Merge pull request #29534 from taosdata/fix/TD-31605/crashLogMain

fix: ts31605, write crash log
This commit is contained in:
Shengliang Guan 2025-01-10 11:22:59 +08:00 committed by GitHub
commit ad6e9c46d8
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
7 changed files with 145 additions and 43 deletions

View File

@ -119,6 +119,11 @@ void taosLogCrashInfo(char *nodeType, char *pMsg, int64_t msgLen, int signum, vo
void taosReadCrashInfo(char *filepath, char **pMsg, int64_t *pMsgLen, TdFilePtr *pFd);
void taosReleaseCrashLogFile(TdFilePtr pFile, bool truncateFile);
int32_t initCrashLogWriter();
void checkAndPrepareCrashInfo();
bool reportThreadSetQuit();
void writeCrashLogToFile(int signum, void *sigInfo, char *nodeType, int64_t clusterId, int64_t startTime);
// clang-format off
#define uFatal(...) { if (uDebugFlag & DEBUG_FATAL) { taosPrintLog("UTL FATAL", DEBUG_FATAL, tsLogEmbedded ? 255 : uDebugFlag, __VA_ARGS__); }}
#define uError(...) { if (uDebugFlag & DEBUG_ERROR) { taosPrintLog("UTL ERROR ", DEBUG_ERROR, tsLogEmbedded ? 255 : uDebugFlag, __VA_ARGS__); }}

View File

@ -43,7 +43,7 @@
#endif
#ifndef CUS_PROMPT
#define CUS_PROMPT "tao"
#define CUS_PROMPT "taos"
#endif
#define TSC_VAR_NOT_RELEASE 1
@ -831,9 +831,17 @@ static void *tscCrashReportThreadFp(void *param) {
return NULL;
}
code = initCrashLogWriter();
if (code) {
tscError("failed to init crash log writer, code:%s", tstrerror(code));
return NULL;
}
while (1) {
if (clientStop > 0) break;
checkAndPrepareCrashInfo();
if (clientStop > 0 && reportThreadSetQuit()) break;
if (loopTimes++ < reportPeriodNum) {
if (loopTimes < 0) loopTimes = reportPeriodNum;
taosMsleep(sleepTime);
continue;
}
@ -921,21 +929,7 @@ void tscStopCrashReport() {
}
void tscWriteCrashInfo(int signum, void *sigInfo, void *context) {
char *pMsg = NULL;
const char *flags = "UTL FATAL ";
ELogLevel level = DEBUG_FATAL;
int32_t dflag = 255;
int64_t msgLen = -1;
if (tsEnableCrashReport) {
if (taosGenCrashJsonMsg(signum, &pMsg, lastClusterId, appInfo.startTime)) {
taosPrintLog(flags, level, dflag, "failed to generate crash json msg");
} else {
msgLen = strlen(pMsg);
}
}
taosLogCrashInfo("taos", pMsg, msgLen, signum, sigInfo);
writeCrashLogToFile(signum, sigInfo, CUS_PROMPT, lastClusterId, appInfo.startTime);
}
void taos_init_imp(void) {
@ -969,7 +963,7 @@ void taos_init_imp(void) {
}
taosHashSetFreeFp(appInfo.pInstMap, destroyAppInst);
const char *logName = CUS_PROMPT "slog";
const char *logName = CUS_PROMPT "log";
ENV_ERR_RET(taosInitLogOutput(&logName), "failed to init log output");
if (taosCreateLog(logName, 10, configDir, NULL, NULL, NULL, NULL, 1) != 0) {
(void)printf(" WARING: Create %s failed:%s. configDir=%s\n", logName, strerror(errno), configDir);

View File

@ -231,6 +231,7 @@ int32_t taosGenCrashJsonMsg(int signum, char** pMsg, int64_t clusterId, int64_t
TAOS_CHECK_GOTO(tjsonAddIntegerToObject(pJson, "crashSig", signum), NULL, _exit);
TAOS_CHECK_GOTO(tjsonAddIntegerToObject(pJson, "crashTs", taosGetTimestampUs()), NULL, _exit);
#if 0
#ifdef _TD_DARWIN_64
taosLogTraceToBuf(tmp, sizeof(tmp), 4);
#elif !defined(WINDOWS)
@ -240,7 +241,7 @@ int32_t taosGenCrashJsonMsg(int signum, char** pMsg, int64_t clusterId, int64_t
#endif
TAOS_CHECK_GOTO(tjsonAddStringToObject(pJson, "stackInfo", tmp), NULL, _exit);
#endif
char* pCont = tjsonToString(pJson);
if (pCont == NULL) {
code = terrno;

View File

@ -131,25 +131,7 @@ void dmLogCrash(int signum, void *sigInfo, void *context) {
if (taosIgnSignal(SIGSEGV) != 0) {
dWarn("failed to ignore signal SIGABRT");
}
char *pMsg = NULL;
const char *flags = "UTL FATAL ";
ELogLevel level = DEBUG_FATAL;
int32_t dflag = 255;
int64_t msgLen = -1;
if (tsEnableCrashReport) {
if (taosGenCrashJsonMsg(signum, &pMsg, dmGetClusterId(), global.startTime)) {
taosPrintLog(flags, level, dflag, "failed to generate crash json msg");
goto _return;
} else {
msgLen = strlen(pMsg);
}
}
_return:
taosLogCrashInfo(CUS_PROMPT "d", pMsg, msgLen, signum, sigInfo);
writeCrashLogToFile(signum, sigInfo, CUS_PROMPT "d", dmGetClusterId(), global.startTime);
#ifdef _TD_DARWIN_64
exit(signum);
@ -177,6 +159,15 @@ static void dmSetSignalHandle() {
if (taosSetSignal(SIGBREAK, dmStopDnode) != 0) {
dWarn("failed to set signal SIGUSR1");
}
if (taosSetSignal(SIGABRT, dmLogCrash) != 0) {
dWarn("failed to set signal SIGUSR1");
}
if (taosSetSignal(SIGFPE, dmLogCrash) != 0) {
dWarn("failed to set signal SIGUSR1");
}
if (taosSetSignal(SIGSEGV, dmLogCrash) != 0) {
dWarn("failed to set signal SIGUSR1");
}
#ifndef WINDOWS
if (taosSetSignal(SIGTSTP, dmStopDnode) != 0) {
dWarn("failed to set signal SIGUSR1");
@ -184,6 +175,9 @@ static void dmSetSignalHandle() {
if (taosSetSignal(SIGQUIT, dmStopDnode) != 0) {
dWarn("failed to set signal SIGUSR1");
}
if (taosSetSignal(SIGBUS, dmLogCrash) != 0) {
dWarn("failed to set signal SIGUSR1");
}
#endif
}

View File

@ -274,14 +274,22 @@ static void *dmCrashReportThreadFp(void *param) {
dError("failed to init telemetry since %s", tstrerror(code));
return NULL;
}
code = initCrashLogWriter();
if (code != 0) {
dError("failed to init crash log writer since %s", tstrerror(code));
return NULL;
}
while (1) {
if (pMgmt->pData->dropped || pMgmt->pData->stopped) break;
checkAndPrepareCrashInfo();
if ((pMgmt->pData->dropped || pMgmt->pData->stopped) && reportThreadSetQuit()) {
break;
}
if (loopTimes++ < reportPeriodNum) {
taosMsleep(sleepTime);
if(loopTimes < 0) loopTimes = reportPeriodNum;
continue;
}
taosReadCrashInfo(filepath, &pMsg, &msgLen, &pFile);
if (pMsg && msgLen > 0) {
if (taosSendTelemReport(&mgt, tsSvrCrashReportUri, tsTelemPort, pMsg, msgLen, HTTP_FLAT) != 0) {

View File

@ -21,6 +21,7 @@
#include "tjson.h"
#include "ttime.h"
#include "tutil.h"
#include "tcommon.h"
#define LOG_MAX_LINE_SIZE (10024)
#define LOG_MAX_LINE_BUFFER_SIZE (LOG_MAX_LINE_SIZE + 3)
@ -1264,6 +1265,8 @@ _return:
taosPrintLog(flags, level, dflag, "crash signal is %d", signum);
// print the stack trace
#if 0
#ifdef _TD_DARWIN_64
taosPrintTrace(flags, level, dflag, 4);
#elif !defined(WINDOWS)
@ -1273,10 +1276,109 @@ _return:
#else
taosPrintTrace(flags, level, dflag, 8);
#endif
#endif
taosMemoryFree(pMsg);
}
typedef enum {
CRASH_LOG_WRITER_UNKNOWN = 0,
CRASH_LOG_WRITER_INIT = 1,
CRASH_LOG_WRITER_WAIT,
CRASH_LOG_WRITER_RUNNING,
CRASH_LOG_WRITER_QUIT
} CrashStatus;
typedef struct crashBasicInfo {
int8_t status;
int64_t clusterId;
int64_t startTime;
char *nodeType;
int signum;
void *sigInfo;
tsem_t sem;
int64_t reportThread;
} crashBasicInfo;
crashBasicInfo gCrashBasicInfo = {0};
void setCrashWriterStatus(int8_t status) { atomic_store_8(&gCrashBasicInfo.status, status); }
bool reportThreadSetQuit() {
CrashStatus status =
atomic_val_compare_exchange_8(&gCrashBasicInfo.status, CRASH_LOG_WRITER_INIT, CRASH_LOG_WRITER_QUIT);
if (status == CRASH_LOG_WRITER_INIT) {
return true;
} else {
return false;
}
}
bool setReportThreadWait() {
CrashStatus status =
atomic_val_compare_exchange_8(&gCrashBasicInfo.status, CRASH_LOG_WRITER_INIT, CRASH_LOG_WRITER_WAIT);
if (status == CRASH_LOG_WRITER_INIT) {
return true;
} else {
return false;
}
}
bool setReportThreadRunning() {
CrashStatus status =
atomic_val_compare_exchange_8(&gCrashBasicInfo.status, CRASH_LOG_WRITER_WAIT, CRASH_LOG_WRITER_RUNNING);
if (status == CRASH_LOG_WRITER_WAIT) {
return true;
} else {
return false;
}
}
static void checkWriteCrashLogToFileInNewThead() {
if (setReportThreadRunning()) {
char *pMsg = NULL;
const char *flags = "UTL FATAL ";
ELogLevel level = DEBUG_FATAL;
int32_t dflag = 255;
int64_t msgLen = -1;
if (tsEnableCrashReport) {
if (taosGenCrashJsonMsg(gCrashBasicInfo.signum, &pMsg, gCrashBasicInfo.clusterId, gCrashBasicInfo.startTime)) {
taosPrintLog(flags, level, dflag, "failed to generate crash json msg");
} else {
msgLen = strlen(pMsg);
}
}
taosLogCrashInfo(gCrashBasicInfo.nodeType, pMsg, msgLen, gCrashBasicInfo.signum, gCrashBasicInfo.sigInfo);
setCrashWriterStatus(CRASH_LOG_WRITER_INIT);
tsem_post(&gCrashBasicInfo.sem);
}
}
void checkAndPrepareCrashInfo() {
return checkWriteCrashLogToFileInNewThead();
}
int32_t initCrashLogWriter() {
int32_t code = tsem_init(&gCrashBasicInfo.sem, 0, 0);
if (code != 0) {
uError("failed to init sem for crashLogWriter, code:%d", code);
return code;
}
gCrashBasicInfo.reportThread = taosGetSelfPthreadId();
setCrashWriterStatus(CRASH_LOG_WRITER_INIT);
return code;
}
void writeCrashLogToFile(int signum, void *sigInfo, char *nodeType, int64_t clusterId, int64_t startTime) {
if (gCrashBasicInfo.reportThread == taosGetSelfPthreadId()) {
return;
}
if (setReportThreadWait()) {
gCrashBasicInfo.clusterId = clusterId;
gCrashBasicInfo.startTime = startTime;
gCrashBasicInfo.nodeType = nodeType;
gCrashBasicInfo.signum = signum;
gCrashBasicInfo.sigInfo = sigInfo;
tsem_wait(&gCrashBasicInfo.sem);
}
}
void taosReadCrashInfo(char *filepath, char **pMsg, int64_t *pMsgLen, TdFilePtr *pFd) {
const char *flags = "UTL FATAL ";
ELogLevel level = DEBUG_FATAL;

View File

@ -49,14 +49,12 @@ int main(int argc, char *argv[]) {
shell.args.local = false;
#endif
#if 0
#if !defined(WINDOWS)
taosSetSignal(SIGBUS, shellCrashHandler);
#endif
taosSetSignal(SIGABRT, shellCrashHandler);
taosSetSignal(SIGFPE, shellCrashHandler);
taosSetSignal(SIGSEGV, shellCrashHandler);
#endif
if (shellCheckIntSize() != 0) {
return -1;