homework-jianmu/src/dnode/src/dnodeVnodes.c

303 lines
9.7 KiB
C

/*
* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#define _DEFAULT_SOURCE
#include "os.h"
#include "ttimer.h"
#include "dnodeEps.h"
#include "dnodeCfg.h"
#include "dnodeMInfos.h"
#include "dnodeVnodes.h"
typedef struct {
pthread_t thread;
int32_t threadIndex;
int32_t failed;
int32_t opened;
int32_t vnodeNum;
int32_t * vnodeList;
} SOpenVnodeThread;
extern void * tsDnodeTmr;
static void * tsStatusTimer = NULL;
static uint32_t tsRebootTime = 0;
static int32_t tsOpenVnodes = 0;
static int32_t tsTotalVnodes = 0;
static void dnodeSendStatusMsg(void *handle, void *tmrId);
static void dnodeProcessStatusRsp(SRpcMsg *pMsg);
int32_t dnodeInitStatusTimer() {
dnodeAddClientRspHandle(TSDB_MSG_TYPE_DM_STATUS_RSP, dnodeProcessStatusRsp);
tsRebootTime = taosGetTimestampSec();
taosTmrReset(dnodeSendStatusMsg, 500, NULL, tsDnodeTmr, &tsStatusTimer);
dInfo("dnode status timer is initialized");
return TSDB_CODE_SUCCESS;
}
void dnodeCleanupStatusTimer() {
if (tsStatusTimer != NULL) {
taosTmrStopA(&tsStatusTimer);
tsStatusTimer = NULL;
}
}
static int32_t dnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes) {
DIR *dir = opendir(tsVnodeDir);
if (dir == NULL) return TSDB_CODE_DND_NO_WRITE_ACCESS;
*numOfVnodes = 0;
struct dirent *de = NULL;
while ((de = readdir(dir)) != NULL) {
if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) continue;
if (de->d_type & DT_DIR) {
if (strncmp("vnode", de->d_name, 5) != 0) continue;
int32_t vnode = atoi(de->d_name + 5);
if (vnode == 0) continue;
(*numOfVnodes)++;
if (*numOfVnodes >= TSDB_MAX_VNODES) {
dError("vgId:%d, too many vnode directory in disk, exist:%d max:%d", vnode, *numOfVnodes, TSDB_MAX_VNODES);
closedir(dir);
return TSDB_CODE_DND_TOO_MANY_VNODES;
} else {
vnodeList[*numOfVnodes - 1] = vnode;
}
}
}
closedir(dir);
return TSDB_CODE_SUCCESS;
}
static void *dnodeOpenVnode(void *param) {
SOpenVnodeThread *pThread = param;
char stepDesc[TSDB_STEP_DESC_LEN] = {0};
dDebug("thread:%d, start to open %d vnodes", pThread->threadIndex, pThread->vnodeNum);
for (int32_t v = 0; v < pThread->vnodeNum; ++v) {
int32_t vgId = pThread->vnodeList[v];
snprintf(stepDesc, TSDB_STEP_DESC_LEN, "vgId:%d, start to restore, %d of %d have been opened", vgId, tsOpenVnodes, tsTotalVnodes);
dnodeReportStep("open-vnodes", stepDesc, 0);
if (vnodeOpen(vgId) < 0) {
dError("vgId:%d, failed to open vnode by thread:%d", vgId, pThread->threadIndex);
pThread->failed++;
} else {
dDebug("vgId:%d, is opened by thread:%d", vgId, pThread->threadIndex);
pThread->opened++;
}
atomic_add_fetch_32(&tsOpenVnodes, 1);
}
dDebug("thread:%d, total vnodes:%d, opened:%d failed:%d", pThread->threadIndex, pThread->vnodeNum, pThread->opened,
pThread->failed);
return NULL;
}
int32_t dnodeInitVnodes() {
int32_t vnodeList[TSDB_MAX_VNODES] = {0};
int32_t numOfVnodes = 0;
int32_t status = dnodeGetVnodeList(vnodeList, &numOfVnodes);
tsTotalVnodes = numOfVnodes;
if (status != TSDB_CODE_SUCCESS) {
dInfo("get dnode list failed");
return status;
}
int32_t threadNum = tsNumOfCores;
int32_t vnodesPerThread = numOfVnodes / threadNum + 1;
SOpenVnodeThread *threads = calloc(threadNum, sizeof(SOpenVnodeThread));
for (int32_t t = 0; t < threadNum; ++t) {
threads[t].threadIndex = t;
threads[t].vnodeList = calloc(vnodesPerThread, sizeof(int32_t));
}
for (int32_t v = 0; v < numOfVnodes; ++v) {
int32_t t = v % threadNum;
SOpenVnodeThread *pThread = &threads[t];
pThread->vnodeList[pThread->vnodeNum++] = vnodeList[v];
}
dInfo("start %d threads to open %d vnodes", threadNum, numOfVnodes);
for (int32_t t = 0; t < threadNum; ++t) {
SOpenVnodeThread *pThread = &threads[t];
if (pThread->vnodeNum == 0) continue;
pthread_attr_t thAttr;
pthread_attr_init(&thAttr);
pthread_attr_setdetachstate(&thAttr, PTHREAD_CREATE_JOINABLE);
if (pthread_create(&pThread->thread, &thAttr, dnodeOpenVnode, pThread) != 0) {
dError("thread:%d, failed to create thread to open vnode, reason:%s", pThread->threadIndex, strerror(errno));
}
pthread_attr_destroy(&thAttr);
}
int32_t openVnodes = 0;
int32_t failedVnodes = 0;
for (int32_t t = 0; t < threadNum; ++t) {
SOpenVnodeThread *pThread = &threads[t];
if (pThread->vnodeNum > 0 && taosCheckPthreadValid(pThread->thread)) {
pthread_join(pThread->thread, NULL);
}
openVnodes += pThread->opened;
failedVnodes += pThread->failed;
free(pThread->vnodeList);
}
free(threads);
dInfo("there are total vnodes:%d, opened:%d", numOfVnodes, openVnodes);
if (failedVnodes != 0) {
dError("there are total vnodes:%d, failed:%d", numOfVnodes, failedVnodes);
return -1;
}
return TSDB_CODE_SUCCESS;
}
void dnodeCleanupVnodes() {
int32_t vnodeList[TSDB_MAX_VNODES]= {0};
int32_t numOfVnodes = 0;
int32_t status;
status = vnodeGetVnodeList(vnodeList, &numOfVnodes);
if (status != TSDB_CODE_SUCCESS) {
dInfo("get dnode list failed");
return;
}
for (int32_t i = 0; i < numOfVnodes; ++i) {
vnodeClose(vnodeList[i]);
}
dInfo("total vnodes:%d are all closed", numOfVnodes);
}
static void dnodeProcessStatusRsp(SRpcMsg *pMsg) {
if (pMsg->code != TSDB_CODE_SUCCESS) {
dError("status rsp is received, error:%s", tstrerror(pMsg->code));
if (pMsg->code == TSDB_CODE_MND_DNODE_NOT_EXIST) {
char clusterId[TSDB_CLUSTER_ID_LEN];
dnodeGetClusterId(clusterId);
if (clusterId[0] != '\0') {
dnodeSetDropped();
dError("exit zombie dropped dnode");
exit(EXIT_FAILURE);
}
}
taosTmrReset(dnodeSendStatusMsg, tsStatusInterval * 1000, NULL, tsDnodeTmr, &tsStatusTimer);
return;
}
SStatusRsp *pStatusRsp = pMsg->pCont;
SMInfos *minfos = &pStatusRsp->mnodes;
dnodeUpdateMInfos(minfos);
SDnodeCfg *pCfg = &pStatusRsp->dnodeCfg;
pCfg->numOfVnodes = htonl(pCfg->numOfVnodes);
pCfg->moduleStatus = htonl(pCfg->moduleStatus);
pCfg->dnodeId = htonl(pCfg->dnodeId);
dnodeUpdateCfg(pCfg);
vnodeSetAccess(pStatusRsp->vgAccess, pCfg->numOfVnodes);
SDnodeEps *pEps = (SDnodeEps *)((char *)pStatusRsp->vgAccess + pCfg->numOfVnodes * sizeof(SVgroupAccess));
dnodeUpdateEps(pEps);
taosTmrReset(dnodeSendStatusMsg, tsStatusInterval * 1000, NULL, tsDnodeTmr, &tsStatusTimer);
}
static void dnodeSendStatusMsg(void *handle, void *tmrId) {
if (tsDnodeTmr == NULL) {
dError("dnode timer is already released");
return;
}
if (tsStatusTimer == NULL) {
taosTmrReset(dnodeSendStatusMsg, tsStatusInterval * 1000, NULL, tsDnodeTmr, &tsStatusTimer);
dError("failed to start status timer");
return;
}
int32_t contLen = sizeof(SStatusMsg) + TSDB_MAX_VNODES * sizeof(SVnodeLoad);
SStatusMsg *pStatus = rpcMallocCont(contLen);
if (pStatus == NULL) {
taosTmrReset(dnodeSendStatusMsg, tsStatusInterval * 1000, NULL, tsDnodeTmr, &tsStatusTimer);
dError("failed to malloc status message");
return;
}
dnodeGetCfg(&pStatus->dnodeId, pStatus->clusterId);
pStatus->dnodeId = htonl(dnodeGetDnodeId());
pStatus->version = htonl(tsVersion);
pStatus->lastReboot = htonl(tsRebootTime);
pStatus->numOfCores = htons((uint16_t) tsNumOfCores);
pStatus->diskAvailable = tsAvailDataDirGB;
pStatus->alternativeRole = tsAlternativeRole;
tstrncpy(pStatus->dnodeEp, tsLocalEp, TSDB_EP_LEN);
// fill cluster cfg parameters
pStatus->clusterCfg.numOfMnodes = htonl(tsNumOfMnodes);
pStatus->clusterCfg.mnodeEqualVnodeNum = htonl(tsMnodeEqualVnodeNum);
pStatus->clusterCfg.offlineThreshold = htonl(tsOfflineThreshold);
pStatus->clusterCfg.statusInterval = htonl(tsStatusInterval);
pStatus->clusterCfg.maxtablesPerVnode = htonl(tsMaxTablePerVnode);
pStatus->clusterCfg.maxVgroupsPerDb = htonl(tsMaxVgroupsPerDb);
tstrncpy(pStatus->clusterCfg.arbitrator, tsArbitrator, TSDB_EP_LEN);
tstrncpy(pStatus->clusterCfg.timezone, tsTimezone, 64);
pStatus->clusterCfg.checkTime = 0;
char timestr[32] = "1970-01-01 00:00:00.00";
(void)taosParseTime(timestr, &pStatus->clusterCfg.checkTime, (int32_t)strlen(timestr), TSDB_TIME_PRECISION_MILLI, 0);
tstrncpy(pStatus->clusterCfg.locale, tsLocale, TSDB_LOCALE_LEN);
tstrncpy(pStatus->clusterCfg.charset, tsCharset, TSDB_LOCALE_LEN);
pStatus->clusterCfg.enableBalance = tsEnableBalance;
pStatus->clusterCfg.flowCtrl = tsEnableFlowCtrl;
pStatus->clusterCfg.slaveQuery = tsEnableSlaveQuery;
pStatus->clusterCfg.adjustMaster = tsEnableAdjustMaster;
vnodeBuildStatusMsg(pStatus);
contLen = sizeof(SStatusMsg) + pStatus->openVnodes * sizeof(SVnodeLoad);
pStatus->openVnodes = htons(pStatus->openVnodes);
SRpcMsg rpcMsg = {
.pCont = pStatus,
.contLen = contLen,
.msgType = TSDB_MSG_TYPE_DM_STATUS
};
SRpcEpSet epSet;
dnodeGetEpSetForPeer(&epSet);
dnodeSendMsgToDnode(&epSet, &rpcMsg);
}
void dnodeSendStatusMsgToMnode() {
if (tsDnodeTmr != NULL && tsStatusTimer != NULL) {
dInfo("force send status msg to mnode");
taosTmrReset(dnodeSendStatusMsg, 3, NULL, tsDnodeTmr, &tsStatusTimer);
}
}