiting for start command + * @param abort Don't perform graceful restart, + * but rather restart immediately + * @param disconnect Returns true if mgmapi client must disconnect from + * server to apply the requested operation. (e.g. + * restart the management server) + * + * + * @return Number of nodes stopped (-1 on error). + */ + int ndb_mgm_restart3(NdbMgmHandle handle, int no_of_nodes, + const int * node_list, int initial, + int nostart, int abort, int *disconnect); + + /** * Start database nodes * * @param handle Management handle. @@ -1030,6 +1077,16 @@ extern "C" { Uint32 ndb_mgm_get_mgmd_nodeid(NdbMgmHandle handle); /** + * Get the version of the mgm server we're talking to. + * Designed to allow switching of protocol depending on version + * so that new clients can speak to old servers in a compat mode + */ + int ndb_mgm_get_version(NdbMgmHandle handle, + int *major, int *minor, int* build, + int len, char* str); + + + /** * Config iterator */ typedef struct ndb_mgm_configuration_iterator ndb_mgm_configuration_iterator; diff --git a/ndb/src/mgmapi/mgmapi.cpp b/ndb/src/mgmapi/mgmapi.cpp index 061360428af..7ad20e504f6 100644 --- a/ndb/src/mgmapi/mgmapi.cpp +++ b/ndb/src/mgmapi/mgmapi.cpp @@ -27,6 +27,7 @@ #include <mgmapi_debug.h> #include "mgmapi_configuration.hpp" #include <socket_io.h> +#include <version.h> #include <NdbOut.hpp> #include <SocketServer.hpp> @@ -103,6 +104,9 @@ struct ndb_mgm_handle { #endif FILE *errstream; char *m_name; + int mgmd_version_major; + int mgmd_version_minor; + int mgmd_version_build; }; #define SET_ERROR(h, e, s) setError(h, e, __LINE__, s) @@ -168,6 +172,10 @@ ndb_mgm_create_handle() h->logfile = 0; #endif + h->mgmd_version_major= -1; + h->mgmd_version_minor= -1; + h->mgmd_version_build= -1; + DBUG_PRINT("info", ("handle=0x%x", (UintPtr)h)); DBUG_RETURN(h); } @@ -361,8 +369,9 @@ ndb_mgm_call(NdbMgmHandle handle, const ParserRow<ParserDummy> *command_reply, * Print some info about why the parser returns NULL */ fprintf(handle->errstream, - "Error in mgm protocol parser. cmd: >%s< status: %d curr: %d\n", - cmd, (Uint32)ctx.m_status, ctx.m_currentToken); + "Error in mgm protocol parser. cmd: >%s< status: %d curr: %s\n", + cmd, (Uint32)ctx.m_status, + (ctx.m_currentToken)?ctx.m_currentToken:"NULL"); DBUG_PRINT("info",("ctx.status: %d, ctx.m_currentToken: %s", ctx.m_status, ctx.m_currentToken)); } @@ -678,7 +687,11 @@ ndb_mgm_get_status(NdbMgmHandle handle) out.println(""); char buf[1024]; - in.gets(buf, sizeof(buf)); + if(!in.gets(buf, sizeof(buf))) + { + SET_ERROR(handle, NDB_MGM_ILLEGAL_SERVER_REPLY, "Probably disconnected"); + return NULL; + } if(buf[strlen(buf)-1] == '\n') buf[strlen(buf)-1] = '\0'; @@ -687,7 +700,11 @@ ndb_mgm_get_status(NdbMgmHandle handle) return NULL; } - in.gets(buf, sizeof(buf)); + if(!in.gets(buf, sizeof(buf))) + { + SET_ERROR(handle, NDB_MGM_ILLEGAL_SERVER_REPLY, "Probably disconnected"); + return NULL; + } if(buf[strlen(buf)-1] == '\n') buf[strlen(buf)-1] = '\0'; @@ -710,6 +727,13 @@ ndb_mgm_get_status(NdbMgmHandle handle) malloc(sizeof(ndb_mgm_cluster_state)+ noOfNodes*(sizeof(ndb_mgm_node_state)+sizeof("000.000.000.000#"))); + if(!state) + { + SET_ERROR(handle, NDB_MGM_OUT_OF_MEMORY, + "Allocating ndb_mgm_cluster_state"); + return NULL; + } + state->no_of_nodes= noOfNodes; ndb_mgm_node_state * ptr = &state->node_states[0]; int nodeId = 0; @@ -719,7 +743,13 @@ ndb_mgm_get_status(NdbMgmHandle handle) } i = -1; ptr--; for(; i<noOfNodes; ){ - in.gets(buf, sizeof(buf)); + if(!in.gets(buf, sizeof(buf))) + { + free(state); + SET_ERROR(handle, NDB_MGM_ILLEGAL_SERVER_REPLY, + "Probably disconnected"); + return NULL; + } tmp.assign(buf); if(tmp.trim() == ""){ @@ -829,37 +859,81 @@ ndb_mgm_stop(NdbMgmHandle handle, int no_of_nodes, const int * node_list) return ndb_mgm_stop2(handle, no_of_nodes, node_list, 0); } - extern "C" -int +int ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, int abort) { - SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_stop2"); - const ParserRow<ParserDummy> stop_reply[] = { + int disconnect; + return ndb_mgm_stop3(handle, no_of_nodes, node_list, abort, &disconnect); +} + + +extern "C" +int +ndb_mgm_stop3(NdbMgmHandle handle, int no_of_nodes, const int * node_list, + int abort, int *disconnect) +{ + SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_stop3"); + const ParserRow<ParserDummy> stop_reply_v1[] = { + MGM_CMD("stop reply", NULL, ""), + MGM_ARG("stopped", Int, Optional, "No of stopped nodes"), + MGM_ARG("result", String, Mandatory, "Error message"), + MGM_END() + }; + const ParserRow<ParserDummy> stop_reply_v2[] = { MGM_CMD("stop reply", NULL, ""), MGM_ARG("stopped", Int, Optional, "No of stopped nodes"), MGM_ARG("result", String, Mandatory, "Error message"), + MGM_ARG("disconnect", Int, Mandatory, "Need to disconnect"), MGM_END() }; + CHECK_HANDLE(handle, -1); CHECK_CONNECTED(handle, -1); - if(no_of_nodes < 0){ + if(handle->mgmd_version_build==-1) + { + char verstr[50]; + if(!ndb_mgm_get_version(handle, + &(handle->mgmd_version_major), + &(handle->mgmd_version_minor), + &(handle->mgmd_version_build), + sizeof(verstr), + verstr)) + { + return -1; + } + } + int use_v2= ((handle->mgmd_version_major==5) + && ( + (handle->mgmd_version_minor==0 && handle->mgmd_version_build>=21) + ||(handle->mgmd_version_minor==1 && handle->mgmd_version_build>=12) + ||(handle->mgmd_version_minor>1) + ) + ) + || (handle->mgmd_version_major>5); + + if(no_of_nodes < -1){ SET_ERROR(handle, NDB_MGM_ILLEGAL_NUMBER_OF_NODES, "Negative number of nodes requested to stop"); return -1; } Uint32 stoppedNoOfNodes = 0; - if(no_of_nodes == 0){ + if(no_of_nodes <= 0){ /** - * All database nodes should be stopped + * All nodes should be stopped (all or just db) */ Properties args; args.put("abort", abort); + if(use_v2) + args.put("stop", (no_of_nodes==-1)?"mgm,db":"db"); const Properties *reply; - reply = ndb_mgm_call(handle, stop_reply, "stop all", &args); + if(use_v2) + reply = ndb_mgm_call(handle, stop_reply_v2, "stop all", &args); + else + reply = ndb_mgm_call(handle, stop_reply_v1, "stop all", &args); CHECK_REPLY(reply, -1); if(!reply->get("stopped", &stoppedNoOfNodes)){ @@ -868,6 +942,10 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, delete reply; return -1; } + if(use_v2) + reply->get("disconnect", (Uint32*)disconnect); + else + *disconnect= 0; BaseString result; reply->get("result", result); if(strcmp(result.c_str(), "Ok") != 0) { @@ -893,7 +971,11 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, args.put("abort", abort); const Properties *reply; - reply = ndb_mgm_call(handle, stop_reply, "stop", &args); + if(use_v2) + reply = ndb_mgm_call(handle, stop_reply_v2, "stop v2", &args); + else + reply = ndb_mgm_call(handle, stop_reply_v1, "stop", &args); + CHECK_REPLY(reply, stoppedNoOfNodes); if(!reply->get("stopped", &stoppedNoOfNodes)){ SET_ERROR(handle, NDB_MGM_STOP_FAILED, @@ -901,6 +983,10 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, delete reply; return -1; } + if(use_v2) + reply->get("disconnect", (Uint32*)disconnect); + else + *disconnect= 0; BaseString result; reply->get("result", result); if(strcmp(result.c_str(), "Ok") != 0) { @@ -914,20 +1000,69 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, extern "C" int +ndb_mgm_restart(NdbMgmHandle handle, int no_of_nodes, const int *node_list) +{ + SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart"); + return ndb_mgm_restart2(handle, no_of_nodes, node_list, 0, 0, 0); +} + +extern "C" +int ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, int initial, int nostart, int abort) { - SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart2"); + int disconnect; + + return ndb_mgm_restart3(handle, no_of_nodes, node_list, initial, nostart, + abort, &disconnect); +} + +extern "C" +int +ndb_mgm_restart3(NdbMgmHandle handle, int no_of_nodes, const int * node_list, + int initial, int nostart, int abort, int *disconnect) +{ + SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart3"); Uint32 restarted = 0; - const ParserRow<ParserDummy> restart_reply[] = { + const ParserRow<ParserDummy> restart_reply_v1[] = { MGM_CMD("restart reply", NULL, ""), MGM_ARG("result", String, Mandatory, "Error message"), MGM_ARG("restarted", Int, Optional, "No of restarted nodes"), MGM_END() }; + const ParserRow<ParserDummy> restart_reply_v2[] = { + MGM_CMD("restart reply", NULL, ""), + MGM_ARG("result", String, Mandatory, "Error message"), + MGM_ARG("restarted", Int, Optional, "No of restarted nodes"), + MGM_ARG("disconnect", Int, Optional, "Disconnect to apply"), + MGM_END() + }; + CHECK_HANDLE(handle, -1); CHECK_CONNECTED(handle, -1); - + + if(handle->mgmd_version_build==-1) + { + char verstr[50]; + if(!ndb_mgm_get_version(handle, + &(handle->mgmd_version_major), + &(handle->mgmd_version_minor), + &(handle->mgmd_version_build), + sizeof(verstr), + verstr)) + { + return -1; + } + } + int use_v2= ((handle->mgmd_version_major==5) + && ( + (handle->mgmd_version_minor==0 && handle->mgmd_version_build>=21) + ||(handle->mgmd_version_minor==1 && handle->mgmd_version_build>=12) + ||(handle->mgmd_version_minor>1) + ) + ) + || (handle->mgmd_version_major>5); + if(no_of_nodes < 0){ SET_ERROR(handle, NDB_MGM_RESTART_FAILED, "Restart requested of negative number of nodes"); @@ -942,7 +1077,7 @@ ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, const Properties *reply; const int timeout = handle->read_timeout; handle->read_timeout= 5*60*1000; // 5 minutes - reply = ndb_mgm_call(handle, restart_reply, "restart all", &args); + reply = ndb_mgm_call(handle, restart_reply_v1, "restart all", &args); handle->read_timeout= timeout; CHECK_REPLY(reply, -1); @@ -978,7 +1113,10 @@ ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, const Properties *reply; const int timeout = handle->read_timeout; handle->read_timeout= 5*60*1000; // 5 minutes - reply = ndb_mgm_call(handle, restart_reply, "restart node", &args); + if(use_v2) + reply = ndb_mgm_call(handle, restart_reply_v2, "restart node v2", &args); + else + reply = ndb_mgm_call(handle, restart_reply_v1, "restart node", &args); handle->read_timeout= timeout; if(reply != NULL) { BaseString result; @@ -989,20 +1127,16 @@ ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, return -1; } reply->get("restarted", &restarted); + if(use_v2) + reply->get("disconnect", (Uint32*)disconnect); + else + *disconnect= 0; delete reply; } return restarted; } -extern "C" -int -ndb_mgm_restart(NdbMgmHandle handle, int no_of_nodes, const int *node_list) -{ - SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart"); - return ndb_mgm_restart2(handle, no_of_nodes, node_list, 0, 0, 0); -} - static const char *clusterlog_severity_names[]= { "enabled", "debug", "info", "warning", "error", "critical", "alert" }; @@ -2361,4 +2495,56 @@ int ndb_mgm_end_session(NdbMgmHandle handle) DBUG_RETURN(0); } +extern "C" +int ndb_mgm_get_version(NdbMgmHandle handle, + int *major, int *minor, int *build, int len, char* str) +{ + DBUG_ENTER("ndb_mgm_get_version"); + CHECK_HANDLE(handle, 0); + CHECK_CONNECTED(handle, 0); + + Properties args; + + const ParserRow<ParserDummy> reply[]= { + MGM_CMD("version", NULL, ""), + MGM_ARG("id", Int, Mandatory, "ID"), + MGM_ARG("major", Int, Mandatory, "Major"), + MGM_ARG("minor", Int, Mandatory, "Minor"), + MGM_ARG("string", String, Mandatory, "String"), + MGM_END() + }; + + const Properties *prop; + prop = ndb_mgm_call(handle, reply, "get version", &args); + CHECK_REPLY(prop, 0); + + Uint32 id; + if(!prop->get("id",&id)){ + fprintf(handle->errstream, "Unable to get value\n"); + return 0; + } + *build= getBuild(id); + + if(!prop->get("major",(Uint32*)major)){ + fprintf(handle->errstream, "Unable to get value\n"); + return 0; + } + + if(!prop->get("minor",(Uint32*)minor)){ + fprintf(handle->errstream, "Unable to get value\n"); + return 0; + } + + BaseString result; + if(!prop->get("string", result)){ + fprintf(handle->errstream, "Unable to get value\n"); + return 0; + } + + strncpy(str, result.c_str(), len); + + delete prop; + DBUG_RETURN(1); +} + template class Vector<const ParserRow<ParserDummy>*>; diff --git a/ndb/src/mgmclient/CommandInterpreter.cpp b/ndb/src/mgmclient/CommandInterpreter.cpp index 14cd3fba42b..58b98671b14 100644 --- a/ndb/src/mgmclient/CommandInterpreter.cpp +++ b/ndb/src/mgmclient/CommandInterpreter.cpp @@ -162,6 +162,7 @@ private: NdbMgmHandle m_mgmsrv; NdbMgmHandle m_mgmsrv2; + const char *m_constr; bool m_connected; int m_verbose; int try_reconnect; @@ -390,22 +391,7 @@ convert(const char* s, int& val) { CommandInterpreter::CommandInterpreter(const char *_host,int verbose) : m_verbose(verbose) { - m_mgmsrv = ndb_mgm_create_handle(); - if(m_mgmsrv == NULL) { - ndbout_c("Cannot create handle to management server."); - exit(-1); - } - m_mgmsrv2 = ndb_mgm_create_handle(); - if(m_mgmsrv2 == NULL) { - ndbout_c("Cannot create 2:nd handle to management server."); - exit(-1); - } - if (ndb_mgm_set_connectstring(m_mgmsrv, _host)) - { - printError(); - exit(-1); - } - + m_constr= _host; m_connected= false; m_event_thread= 0; try_reconnect = 0; @@ -422,8 +408,6 @@ CommandInterpreter::CommandInterpreter(const char *_host,int verbose) CommandInterpreter::~CommandInterpreter() { disconnect(); - ndb_mgm_destroy_handle(&m_mgmsrv); - ndb_mgm_destroy_handle(&m_mgmsrv2); } static bool @@ -445,15 +429,14 @@ emptyString(const char* s) void CommandInterpreter::printError() { - if (ndb_mgm_check_connection(m_mgmsrv)) - { - m_connected= false; - disconnect(); - } ndbout_c("* %5d: %s", ndb_mgm_get_latest_error(m_mgmsrv), ndb_mgm_get_latest_error_msg(m_mgmsrv)); ndbout_c("* %s", ndb_mgm_get_latest_error_desc(m_mgmsrv)); + if (ndb_mgm_check_connection(m_mgmsrv)) + { + disconnect(); + } } //***************************************************************************** @@ -497,78 +480,97 @@ event_thread_run(void* m) } bool -CommandInterpreter::connect() +CommandInterpreter::connect() { DBUG_ENTER("CommandInterpreter::connect"); - if(!m_connected) + + if(m_connected) + DBUG_RETURN(m_connected); + + m_mgmsrv = ndb_mgm_create_handle(); + if(m_mgmsrv == NULL) { + ndbout_c("Cannot create handle to management server."); + exit(-1); + } + m_mgmsrv2 = ndb_mgm_create_handle(); + if(m_mgmsrv2 == NULL) { + ndbout_c("Cannot create 2:nd handle to management server."); + exit(-1); + } + + if (ndb_mgm_set_connectstring(m_mgmsrv, m_constr)) { - if(!ndb_mgm_connect(m_mgmsrv, try_reconnect-1, 5, 1)) + printError(); + exit(-1); + } + + if(ndb_mgm_connect(m_mgmsrv, try_reconnect-1, 5, 1)) + DBUG_RETURN(m_connected); // couldn't connect, always false + + const char *host= ndb_mgm_get_connected_host(m_mgmsrv); + unsigned port= ndb_mgm_get_connected_port(m_mgmsrv); + BaseString constr; + constr.assfmt("%s:%d",host,port); + if(!ndb_mgm_set_connectstring(m_mgmsrv2, constr.c_str()) && + !ndb_mgm_connect(m_mgmsrv2, try_reconnect-1, 5, 1)) + { + DBUG_PRINT("info",("2:ndb connected to Management Server ok at: %s:%d", + host, port)); + assert(m_event_thread == 0); + assert(do_event_thread == 0); + do_event_thread= 0; + m_event_thread = NdbThread_Create(event_thread_run, + (void**)&m_mgmsrv2, + 32768, + "CommandInterpreted_event_thread", + NDB_THREAD_PRIO_LOW); + if (m_event_thread != 0) { - const char *host= ndb_mgm_get_connected_host(m_mgmsrv); - unsigned port= ndb_mgm_get_connected_port(m_mgmsrv); - BaseString constr; - constr.assfmt("%s:%d",host,port); - if(!ndb_mgm_set_connectstring(m_mgmsrv2, constr.c_str()) && - !ndb_mgm_connect(m_mgmsrv2, try_reconnect-1, 5, 1)) - { - DBUG_PRINT("info",("2:ndb connected to Management Server ok at: %s:%d", - host, port)); - assert(m_event_thread == 0); - assert(do_event_thread == 0); - do_event_thread= 0; - m_event_thread = NdbThread_Create(event_thread_run, - (void**)&m_mgmsrv2, - 32768, - "CommandInterpreted_event_thread", - NDB_THREAD_PRIO_LOW); - if (m_event_thread != 0) - { - DBUG_PRINT("info",("Thread created ok, waiting for started...")); - int iter= 1000; // try for 30 seconds - while(do_event_thread == 0 && - iter-- > 0) - NdbSleep_MilliSleep(30); - } - if (m_event_thread == 0 || - do_event_thread == 0 || - do_event_thread == -1) - { - DBUG_PRINT("info",("Warning, event thread startup failed, " - "degraded printouts as result, errno=%d", - errno)); - printf("Warning, event thread startup failed, " - "degraded printouts as result, errno=%d\n", errno); - do_event_thread= 0; - if (m_event_thread) - { - void *res; - NdbThread_WaitFor(m_event_thread, &res); - NdbThread_Destroy(&m_event_thread); - } - ndb_mgm_disconnect(m_mgmsrv2); - } - } - else - { - DBUG_PRINT("warning", - ("Could not do 2:nd connect to mgmtserver for event listening")); - DBUG_PRINT("info", ("code: %d, msg: %s", - ndb_mgm_get_latest_error(m_mgmsrv2), - ndb_mgm_get_latest_error_msg(m_mgmsrv2))); - printf("Warning, event connect failed, degraded printouts as result\n"); - printf("code: %d, msg: %s\n", - ndb_mgm_get_latest_error(m_mgmsrv2), - ndb_mgm_get_latest_error_msg(m_mgmsrv2)); - } - m_connected= true; - DBUG_PRINT("info",("Connected to Management Server at: %s:%d", host, port)); - if (m_verbose) + DBUG_PRINT("info",("Thread created ok, waiting for started...")); + int iter= 1000; // try for 30 seconds + while(do_event_thread == 0 && + iter-- > 0) + NdbSleep_MilliSleep(30); + } + if (m_event_thread == 0 || + do_event_thread == 0 || + do_event_thread == -1) + { + DBUG_PRINT("info",("Warning, event thread startup failed, " + "degraded printouts as result, errno=%d", + errno)); + printf("Warning, event thread startup failed, " + "degraded printouts as result, errno=%d\n", errno); + do_event_thread= 0; + if (m_event_thread) { - printf("Connected to Management Server at: %s:%d\n", - host, port); + void *res; + NdbThread_WaitFor(m_event_thread, &res); + NdbThread_Destroy(&m_event_thread); } + ndb_mgm_disconnect(m_mgmsrv2); } } + else + { + DBUG_PRINT("warning", + ("Could not do 2:nd connect to mgmtserver for event listening")); + DBUG_PRINT("info", ("code: %d, msg: %s", + ndb_mgm_get_latest_error(m_mgmsrv2), + ndb_mgm_get_latest_error_msg(m_mgmsrv2))); + printf("Warning, event connect failed, degraded printouts as result\n"); + printf("code: %d, msg: %s\n", + ndb_mgm_get_latest_error(m_mgmsrv2), + ndb_mgm_get_latest_error_msg(m_mgmsrv2)); + } + m_connected= true; + DBUG_PRINT("info",("Connected to Management Server at: %s:%d", host, port)); + if (m_verbose) + { + printf("Connected to Management Server at: %s:%d\n", + host, port); + } + DBUG_RETURN(m_connected); } @@ -576,20 +578,18 @@ bool CommandInterpreter::disconnect() { DBUG_ENTER("CommandInterpreter::disconnect"); + if (m_event_thread) { void *res; do_event_thread= 0; NdbThread_WaitFor(m_event_thread, &res); NdbThread_Destroy(&m_event_thread); m_event_thread= 0; - ndb_mgm_disconnect(m_mgmsrv2); + ndb_mgm_destroy_handle(&m_mgmsrv2); } if (m_connected) { - if (ndb_mgm_disconnect(m_mgmsrv) == -1) { - ndbout_c("Could not disconnect from management server"); - printError(); - } + ndb_mgm_destroy_handle(&m_mgmsrv); m_connected= false; } DBUG_RETURN(true); @@ -1057,7 +1057,8 @@ CommandInterpreter::executeShutdown(char* parameters) NdbAutoPtr<char> ap1((char*)state); int result = 0; - result = ndb_mgm_stop(m_mgmsrv, 0, 0); + int need_disconnect; + result = ndb_mgm_stop3(m_mgmsrv, -1, 0, 0, &need_disconnect); if (result < 0) { ndbout << "Shutdown of NDB Cluster node(s) failed." << endl; printError(); @@ -1066,28 +1067,11 @@ CommandInterpreter::executeShutdown(char* parameters) ndbout << result << " NDB Cluster node(s) have shutdown." << endl; - int mgm_id= 0; - mgm_id= ndb_mgm_get_mgmd_nodeid(m_mgmsrv); - if (mgm_id == 0) - { - ndbout << "Unable to locate management server, " - << "shutdown manually with <id> STOP" + if(need_disconnect) { + ndbout << "Disconnecting to allow management server to shutdown." << endl; - return 1; - } - - result = ndb_mgm_stop(m_mgmsrv, 1, &mgm_id); - if (result <= 0) { - ndbout << "Shutdown of NDB Cluster management server failed." << endl; - printError(); - if (result == 0) - return 1; - return result; + disconnect(); } - - m_connected= false; - disconnect(); - ndbout << "NDB Cluster management server shutdown." << endl; return 0; } @@ -1311,12 +1295,7 @@ CommandInterpreter::executeConnect(char* parameters) { disconnect(); if (!emptyString(parameters)) { - if (ndb_mgm_set_connectstring(m_mgmsrv, - BaseString(parameters).trim().c_str())) - { - printError(); - return; - } + m_constr= BaseString(parameters).trim().c_str(); } connect(); } @@ -1481,6 +1460,7 @@ CommandInterpreter::executeStop(Vector<BaseString> &command_list, unsigned command_pos, int *node_ids, int no_of_nodes) { + int need_disconnect; int abort= 0; for (; command_pos < command_list.size(); command_pos++) { @@ -1495,7 +1475,8 @@ CommandInterpreter::executeStop(Vector<BaseString> &command_list, return; } - int result= ndb_mgm_stop2(m_mgmsrv, no_of_nodes, node_ids, abort); + int result= ndb_mgm_stop3(m_mgmsrv, no_of_nodes, node_ids, abort, + &need_disconnect); if (result < 0) { ndbout_c("Shutdown failed."); @@ -1509,10 +1490,17 @@ CommandInterpreter::executeStop(Vector<BaseString> &command_list, { ndbout << "Node"; for (int i= 0; i < no_of_nodes; i++) - ndbout << " " << node_ids[i]; + ndbout << " " << node_ids[i]; ndbout_c(" has shutdown."); } } + + if(need_disconnect) + { + ndbout << "Disconnecting to allow Management Server to shutdown" << endl; + disconnect(); + } + } void @@ -1603,6 +1591,7 @@ CommandInterpreter::executeRestart(Vector<BaseString> &command_list, int nostart= 0; int initialstart= 0; int abort= 0; + int need_disconnect= 0; for (; command_pos < command_list.size(); command_pos++) { @@ -1627,9 +1616,9 @@ CommandInterpreter::executeRestart(Vector<BaseString> &command_list, return; } - result= ndb_mgm_restart2(m_mgmsrv, no_of_nodes, node_ids, - initialstart, nostart, abort); - + result= ndb_mgm_restart3(m_mgmsrv, no_of_nodes, node_ids, + initialstart, nostart, abort, &need_disconnect); + if (result <= 0) { ndbout_c("Restart failed."); printError(); @@ -1645,6 +1634,8 @@ CommandInterpreter::executeRestart(Vector<BaseString> &command_list, ndbout << " " << node_ids[i]; ndbout_c(" is being restarted"); } + if(need_disconnect) + disconnect(); } } diff --git a/ndb/src/mgmsrv/MgmtSrvr.cpp b/ndb/src/mgmsrv/MgmtSrvr.cpp index f94042b2abd..2505515c086 100644 --- a/ndb/src/mgmsrv/MgmtSrvr.cpp +++ b/ndb/src/mgmsrv/MgmtSrvr.cpp @@ -60,9 +60,6 @@ #include <SignalSender.hpp> -extern bool g_StopServer; -extern bool g_RestartServer; - //#define MGM_SRV_DEBUG #ifdef MGM_SRV_DEBUG #define DEBUG(x) do ndbout << x << endl; while(0) @@ -942,6 +939,13 @@ int MgmtSrvr::sendStopMgmd(NodeId nodeId, * client connection to that mgmd and stop it that way. * This allows us to stop mgm servers when there isn't any real * distributed communication up. + * + * node_ids.size()==0 means to stop all DB nodes. + * MGM nodes will *NOT* be stopped. + * + * If we work out we should be stopping or restarting ourselves, + * we return <0 in stopSelf for restart, >0 for stop + * and 0 for do nothing. */ int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids, @@ -951,7 +955,8 @@ int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids, bool stop, bool restart, bool nostart, - bool initialStart) + bool initialStart, + int* stopSelf) { int error = 0; DBUG_ENTER("MgmtSrvr::sendSTOP_REQ"); @@ -1000,12 +1005,13 @@ int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids, NodeId nodeId= 0; int use_master_node= 0; int do_send= 0; - int do_stop_self= 0; + *stopSelf= 0; NdbNodeBitmask nodes_to_stop; { for (unsigned i= 0; i < node_ids.size(); i++) { nodeId= node_ids[i]; + ndbout << "asked to stop " << nodeId << endl; if (getNodeType(nodeId) != NDB_MGM_NODE_TYPE_MGM) nodes_to_stop.set(nodeId); else if (nodeId != getOwnNodeId()) @@ -1016,7 +1022,11 @@ int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids, stoppedNodes.set(nodeId); } else - do_stop_self= 1;; + { + ndbout << "which is me" << endl; + *stopSelf= (restart)? -1 : 1; + stoppedNodes.set(nodeId); + } } } int no_of_nodes_to_stop= nodes_to_stop.count(); @@ -1049,14 +1059,6 @@ int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids, nodes.set(nodeId); } } - nodeId= 0; - while(getNextNodeId(&nodeId, NDB_MGM_NODE_TYPE_MGM)) - { - if(nodeId==getOwnNodeId()) - continue; - if(sendStopMgmd(nodeId, abort, stop, restart, nostart, initialStart)==0) - stoppedNodes.set(nodeId); - } } // now wait for the replies @@ -1148,11 +1150,9 @@ int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids, DBUG_RETURN(SEND_OR_RECEIVE_FAILED); } } - if (!error && do_stop_self) + if (error && *stopSelf) { - if (restart) - g_RestartServer= true; - g_StopServer= true; + *stopSelf= 0; } DBUG_RETURN(error); } @@ -1162,7 +1162,7 @@ int MgmtSrvr::sendSTOP_REQ(const Vector<NodeId> &node_ids, */ int MgmtSrvr::stopNodes(const Vector<NodeId> &node_ids, - int *stopCount, bool abort) + int *stopCount, bool abort, int* stopSelf) { if (!abort) { @@ -1184,20 +1184,46 @@ int MgmtSrvr::stopNodes(const Vector<NodeId> &node_ids, false, false, false, - false); + false, + stopSelf); if (stopCount) *stopCount= nodes.count(); return ret; } +int MgmtSrvr::shutdownMGM(int *stopCount, bool abort, int *stopSelf) +{ + NodeId nodeId = 0; + int error; + + while(getNextNodeId(&nodeId, NDB_MGM_NODE_TYPE_MGM)) + { + if(nodeId==getOwnNodeId()) + continue; + error= sendStopMgmd(nodeId, abort, true, false, + false, false); + if (error == 0) + *stopCount++; + } + + *stopSelf= 1; + *stopCount++; + + return 0; +} + /* - * Perform system shutdown + * Perform DB nodes shutdown. + * MGM servers are left in their current state */ -int MgmtSrvr::stop(int * stopCount, bool abort) +int MgmtSrvr::shutdownDB(int * stopCount, bool abort) { NodeBitmask nodes; Vector<NodeId> node_ids; + + int tmp; + int ret = sendSTOP_REQ(node_ids, nodes, 0, @@ -1205,7 +1231,8 @@ int MgmtSrvr::stop(int * stopCount, bool abort) true, false, false, - false); + false, + &tmp); if (stopCount) *stopCount = nodes.count(); return ret; @@ -1230,6 +1257,7 @@ int MgmtSrvr::enterSingleUser(int * stopCount, Uint32 singleUserNodeId) } NodeBitmask nodes; Vector<NodeId> node_ids; + int stopSelf; int ret = sendSTOP_REQ(node_ids, nodes, singleUserNodeId, @@ -1237,7 +1265,8 @@ int MgmtSrvr::enterSingleUser(int * stopCount, Uint32 singleUserNodeId) false, false, false, - false); + false, + &stopSelf); if (stopCount) *stopCount = nodes.count(); return ret; @@ -1249,7 +1278,8 @@ int MgmtSrvr::enterSingleUser(int * stopCount, Uint32 singleUserNodeId) int MgmtSrvr::restartNodes(const Vector<NodeId> &node_ids, int * stopCount, bool nostart, - bool initialStart, bool abort) + bool initialStart, bool abort, + int *stopSelf) { NodeBitmask nodes; int ret= sendSTOP_REQ(node_ids, @@ -1259,7 +1289,8 @@ int MgmtSrvr::restartNodes(const Vector<NodeId> &node_ids, false, true, true, - initialStart); + initialStart, + stopSelf); if (ret) return ret; @@ -1302,14 +1333,16 @@ int MgmtSrvr::restartNodes(const Vector<NodeId> &node_ids, } /* - * Perform system restart + * Perform restart of all DB nodes */ -int MgmtSrvr::restart(bool nostart, bool initialStart, - bool abort, int * stopCount ) +int MgmtSrvr::restartDB(bool nostart, bool initialStart, + bool abort, int * stopCount) { NodeBitmask nodes; Vector<NodeId> node_ids; + int tmp; + int ret = sendSTOP_REQ(node_ids, nodes, 0, @@ -1317,7 +1350,8 @@ int MgmtSrvr::restart(bool nostart, bool initialStart, true, true, true, - initialStart); + initialStart, + &tmp); if (ret) return ret; diff --git a/ndb/src/mgmsrv/MgmtSrvr.hpp b/ndb/src/mgmsrv/MgmtSrvr.hpp index 57d51192498..7811cf0e5d1 100644 --- a/ndb/src/mgmsrv/MgmtSrvr.hpp +++ b/ndb/src/mgmsrv/MgmtSrvr.hpp @@ -255,12 +255,15 @@ public: * @param processId: Id of the DB process to stop * @return 0 if succeeded, otherwise: as stated above, plus: */ - int stopNodes(const Vector<NodeId> &node_ids, int *stopCount, bool abort); + int stopNodes(const Vector<NodeId> &node_ids, int *stopCount, bool abort, + int *stopSelf); + + int shutdownMGM(int *stopCount, bool abort, int *stopSelf); /** - * Stop the system + * shutdown the DB nodes */ - int stop(int * cnt = 0, bool abort = false); + int shutdownDB(int * cnt = 0, bool abort = false); /** * print version info about a node @@ -294,14 +297,14 @@ public: */ int restartNodes(const Vector<NodeId> &node_ids, int *stopCount, bool nostart, - bool initialStart, bool abort); + bool initialStart, bool abort, int *stopSelf); /** - * Restart the system + * Restart all DB nodes */ - int restart(bool nostart, bool initialStart, - bool abort = false, - int * stopCount = 0); + int restartDB(bool nostart, bool initialStart, + bool abort = false, + int * stopCount = 0); struct BackupEvent { enum Event { @@ -505,7 +508,8 @@ private: bool stop, bool restart, bool nostart, - bool initialStart); + bool initialStart, + int *stopSelf); /** * Check if it is possible to send a signal to a (DB) process diff --git a/ndb/src/mgmsrv/Services.cpp b/ndb/src/mgmsrv/Services.cpp index 83e289cb409..0524aba4c32 100644 --- a/ndb/src/mgmsrv/Services.cpp +++ b/ndb/src/mgmsrv/Services.cpp @@ -35,6 +35,7 @@ #include <base64.h> extern bool g_StopServer; +extern bool g_RestartServer; extern EventLogger g_eventLogger; static const unsigned int MAX_READ_TIMEOUT = 1000 ; @@ -146,7 +147,13 @@ ParserRow<MgmApiSession> commands[] = { MGM_CMD("get info clusterlog", &MgmApiSession::getInfoClusterLog, ""), - MGM_CMD("restart node", &MgmApiSession::restart, ""), + MGM_CMD("restart node", &MgmApiSession::restart_v1, ""), + MGM_ARG("node", String, Mandatory, "Nodes to restart"), + MGM_ARG("initialstart", Int, Optional, "Initial start"), + MGM_ARG("nostart", Int, Optional, "No start"), + MGM_ARG("abort", Int, Optional, "Abort"), + + MGM_CMD("restart node v2", &MgmApiSession::restart_v2, ""), MGM_ARG("node", String, Mandatory, "Nodes to restart"), MGM_ARG("initialstart", Int, Optional, "Initial start"), MGM_ARG("nostart", Int, Optional, "No start"), @@ -187,13 +194,18 @@ ParserRow<MgmApiSession> commands[] = { MGM_CMD("abort backup", &MgmApiSession::abortBackup, ""), MGM_ARG("id", Int, Mandatory, "Backup id"), - MGM_CMD("stop", &MgmApiSession::stop, ""), + MGM_CMD("stop", &MgmApiSession::stop_v1, ""), + MGM_ARG("node", String, Mandatory, "Node"), + MGM_ARG("abort", Int, Mandatory, "Node"), + + MGM_CMD("stop v2", &MgmApiSession::stop_v2, ""), MGM_ARG("node", String, Mandatory, "Node"), MGM_ARG("abort", Int, Mandatory, "Node"), MGM_CMD("stop all", &MgmApiSession::stopAll, ""), MGM_ARG("abort", Int, Mandatory, "Node"), - + MGM_ARG("stop", String, Optional, "MGM/DB or both"), + MGM_CMD("enter single user", &MgmApiSession::enterSingleUser, ""), MGM_ARG("nodeId", Int, Mandatory, "Node"), @@ -278,6 +290,7 @@ MgmApiSession::MgmApiSession(class MgmtSrvr & mgm, NDB_SOCKET_TYPE sock) m_output = new SocketOutputStream(sock); m_parser = new Parser_t(commands, *m_input, true, true, true); m_allocated_resources= new MgmtSrvr::Allocated_resources(m_mgmsrv); + m_stopSelf= 0; DBUG_VOID_RETURN; } @@ -297,6 +310,10 @@ MgmApiSession::~MgmApiSession() NDB_CLOSE_SOCKET(m_socket); m_socket= NDB_INVALID_SOCKET; } + if(m_stopSelf < 0) + g_RestartServer= true; + if(m_stopSelf) + g_StopServer= true; DBUG_VOID_RETURN; } @@ -885,8 +902,19 @@ MgmApiSession::stopSignalLog(Parser<MgmApiSession>::Context &, } void -MgmApiSession::restart(Parser<MgmApiSession>::Context &, +MgmApiSession::restart_v1(Parser<MgmApiSession>::Context &, + Properties const &args) { + restart(args,1); +} + +void +MgmApiSession::restart_v2(Parser<MgmApiSession>::Context &, Properties const &args) { + restart(args,2); +} + +void +MgmApiSession::restart(Properties const &args, int version) { Uint32 nostart = 0, initialstart = 0, @@ -911,7 +939,8 @@ MgmApiSession::restart(Parser<MgmApiSession>::Context &, &restarted, nostart != 0, initialstart != 0, - abort != 0); + abort != 0, + &m_stopSelf); m_output->println("restart reply"); if(result != 0){ @@ -919,6 +948,8 @@ MgmApiSession::restart(Parser<MgmApiSession>::Context &, } else m_output->println("result: Ok"); m_output->println("restarted: %d", restarted); + if(version>1) + m_output->println("disconnect: %d", (m_stopSelf)?1:0); m_output->println(""); } @@ -935,7 +966,7 @@ MgmApiSession::restartAll(Parser<MgmApiSession>::Context &, args.get("nostart", &nostart); int count = 0; - int result = m_mgmsrv.restart(nostart, initialstart, abort, &count); + int result = m_mgmsrv.restartDB(nostart, initialstart, abort, &count); m_output->println("restart reply"); if(result != 0) @@ -1028,8 +1059,19 @@ MgmApiSession::getInfoClusterLog(Parser<MgmApiSession>::Context &, } void -MgmApiSession::stop(Parser<MgmApiSession>::Context &, - Properties const &args) { +MgmApiSession::stop_v1(Parser<MgmApiSession>::Context &, + Properties const &args) { + stop(args,1); +} + +void +MgmApiSession::stop_v2(Parser<MgmApiSession>::Context &, + Properties const &args) { + stop(args,2); +} + +void +MgmApiSession::stop(Properties const &args, int version) { Uint32 abort; char *nodes_str; Vector<NodeId> nodes; @@ -1054,7 +1096,7 @@ MgmApiSession::stop(Parser<MgmApiSession>::Context &, int stopped= 0; int result= 0; if (nodes.size()) - result= m_mgmsrv.stopNodes(nodes, &stopped, abort != 0); + result= m_mgmsrv.stopNodes(nodes, &stopped, abort != 0, &m_stopSelf); m_output->println("stop reply"); if(result != 0) @@ -1062,25 +1104,41 @@ MgmApiSession::stop(Parser<MgmApiSession>::Context &, else m_output->println("result: Ok"); m_output->println("stopped: %d", stopped); + if(version>1) + m_output->println("disconnect: %d", (m_stopSelf)?1:0); m_output->println(""); } - void MgmApiSession::stopAll(Parser<MgmApiSession>::Context &, - Properties const &args) { - int stopped = 0; + Properties const &args) { + int stopped[2] = {0,0}; Uint32 abort; args.get("abort", &abort); - int result = m_mgmsrv.stop(&stopped, abort != 0); + BaseString stop; + const char* tostop= "db"; + int ver=1; + if (args.get("stop", stop)) + { + tostop= stop.c_str(); + ver= 2; + } + + int result= 0; + if(strstr(tostop,"db")) + result= m_mgmsrv.shutdownDB(&stopped[0], abort != 0); + if(!result && strstr(tostop,"mgm")) + result= m_mgmsrv.shutdownMGM(&stopped[1], abort!=0, &m_stopSelf); m_output->println("stop reply"); if(result != 0) m_output->println("result: %s", get_error_text(result)); else m_output->println("result: Ok"); - m_output->println("stopped: %d", stopped); + m_output->println("stopped: %d", stopped[0]+stopped[1]); + if(ver >1) + m_output->println("disconnect: %d", (m_stopSelf)?1:0); m_output->println(""); } diff --git a/ndb/src/mgmsrv/Services.hpp b/ndb/src/mgmsrv/Services.hpp index 975202b96df..6e0bb701d7b 100644 --- a/ndb/src/mgmsrv/Services.hpp +++ b/ndb/src/mgmsrv/Services.hpp @@ -41,6 +41,7 @@ private: Parser_t *m_parser; MgmtSrvr::Allocated_resources *m_allocated_resources; char m_err_str[1024]; + int m_stopSelf; // -1 is restart, 0 do nothing, 1 stop void getConfig_common(Parser_t::Context &ctx, const class Properties &args, @@ -63,7 +64,9 @@ public: void getVersion(Parser_t::Context &ctx, const class Properties &args); void getStatus(Parser_t::Context &ctx, const class Properties &args); void getInfoClusterLog(Parser_t::Context &ctx, const class Properties &args); - void restart(Parser_t::Context &ctx, const class Properties &args); + void restart(const class Properties &args, int version); + void restart_v1(Parser_t::Context &ctx, const class Properties &args); + void restart_v2(Parser_t::Context &ctx, const class Properties &args); void restartAll(Parser_t::Context &ctx, const class Properties &args); void insertError(Parser_t::Context &ctx, const class Properties &args); void setTrace(Parser_t::Context &ctx, const class Properties &args); @@ -75,7 +78,9 @@ public: void abortBackup(Parser_t::Context &ctx, const class Properties &args); void enterSingleUser(Parser_t::Context &ctx, const class Properties &args); void exitSingleUser(Parser_t::Context &ctx, const class Properties &args); - void stop(Parser_t::Context &ctx, const class Properties &args); + void stop_v1(Parser_t::Context &ctx, const class Properties &args); + void stop_v2(Parser_t::Context &ctx, const class Properties &args); + void stop(const class Properties &args, int version); void stopAll(Parser_t::Context &ctx, const class Properties &args); void start(Parser_t::Context &ctx, const class Properties &args); void startAll(Parser_t::Context &ctx, const class Properties &args); |