未验证 提交 a9161a5c 编写于 作者: C Cary Xu 提交者: GitHub

Merge pull request #14471 from taosdata/feature/TD-11274-3.0

refactor: rsma commit and recovery
......@@ -62,12 +62,10 @@ struct STSmaStat {
struct SRSmaStat {
SSma *pSma;
int64_t refId; // shared by persistence/fetch tasks
void *tmrHandle; // for persistence task
tmr_h tmrId; // for persistence task
int32_t tmrSeconds; // for persistence task
int8_t triggerStat; // for persistence task
int8_t runningStat; // for persistence task
int64_t refId; // shared by fetch tasks
void *tmrHandle; // shared by fetch tasks
int8_t triggerStat; // shared by fetch tasks
int8_t runningStat; // for persistence task
SHashObj *rsmaInfoHash; // key: stbUid, value: SRSmaInfo;
};
......@@ -82,7 +80,6 @@ struct SSmaStat {
#define SMA_TSMA_STAT(s) (&(s)->tsmaStat)
#define SMA_RSMA_STAT(s) (&(s)->rsmaStat)
#define RSMA_INFO_HASH(r) ((r)->rsmaInfoHash)
#define RSMA_TMR_ID(r) ((r)->tmrId)
#define RSMA_TMR_HANDLE(r) ((r)->tmrHandle)
#define RSMA_TRIGGER_STAT(r) (&(r)->triggerStat)
#define RSMA_RUNNING_STAT(r) (&(r)->runningStat)
......@@ -185,9 +182,11 @@ static FORCE_INLINE void tdSmaStatSetDropped(STSmaStat *pTStat) {
static int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType);
void *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType);
void *tdFreeRSmaInfo(SRSmaInfo *pInfo);
int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat);
int32_t tdProcessRSmaCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, const char *tbName);
int32_t tdProcessRSmaRestoreImpl(SSma *pSma);
int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg);
int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg);
int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days);
......@@ -244,8 +243,8 @@ void tdUpdateTFileMagic(STFile *pTFile, void *pCksm);
void tdCloseTFile(STFile *pTFile);
void tdDestroyTFile(STFile *pTFile);
void tdGetVndFileName(int32_t vgId, const char *dname, const char *fname, int64_t version, char *outputName);
void tdGetVndDirName(int32_t vgId, const char *dname, char *outputName);
void tdGetVndFileName(int32_t vgId, const char *pdname, const char *dname, const char *fname, int64_t version, char *outputName);
void tdGetVndDirName(int32_t vgId,const char *pdname, const char *dname, bool endWithSep, char *outputName);
#ifdef __cplusplus
}
......
......@@ -163,8 +163,8 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pSchema, bool
// sma
int32_t smaOpen(SVnode* pVnode);
int32_t smaCloseEnv(SSma* pSma);
int32_t smaCloseEx(SSma* pSma);
int32_t smaClose(SSma* pSma);
int32_t smaBegin(SSma* pSma);
int32_t smaPreCommit(SSma* pSma);
int32_t smaCommit(SSma* pSma);
int32_t smaPostCommit(SSma* pSma);
......
......@@ -43,13 +43,48 @@ int32_t smaCommit(SSma *pSma) { return tdProcessRSmaCommitImpl(pSma); }
*/
int32_t smaPostCommit(SSma *pSma) { return tdProcessRSmaPostCommitImpl(pSma); }
/**
* @brief set rsma trigger stat active
*
* @param pSma
* @return int32_t
*/
int32_t smaBegin(SSma *pSma) {
SSmaEnv *pSmaEnv = SMA_RSMA_ENV(pSma);
if (!pSmaEnv) {
return TSDB_CODE_SUCCESS;
}
SSmaStat *pStat = SMA_ENV_STAT(pSmaEnv);
SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pStat);
int8_t rsmaTriggerStat =
atomic_val_compare_exchange_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_PAUSED, TASK_TRIGGER_STAT_ACTIVE);
switch (rsmaTriggerStat) {
case TASK_TRIGGER_STAT_PAUSED: {
smaDebug("vgId:%d rsma trigger stat from paused to active", SMA_VID(pSma));
break;
}
case TASK_TRIGGER_STAT_INIT: {
atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_ACTIVE);
smaDebug("vgId:%d rsma trigger stat from init to active", SMA_VID(pSma));
break;
}
default: {
atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_ACTIVE);
smaWarn("vgId:%d rsma trigger stat %" PRIi8 " is unexpected", SMA_VID(pSma), rsmaTriggerStat);
ASSERT(0);
break;
}
}
return TSDB_CODE_SUCCESS;
}
/**
* @brief pre-commit for rollup sma.
* 1) set trigger stat of rsma timer TASK_TRIGGER_STAT_PAUSED.
* 2) perform persist task for qTaskInfo
* 3) wait all triggered fetch tasks finished
* 4) set trigger stat of rsma timer TASK_TRIGGER_STAT_ACTIVE.
* 5) finish
*
* @param pSma
* @return int32_t
......@@ -63,10 +98,30 @@ static int32_t tdProcessRSmaPreCommitImpl(SSma *pSma) {
SSmaStat *pStat = SMA_ENV_STAT(pSmaEnv);
SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pStat);
// step 1
// step 1: set persistence task paused
atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_PAUSED);
// step 2
// step 2: perform persist task for qTaskInfo
tdRSmaPersistExecImpl(pRSmaStat);
// step 3: wait all triggered fetch tasks finished
int32_t nLoops = 0;
while (1) {
if (T_REF_VAL_GET(pStat) == 0) {
smaDebug("vgId:%d, rsma fetch tasks all finished", SMA_VID(pSma));
break;
} else {
smaDebug("vgId:%d, rsma fetch tasks not all finished yet", SMA_VID(pSma));
}
++nLoops;
if (nLoops > 1000) {
sched_yield();
nLoops = 0;
}
}
smaDebug("vgId:%d, rsma pre commit succeess", SMA_VID(pSma));
return TSDB_CODE_SUCCESS;
}
......@@ -103,48 +158,68 @@ static int32_t tdProcessRSmaPostCommitImpl(SSma *pSma) {
TdDirPtr pDir = NULL;
TdDirEntryPtr pDirEntry = NULL;
char dir[TSDB_FILENAME_LEN];
char bname[TSDB_FILENAME_LEN];
const char *pattern = "^v[0-9]+qtaskinfo\\.ver([0-9]+)?$";
const char *pattern = "v[0-9]+qtaskinfo\\.ver([0-9]+)?$";
regex_t regex;
int code = 0;
tdGetVndDirName(TD_VID(pVnode), VNODE_RSMA_DIR, dir);
tdGetVndDirName(TD_VID(pVnode), tfsGetPrimaryPath(pVnode->pTfs), VNODE_RSMA_DIR, true, dir);
// Resource allocation and init
regcomp(&regex, pattern, REG_EXTENDED);
if ((code = regcomp(&regex, pattern, REG_EXTENDED)) != 0) {
char errbuf[128];
regerror(code, &regex, errbuf, sizeof(errbuf));
smaWarn("vgId:%d, rsma post commit, regcomp for %s failed since %s", TD_VID(pVnode), dir, errbuf);
return TSDB_CODE_FAILED;
}
if ((pDir = taosOpenDir(dir)) == NULL) {
terrno = TAOS_SYSTEM_ERROR(errno);
smaWarn("rsma post-commit open dir %s failed since %s", dir, terrstr());
smaWarn("vgId:%d, rsma post commit, open dir %s failed since %s", TD_VID(pVnode), dir, terrstr());
return TSDB_CODE_FAILED;
}
int32_t dirLen = strlen(dir);
char *dirEnd = POINTER_SHIFT(dir, dirLen);
regmatch_t regMatch[2];
while ((pDirEntry = taosReadDir(pDir)) != NULL) {
char *entryName = taosGetDirEntryName(pDirEntry);
if (!entryName) {
continue;
}
char *fileName = taosDirEntryBaseName(entryName);
int code = regexec(&regex, bname, 2, regMatch, 0);
code = regexec(&regex, entryName, 2, regMatch, 0);
if (code == 0) {
// match
printf("match 0 = %s\n", (char *)POINTER_SHIFT(fileName, regMatch[0].rm_so));
printf("match 1 = %s\n", (char *)POINTER_SHIFT(fileName, regMatch[1].rm_so));
int64_t version = -1;
sscanf((const char *)POINTER_SHIFT(entryName, regMatch[1].rm_so), "%" PRIi64, &version);
if ((version < committed) && (version > -1)) {
strncpy(dirEnd, entryName, TSDB_FILENAME_LEN - dirLen);
if (taosRemoveFile(dir) != 0) {
terrno = TAOS_SYSTEM_ERROR(errno);
smaWarn("vgId:%d, committed version:%" PRIi64 ", failed to remove %s since %s", TD_VID(pVnode), committed,
dir, terrstr());
} else {
smaDebug("vgId:%d, committed version:%" PRIi64 ", success to remove %s", TD_VID(pVnode), committed, dir);
}
}
} else if (code == REG_NOMATCH) {
// not match
smaInfo("rsma post-commit not match %s", fileName);
smaTrace("vgId:%d, rsma post commit, not match %s", TD_VID(pVnode), entryName);
continue;
} else {
// has other error
terrno = TAOS_SYSTEM_ERROR(code);
smaWarn("rsma post-commit regexec failed since %s", terrstr());
char errbuf[128];
regerror(code, &regex, errbuf, sizeof(errbuf));
smaWarn("vgId:%d, rsma post commit, regexec failed since %s", TD_VID(pVnode), errbuf);
taosCloseDir(&pDir);
regfree(&regex);
return TSDB_CODE_FAILED;
}
}
taosCloseDir(&pDir);
regfree(&regex);
return TSDB_CODE_SUCCESS;
}
......@@ -132,6 +132,7 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pS
if (smaType == TSDB_SMA_TYPE_ROLLUP) {
SRSmaStat *pRSmaStat = (SRSmaStat *)(*pSmaStat);
pRSmaStat->pSma = (SSma *)pSma;
atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_INIT);
// init smaMgmt
smaMgmt.smaRef = taosOpenRef(SMA_MGMT_REF_NUM, tdDestroyRSmaStat);
......@@ -192,22 +193,20 @@ static void *tdFreeTSmaStat(STSmaStat *pStat) {
static void tdDestroyRSmaStat(void *pRSmaStat) {
if (pRSmaStat) {
SRSmaStat *pStat = (SRSmaStat *)pRSmaStat;
smaDebug("vgId:%d %s:%d destroy rsma stat %p", SMA_VID(pStat->pSma), __func__, __LINE__, pRSmaStat);
// step 1: set persistence task cancelled
SSma *pSma = pStat->pSma;
smaDebug("vgId:%d, destroy rsma stat %p", SMA_VID(pSma), pRSmaStat);
// step 1: set rsma trigger stat cancelled
atomic_store_8(RSMA_TRIGGER_STAT(pStat), TASK_TRIGGER_STAT_CANCELLED);
// step 2: stop the persistence timer
taosTmrStopA(&RSMA_TMR_ID(pStat));
// step 3: wait the persistence thread to finish
// step 2: wait the persistence thread to finish
int32_t nLoops = 0;
if (atomic_load_8(RSMA_RUNNING_STAT(pStat)) == 1) {
while (1) {
if (atomic_load_8(RSMA_TRIGGER_STAT(pStat)) == TASK_TRIGGER_STAT_FINISHED) {
smaDebug("rsma, persist task finished already");
smaDebug("vgId:%d, rsma persist task finished already", SMA_VID(pSma));
break;
} else {
smaDebug("rsma, persist task not finished yet since rsma stat in %" PRIi8,
smaDebug("vgId:%d, rsma persist task not finished yet since rsma stat in %" PRIi8, SMA_VID(pSma),
atomic_load_8(RSMA_TRIGGER_STAT(pStat)));
}
++nLoops;
......@@ -218,13 +217,15 @@ static void tdDestroyRSmaStat(void *pRSmaStat) {
}
}
// step 4: destroy the rsma info and associated fetch tasks
// step 3: destroy the rsma info and associated fetch tasks
// TODO: use taosHashSetFreeFp when taosHashSetFreeFp is ready.
void *infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), NULL);
while (infoHash) {
SRSmaInfo *pSmaInfo = *(SRSmaInfo **)infoHash;
tdFreeRSmaInfo(pSmaInfo);
infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), infoHash);
if (taosHashGetSize(RSMA_INFO_HASH(pStat)) > 0) {
void *infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), NULL);
while (infoHash) {
SRSmaInfo *pSmaInfo = *(SRSmaInfo **)infoHash;
tdFreeRSmaInfo(pSmaInfo);
infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), infoHash);
}
}
taosHashCleanup(RSMA_INFO_HASH(pStat));
......@@ -232,10 +233,10 @@ static void tdDestroyRSmaStat(void *pRSmaStat) {
nLoops = 0;
while (1) {
if (T_REF_VAL_GET((SSmaStat *)pStat) == 0) {
smaDebug("rsma, all fetch task finished already");
smaDebug("vgId:%d, rsma fetch tasks all finished", SMA_VID(pSma));
break;
} else {
smaDebug("rsma, fetch tasks not all finished yet");
smaDebug("vgId:%d, rsma fetch tasks not all finished yet", SMA_VID(pSma));
}
++nLoops;
if (nLoops > 1000) {
......@@ -275,7 +276,7 @@ int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType) {
} else if (smaType == TSDB_SMA_TYPE_ROLLUP) {
SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pSmaStat);
if (taosRemoveRef(smaMgmt.smaRef, RSMA_REF_ID(pRSmaStat)) < 0) {
smaError("remove refId from smaRef failed, refId:0x%" PRIx64, RSMA_REF_ID(pRSmaStat));
smaError("remove refId from rsmaRef:0x%" PRIx64 " failed since %s", RSMA_REF_ID(pRSmaStat), terrstr());
}
} else {
ASSERT(0);
......
......@@ -135,17 +135,11 @@ _err:
return -1;
}
int32_t smaCloseEnv(SSma *pSma) {
int32_t smaClose(SSma *pSma) {
if (pSma) {
taosThreadMutexDestroy(&pSma->mutex);
SMA_TSMA_ENV(pSma) = tdFreeSmaEnv(SMA_TSMA_ENV(pSma));
SMA_RSMA_ENV(pSma) = tdFreeSmaEnv(SMA_RSMA_ENV(pSma));
}
return 0;
}
int32_t smaCloseEx(SSma *pSma) {
if (pSma) {
taosThreadMutexDestroy(&pSma->mutex);
if SMA_RSMA_TSDB0 (pSma) tsdbClose(&SMA_RSMA_TSDB0(pSma));
if SMA_RSMA_TSDB1 (pSma) tsdbClose(&SMA_RSMA_TSDB1(pSma));
if SMA_RSMA_TSDB2 (pSma) tsdbClose(&SMA_RSMA_TSDB2(pSma));
......
......@@ -15,9 +15,8 @@
#include "sma.h"
#define RSMA_QTASKINFO_PERSIST_MS 7200000
#define RSMA_QTASKINFO_BUFSIZE 32768
#define RSMA_QTASKINFO_HEAD_LEN (sizeof(int32_t) + sizeof(int8_t) + sizeof(int64_t)) // len + type + suid
#define RSMA_QTASKINFO_BUFSIZE 32768
#define RSMA_QTASKINFO_HEAD_LEN (sizeof(int32_t) + sizeof(int8_t) + sizeof(int64_t)) // len + type + suid
SSmaMgmt smaMgmt = {
.smaRef = -1,
......@@ -43,9 +42,9 @@ static int32_t tdRSmaQTaskInfoIterNextBlock(SRSmaQTaskInfoIter *pIter, bool *isF
static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter);
static int32_t tdRSmaQTaskInfoItemRestore(SSma *pSma, const SRSmaQTaskInfoItem *infoItem);
static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma);
static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma);
static int32_t tdRSmaRestoreTSDataReload(SSma *pSma);
static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma, int64_t *nTables);
static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma, int64_t *committed);
static int32_t tdRSmaRestoreTSDataReload(SSma *pSma, int64_t committed);
struct SRSmaInfoItem {
SRSmaInfo *pRsmaInfo;
......@@ -88,7 +87,7 @@ struct SRSmaQTaskInfoIter {
};
static void tdRSmaQTaskInfoGetFName(int32_t vgId, int64_t version, char *outputName) {
tdGetVndFileName(vgId, VNODE_RSMA_DIR, TD_QTASKINFO_FNAME_PREFIX, version, outputName);
tdGetVndFileName(vgId, NULL, VNODE_RSMA_DIR, TD_QTASKINFO_FNAME_PREFIX, version, outputName);
}
static FORCE_INLINE int32_t tdRSmaQTaskInfoContLen(int32_t lenWithHead) {
......@@ -114,12 +113,14 @@ void *tdFreeRSmaInfo(SRSmaInfo *pInfo) {
for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
SRSmaInfoItem *pItem = &pInfo->items[i];
if (pItem->taskInfo) {
smaDebug("vgId:%d, stb %" PRIi64 " stop fetch-timer %p level %d", SMA_VID(pSma), pInfo->suid, pItem->tmrId,
i + 1);
taosTmrStopA(&pItem->tmrId);
if (pItem->tmrId) {
smaDebug("vgId:%d, table %" PRIi64 " stop fetch timer %p level %d", SMA_VID(pSma), pInfo->suid, pItem->tmrId,
i + 1);
taosTmrStopA(&pItem->tmrId);
}
tdFreeTaskHandle(&pItem->taskInfo, SMA_VID(pSma), i + 1);
} else {
smaDebug("vgId:%d, stb %" PRIi64 " no need to destroy rsma info level %d since empty taskInfo", SMA_VID(pSma),
smaDebug("vgId:%d, table %" PRIi64 " no need to destroy rsma info level %d since empty taskInfo", SMA_VID(pSma),
pInfo->suid, i + 1);
}
}
......@@ -358,13 +359,7 @@ int32_t tdProcessRSmaCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, con
goto _err;
}
smaDebug("vgId:%d, register rsma info succeed for suid:%" PRIi64, SMA_VID(pSma), suid);
// start the persist timer
if (TASK_TRIGGER_STAT_INIT ==
atomic_val_compare_exchange_8(RSMA_TRIGGER_STAT(pStat), TASK_TRIGGER_STAT_INIT, TASK_TRIGGER_STAT_ACTIVE)) {
taosTmrStart(tdRSmaPersistTrigger, RSMA_QTASKINFO_PERSIST_MS, pStat, RSMA_TMR_HANDLE(pStat));
}
smaDebug("vgId:%d, register rsma info succeed for table %" PRIi64, SMA_VID(pSma), suid);
return TSDB_CODE_SUCCESS;
_err:
......@@ -748,7 +743,7 @@ int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
return TSDB_CODE_SUCCESS;
}
static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma) {
static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma, int64_t *nTables) {
SVnode *pVnode = pSma->pVnode;
SArray *suidList = taosArrayInit(1, sizeof(tb_uid_t));
......@@ -758,7 +753,12 @@ static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma) {
return TSDB_CODE_FAILED;
}
int32_t arrSize = taosArrayGetSize(suidList);
int64_t arrSize = taosArrayGetSize(suidList);
if (nTables) {
*nTables = arrSize;
}
if (arrSize == 0) {
taosArrayDestroy(suidList);
smaDebug("vgId:%d, no need to restore rsma env since empty stb id list", TD_VID(pVnode));
......@@ -767,9 +767,9 @@ static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma) {
SMetaReader mr = {0};
metaReaderInit(&mr, SMA_META(pSma), 0);
for (int32_t i = 0; i < arrSize; ++i) {
for (int64_t i = 0; i < arrSize; ++i) {
tb_uid_t suid = *(tb_uid_t *)taosArrayGet(suidList, i);
smaDebug("vgId:%d, rsma restore, suid[%d] is %" PRIi64, TD_VID(pVnode), i, suid);
smaDebug("vgId:%d, rsma restore, suid is %" PRIi64, TD_VID(pVnode), suid);
if (metaGetTableEntryByUid(&mr, suid) < 0) {
smaError("vgId:%d, rsma restore, failed to get table meta for %" PRIi64 " since %s", TD_VID(pVnode), suid,
terrstr());
......@@ -803,7 +803,7 @@ _err:
return TSDB_CODE_FAILED;
}
static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma) {
static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma, int64_t *committed) {
SVnode *pVnode = pSma->pVnode;
STFile tFile = {0};
char qTaskInfoFName[TSDB_FILENAME_LEN] = {0};
......@@ -814,6 +814,13 @@ static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma) {
}
if (!taosCheckExistFile(TD_TFILE_FULL_NAME(&tFile))) {
if (pVnode->state.committed > 0) {
smaWarn("vgId:%d, rsma restore for version %" PRIi64 ", not start as %s not exist", TD_VID(pVnode),
pVnode->state.committed, TD_TFILE_FULL_NAME(&tFile));
} else {
smaDebug("vgId:%d, rsma restore for version %" PRIi64 ", no need as %s not exist", TD_VID(pVnode),
pVnode->state.committed, TD_TFILE_FULL_NAME(&tFile));
}
return TSDB_CODE_SUCCESS;
}
......@@ -839,9 +846,14 @@ static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma) {
tdRSmaQTaskInfoIterDestroy(&fIter);
tdCloseTFile(&tFile);
tdDestroyTFile(&tFile);
// restored successfully from committed
*committed = pVnode->state.committed;
return TSDB_CODE_SUCCESS;
_err:
smaError("rsma restore, qtaskinfo reload failed since %s", terrstr());
smaError("vgId:%d, rsma restore for version %" PRIi64 ", qtaskinfo reload failed since %s", TD_VID(pVnode),
pVnode->state.committed, terrstr());
return TSDB_CODE_FAILED;
}
......@@ -849,35 +861,45 @@ _err:
* @brief reload ts data from checkpoint
*
* @param pSma
* @param committed restore from committed version
* @return int32_t
*/
static int32_t tdRSmaRestoreTSDataReload(SSma *pSma) {
static int32_t tdRSmaRestoreTSDataReload(SSma *pSma, int64_t committed) {
// TODO
smaDebug("vgId:%d, rsma restore from %" PRIi64 ", ts data reload success", SMA_VID(pSma), committed);
return TSDB_CODE_SUCCESS;
_err:
smaError("rsma restore, ts data reload failed since %s", terrstr());
smaError("vgId:%d, rsma restore from %" PRIi64 ", ts data reload failed since %s", SMA_VID(pSma), committed,
terrstr());
return TSDB_CODE_FAILED;
}
int32_t tdProcessRSmaRestoreImpl(SSma *pSma) {
// step 1: iterate all stables to restore the rsma env
if (tdRSmaRestoreQTaskInfoInit(pSma) < 0) {
int64_t nTables = 0;
if (tdRSmaRestoreQTaskInfoInit(pSma, &nTables) < 0) {
goto _err;
}
if (nTables <= 0) {
smaDebug("vgId:%d, no need to restore rsma task since no tables", SMA_VID(pSma));
return TSDB_CODE_SUCCESS;
}
// step 2: retrieve qtaskinfo items from the persistence file(rsma/qtaskinfo) and restore
if (tdRSmaRestoreQTaskInfoReload(pSma) < 0) {
int64_t committed = -1;
if (tdRSmaRestoreQTaskInfoReload(pSma, &committed) < 0) {
goto _err;
}
// step 3: reload ts data from checkpoint
if (tdRSmaRestoreTSDataReload(pSma) < 0) {
if (tdRSmaRestoreTSDataReload(pSma, committed) < 0) {
goto _err;
}
return TSDB_CODE_SUCCESS;
_err:
smaError("failed to restore rsma task since %s", terrstr());
smaError("vgId:%d failed to restore rsma task since %s", SMA_VID(pSma), terrstr());
return TSDB_CODE_FAILED;
}
......@@ -1017,7 +1039,8 @@ static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter) {
pIter->qBuf = taosDecodeFixedI32(pIter->qBuf, &qTaskInfoLenWithHead);
if (qTaskInfoLenWithHead < RSMA_QTASKINFO_HEAD_LEN) {
terrno = TSDB_CODE_TDB_FILE_CORRUPTED;
smaError("restore rsma qtaskinfo file %s failed since %s", TD_TFILE_FULL_NAME(pIter->pTFile), terrstr());
smaError("vgId:%d, restore rsma qtaskinfo file %s failed since %s", SMA_VID(pSma),
TD_TFILE_FULL_NAME(pIter->pTFile), terrstr());
return TSDB_CODE_FAILED;
}
......@@ -1054,13 +1077,17 @@ static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter) {
return TSDB_CODE_SUCCESS;
}
static int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat) {
int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat) {
SSma *pSma = pRSmaStat->pSma;
SVnode *pVnode = pSma->pVnode;
int32_t vid = SMA_VID(pSma);
int64_t toffset = 0;
bool isFileCreated = false;
if (taosHashGetSize(RSMA_INFO_HASH(pRSmaStat)) <= 0) {
return TSDB_CODE_SUCCESS;
}
void *infoHash = taosHashIterate(RSMA_INFO_HASH(pRSmaStat), NULL);
if (!infoHash) {
return TSDB_CODE_SUCCESS;
......@@ -1099,11 +1126,15 @@ static int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat) {
char qTaskInfoFName[TSDB_FILENAME_LEN];
tdRSmaQTaskInfoGetFName(vid, pSma->pVnode->state.applied, qTaskInfoFName);
if (tdInitTFile(&tFile, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFName) < 0) {
smaError("vgId:%d, rsma persit, init %s failed since %s", vid, qTaskInfoFName, terrstr());
goto _err;
}
if (tdCreateTFile(&tFile, true, -1) < 0) {
smaError("vgId:%d, rsma persit, create %s failed since %s", vid, TD_TFILE_FULL_NAME(&tFile), terrstr());
goto _err;
}
smaDebug("vgId:%d, rsma, table %" PRIi64 " level %d serialize qTaskInfo, file %s created", vid, pRSmaInfo->suid,
i + 1, TD_TFILE_FULL_NAME(&tFile));
isFileCreated = true;
}
......@@ -1143,6 +1174,7 @@ static int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat) {
}
return TSDB_CODE_SUCCESS;
_err:
smaError("vgId:%d, rsma persit failed since %s", vid, terrstr());
if (isFileCreated) {
tdRemoveTFile(&tFile);
tdDestroyTFile(&tFile);
......@@ -1238,8 +1270,8 @@ static void tdRSmaPersistTrigger(void *param, void *tmrId) {
// start persist task
tdRSmaPersistTask(pRSmaStat);
taosTmrReset(tdRSmaPersistTrigger, RSMA_QTASKINFO_PERSIST_MS, pRSmaStat, pRSmaStat->tmrHandle,
&pRSmaStat->tmrId);
// taosTmrReset(tdRSmaPersistTrigger, 5000, pRSmaStat, pRSmaStat->tmrHandle,
// RSMA_TMR_ID(pRSmaStat));
} else {
atomic_store_8(RSMA_RUNNING_STAT(pRSmaStat), 0);
}
......
......@@ -140,7 +140,7 @@ int64_t tdAppendTFile(STFile *pTFile, void *buf, int64_t nbyte, int64_t *offset)
return -1;
}
#if 1
#if 0
smaDebug("append to file %s, offset:%" PRIi64 " nbyte:%" PRIi64 " fsize:%" PRIi64, TD_TFILE_FULL_NAME(pTFile),
toffset, nbyte, toffset + nbyte);
#endif
......@@ -181,16 +181,43 @@ void tdCloseTFile(STFile *pTFile) {
void tdDestroyTFile(STFile *pTFile) { taosMemoryFreeClear(TD_TFILE_FULL_NAME(pTFile)); }
void tdGetVndFileName(int32_t vgId, const char *dname, const char *fname, int64_t version, char *outputName) {
void tdGetVndFileName(int32_t vgId, const char *pdname, const char *dname, const char *fname, int64_t version,
char *outputName) {
if (version < 0) {
snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s/v%d%s", vgId, dname, vgId, fname);
if (pdname) {
snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s%sv%d%s", pdname, TD_DIRSEP, TD_DIRSEP, vgId,
TD_DIRSEP, dname, TD_DIRSEP, vgId, fname);
} else {
snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s%sv%d%s", TD_DIRSEP, vgId, TD_DIRSEP, dname, TD_DIRSEP,
vgId, fname);
}
} else {
snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s/v%d%s%" PRIi64, vgId, dname, vgId, fname, version);
if (pdname) {
snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s%sv%d%s%" PRIi64, pdname, TD_DIRSEP, TD_DIRSEP,
vgId, TD_DIRSEP, dname, TD_DIRSEP, vgId, fname, version);
} else {
snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s%sv%d%s%" PRIi64, TD_DIRSEP, vgId, TD_DIRSEP, dname,
TD_DIRSEP, vgId, fname, version);
}
}
}
void tdGetVndDirName(int32_t vgId, const char *dname, char *outputName) {
snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s", vgId, dname);
void tdGetVndDirName(int32_t vgId, const char *pdname, const char *dname, bool endWithSep, char *outputName) {
if (pdname) {
if (endWithSep) {
snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s%s", pdname, TD_DIRSEP, TD_DIRSEP, vgId, TD_DIRSEP,
dname, TD_DIRSEP);
} else {
snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s", pdname, TD_DIRSEP, TD_DIRSEP, vgId, TD_DIRSEP,
dname);
}
} else {
if (endWithSep) {
snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s%s", TD_DIRSEP, vgId, TD_DIRSEP, dname, TD_DIRSEP);
} else {
snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s", TD_DIRSEP, vgId, TD_DIRSEP, dname);
}
}
}
int32_t tdInitTFile(STFile *pTFile, const char *dname, const char *fname) {
......@@ -215,35 +242,36 @@ int32_t tdInitTFile(STFile *pTFile, const char *dname, const char *fname) {
int32_t tdCreateTFile(STFile *pTFile, bool updateHeader, int8_t fType) {
ASSERT(pTFile->info.fsize == 0 && pTFile->info.magic == TD_FILE_INIT_MAGIC);
pTFile->pFile = taosOpenFile(TD_TFILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
if (pTFile->pFile == NULL) {
if (errno == ENOENT) {
// Try to create directory recursively
if (taosMulMkDir(taosDirName(TD_TFILE_FULL_NAME(pTFile))) != 0) {
char *s = strdup(TD_TFILE_FULL_NAME(pTFile));
if (taosMulMkDir(taosDirName(s)) != 0) {
terrno = TAOS_SYSTEM_ERROR(errno);
taosMemoryFree(s);
return -1;
}
taosMemoryFree(s);
pTFile->pFile = taosOpenFile(TD_TFILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
if (pTFile->pFile == NULL) {
terrno = TAOS_SYSTEM_ERROR(errno);
return -1;
} else {
pTFile->pFile = taosOpenFile(TD_TFILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
if (pTFile->pFile == NULL) {
terrno = TAOS_SYSTEM_ERROR(errno);
return -1;
}
}
}
}
if (!updateHeader) {
return 0;
}
if (!updateHeader) {
return 0;
}
pTFile->info.fsize += TD_FILE_HEAD_SIZE;
pTFile->info.fver = 0;
pTFile->info.fsize += TD_FILE_HEAD_SIZE;
pTFile->info.fver = 0;
if (tdUpdateTFileHeader(pTFile) < 0) {
tdCloseTFile(pTFile);
tdRemoveTFile(pTFile);
return -1;
}
if (tdUpdateTFileHeader(pTFile) < 0) {
tdCloseTFile(pTFile);
tdRemoveTFile(pTFile);
return -1;
}
return 0;
......
......@@ -69,6 +69,9 @@ int vnodeBegin(SVnode *pVnode) {
}
}
// begin sma
smaBegin(pVnode->pSma); // TODO: refactor to include the rsma1/rsma2 tsdbBegin() after tsdb_refact branch merged
return 0;
}
......@@ -230,8 +233,8 @@ int vnodeCommit(SVnode *pVnode) {
}
// preCommit
// TODO
smaPreCommit(pVnode->pSma);
// commit each sub-system
if (metaCommit(pVnode->pMeta) < 0) {
ASSERT(0);
......
......@@ -152,12 +152,11 @@ SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb) {
return pVnode;
_err:
if (pVnode->pSma) smaCloseEnv(pVnode->pSma);
if (pVnode->pQuery) vnodeQueryClose(pVnode);
if (pVnode->pTq) tqClose(pVnode->pTq);
if (pVnode->pWal) walClose(pVnode->pWal);
if (pVnode->pTsdb) tsdbClose(&pVnode->pTsdb);
if (pVnode->pSma) smaCloseEx(pVnode->pSma);
if (pVnode->pSma) smaClose(pVnode->pSma);
if (pVnode->pMeta) metaClose(pVnode->pMeta);
tsem_destroy(&(pVnode->canCommit));
......@@ -167,14 +166,13 @@ _err:
void vnodeClose(SVnode *pVnode) {
if (pVnode) {
smaCloseEnv(pVnode->pSma);
vnodeCommit(pVnode);
vnodeSyncClose(pVnode);
vnodeQueryClose(pVnode);
walClose(pVnode->pWal);
tqClose(pVnode->pTq);
if (pVnode->pTsdb) tsdbClose(&pVnode->pTsdb);
smaCloseEx(pVnode->pSma);
smaClose(pVnode->pSma);
metaClose(pVnode->pMeta);
vnodeCloseBufPool(pVnode);
// destroy handle
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册