189 - Generate a dump when the size of the generation exceeds thresho…

…ld (#197)

* 189 - Generate a dump when the size of the generation exceeds threshold

* Add another GC.Collect for test

README.md

@@ -25,7 +25,7 @@ procdump [-n Count]
         [-s Seconds]
         [-c|-cl CPU_Usage]
         [-m|-ml Commit_Usage1[,Commit_Usage2,...]]
-        [-gcm Memory_Usage1[,Memory_Usage2...]]
+        [-gcm [<GCGeneration>: | LOH: | POH:]Memory_Usage1[,Memory_Usage2...]]
         [-gcgen Generation]
         [-tc Thread_Threshold]
         [-fc FileDescriptor_Threshold]
@@ -46,7 +46,7 @@ Options:
    -cl     CPU threshold below which to create a dump of the process.
    -m      Memory commit threshold(s) (MB) above which to create dumps.
    -ml     Memory commit threshold(s) (MB) below which to create dumps.
-   -gcm    [.NET] GC memory threshold(s) (MB) above which to create dumps.
+   -gcm    [.NET] GC memory threshold(s) (MB) above which to create dumps for the specified generation or heap (default is total .NET memory usage).
    -gcgen  [.NET] Create dump when the garbage collection of the specified generation starts and finishes.
    -tc     Thread count threshold above which to create a dump of the process.
    -fc     File descriptor count threshold above which to create a dump of the process.
@@ -94,10 +94,18 @@ The following will create a core dump when memory usage is >= 100 MB followed by
 ```
 sudo procdump -m 100,200 1234
 ```
-The following will create a core dump when .NET memory usage is >= 100 MB followed by another dump when memory usage is >= 200MB.
+The following will create a core dump when the total .NET memory usage is >= 100 MB followed by another dump when memory usage is >= 200MB.
 ```
 sudo procdump -gcm 100,200 1234
 ```
+The following will create a core dump when .NET memory usage for generation 1 is >= 1 MB followed by another dump when memory usage is >= 2MB.
+```
+sudo procdump -gcm 1:1,2 1234
+```
+The following will create a core dump when .NET Large Object Heap memory usage is >= 100 MB followed by another dump when memory usage is >= 200MB.
+```
+sudo procdump -gcm LOH:100,200 1234
+```
 The following will create a core dump at the start and end of a .NET generation 1 garbage collection.
 ```
 sudo procdump -gcgen 1

include/Monitor.h

@@ -33,8 +33,7 @@ bool BeginMonitoring(struct ProcDumpConfiguration *self);
 bool MonitorDotNet(struct ProcDumpConfiguration *self);
 char* GetThresholds(struct ProcDumpConfiguration *self);
 char* GetClientData(struct ProcDumpConfiguration *self, char* fullDumpPath);
-char* GetClientDataInt(enum TriggerType triggerType, char* path, int value);
-char* GetClientDataString(enum TriggerType triggerType, char* path, char* value);
+char* GetClientDataHelper(enum TriggerType triggerType, char* path, const char* format, ...);
 bool ExitProcessMonitor(struct ProcDumpConfiguration* config, pthread_t processMonitor);
 
 // Monitor worker threads

include/ProfilerCommon.h

@@ -9,6 +9,9 @@
 #ifndef PROFILERCOMMON_H
 #define PROFILERCOMMON_H
 
+#define CUMULATIVE_GC_SIZE  2008
+#define MAX_GC_GEN  2
+
 enum TriggerType
 {
     Processor,

procdump.1

@@ -7,7 +7,7 @@ procdump [-n Count]
         [-s Seconds]
         [-c|-cl CPU_Usage]
         [-m|-ml Commit_Usage1[,Commit_Usage2,...]]
-        [-gcm Memory_Usage1[,Memory_Usage2...]]
+        [-gcm [<GCGeneration>: | LOH: | POH:]Memory_Usage1[,Memory_Usage2...]]
         [-gcgen Generation]
         [-tc Thread_Threshold]
         [-fc FileDescriptor_Threshold]
@@ -28,7 +28,7 @@ Options:
    -cl     CPU threshold below which to create a dump of the process.
    -m      Memory commit thresholds (MB) above which to create dumps.
    -ml     Memory commit thresholds (MB) below which to create dumps.
-   -gcm    [.NET] GC memory threshold(s) (MB) above which to create dumps.
+   -gcm    [.NET] GC memory threshold(s) (MB) above which to create dumps for the specified generation or heap (default is total .NET memory usage).
    -gcgen  [.NET] Create dump when the garbage collection of the specified generation starts and finishes.
    -tc     Thread count threshold above which to create a dump of the process.
    -fc     File descriptor count threshold above which to create a dump of the process.

profiler/inc/ProcDumpProfiler.h

@@ -111,7 +111,6 @@ class CorProfiler : public ICorProfilerCallback8
     pthread_mutex_t endDumpCondition;
     enum TriggerType triggerType;
     int currentThresholdIndex;
-    bool gen2Collection;
     int gcGeneration;
     bool gcGenStarted;
 
@@ -129,7 +128,7 @@ class CorProfiler : public ICorProfilerCallback8
     int send_all(int socket, void* buffer, size_t length);
     int recv_all(int socket, void* buffer, size_t length);
     bool WildcardSearch(WCHAR*, WCHAR*);
-    u_int64_t GetGCHeapSize();
+    u_int64_t GetGCHeapSize(int generation);
     bool WriteDumpHelper(std::string dumpName);
     bool IsHighPerfBasicGC();
 

profiler/src/ProcDumpProfiler.cpp

@@ -207,7 +207,7 @@ bool CorProfiler::IsHighPerfBasicGC()
 //------------------------------------------------------------------------------------------------------------------------------------------------------
 CorProfiler::CorProfiler() :
     refCount(0), corProfilerInfo8(nullptr), corProfilerInfo3(nullptr), corProfilerInfo(nullptr),
-    procDumpPid(0), currentThresholdIndex(0), gen2Collection(false), gcGeneration(-1), gcGenStarted(false)
+    procDumpPid(0), currentThresholdIndex(0), gcGeneration(-1), gcGenStarted(false)
 {
     // Configure logging
     el::Loggers::reconfigureAllLoggers(el::ConfigurationType::Filename, LOG_FILE);
@@ -346,13 +346,13 @@ WCHAR* CorProfiler::GetUint16(char* buffer)
 // Syntax of client data: <trigger_type><fullpathtodumplocation>;...
 //
 //      DOTNET_EXCEPTION_TRIGGER;<fullpathtodumplocation>;<pidofprocdump>;<exception>:<numdumps>;<exception>:<numdumps>,...
-//      DOTNET_GC_THRESHOLD_TRIGGER;<fullpathtodumplocation>;<pidofprocdump>;Threshold1;Threshold2,...
+//      DOTNET_GC_THRESHOLD_TRIGGER;<fullpathtodumplocation>;<pidofprocdump>;[<generation> | 3(LOH) | 4(POH) | 2008 (total mem)];Threshold1;Threshold2,...
 //      DOTNET_GC_GEN_TRIGGER;<fullpathtodumplocation>;<pidofprocdump>;GCGeneration
 //
 //------------------------------------------------------------------------------------------------------------------------------------------------------
 bool CorProfiler::ParseClientData(char* clientData)
 {
-    LOG(TRACE) << "CorProfiler::ParseClientData: Enter";
+    LOG(TRACE) << "CorProfiler::ParseClientData: Enter clientData = " << clientData;
     std::stringstream clientDataStream(clientData);
     std::string segment;
     std::vector<std::string> dataList;
@@ -415,7 +415,16 @@ bool CorProfiler::ParseClientData(char* clientData)
         else if (triggerType == GCThreshold)
         {
             // GC threshold list
-            gcMemoryThresholdMonitorList.push_back(std::stoi(dataItem) << 20);
+            if(i == 3)
+            {
+                // first element is the generation
+                gcGeneration = std::stoi(dataItem);
+                i++;
+            }
+            else
+            {
+                gcMemoryThresholdMonitorList.push_back(std::stoi(dataItem) << 20);
+            }
         }
         else if (triggerType == GCGeneration)
         {
@@ -442,7 +451,14 @@ bool CorProfiler::ParseClientData(char* clientData)
 
     if(gcGeneration != -1)
     {
-        LOG(TRACE) << "CorProfiler::ParseClientData:GCGeneration  " << gcGeneration;
+        if( gcGeneration == CUMULATIVE_GC_SIZE)
+        {
+            LOG(TRACE) << "CorProfiler::ParseClientData:GCGeneration " << "Cumulative";
+        }
+        else
+        {
+            LOG(TRACE) << "CorProfiler::ParseClientData:GCGeneration " << gcGeneration;
+        }
     }
 
     LOG(TRACE) << "CorProfiler::ParseClientData: Exit";
@@ -720,7 +736,7 @@ std::string CorProfiler::GetDumpName(u_int16_t dumpCount, std::string name)
 //------------------------------------------------------------------------------------------------------------------------------------------------------
 // CorProfiler::GetGCHeapSize
 //------------------------------------------------------------------------------------------------------------------------------------------------------
-uint64_t CorProfiler::GetGCHeapSize()
+uint64_t CorProfiler::GetGCHeapSize(int generation)
 {
     LOG(TRACE) << "CorProfiler::GetGCHeapSize: Enter";
     uint64_t gcHeapSize = 0;
@@ -766,7 +782,17 @@ uint64_t CorProfiler::GetGCHeapSize()
 
     for (int i = nObjectRanges - 1; i >= 0; i--)
     {
-        gcHeapSize += pObjectRanges[i].rangeLength;
+        // Uncomment this to help track down .NET memory usage while debugging.
+        //LOG(TRACE) << "Range Len: " << pObjectRanges[i].rangeLength << " Gen: " << pObjectRanges[i].generation;
+
+        if(generation == CUMULATIVE_GC_SIZE)
+        {
+            gcHeapSize += pObjectRanges[i].rangeLength;
+        }
+        else if(pObjectRanges[i].generation == generation)
+        {
+            gcHeapSize += pObjectRanges[i].rangeLength;
+        }
     }
 
     if(fHeapAlloc == true)
@@ -827,27 +853,29 @@ HRESULT STDMETHODCALLTYPE CorProfiler::GarbageCollectionStarted(int cGenerations
 {
     LOG(TRACE) << "CorProfiler::GarbageCollectionStarted: Enter";
 
-    if(gcMemoryThresholdMonitorList.size() > 0 && generationCollected[2] == true)
+    if(gcGeneration != -1 &&
+       gcGenStarted == false &&
+       gcGeneration == CUMULATIVE_GC_SIZE ||
+       (gcGeneration < cGenerations &&
+       generationCollected[gcGeneration] == true))
     {
-        // GC memory threshold dump
-        gen2Collection = true;
-    }
-    else if(gcGeneration != -1 && gcGenStarted == false && gcGeneration < cGenerations && generationCollected[gcGeneration] == true)
-    {
-        // GC Generation dump
-        LOG(TRACE) << "CorProfiler::GarbageCollectionStarted: Dump on generation: " << gcGeneration << " and cGenerations = " << cGenerations;
-        std::string dump = GetDumpName(1, convertString<std::string,std::wstring>(L"gc_gen"));
-        if(WriteDumpHelper(dump) == false)
+        gcGenStarted = true;
+
+        if(gcMemoryThresholdMonitorList.size() == 0)
         {
-            SendCatastrophicFailureStatus();
-            return E_FAIL;
+            // GC Generation dump
+            LOG(TRACE) << "CorProfiler::GarbageCollectionStarted: Dump on generation: " << gcGeneration << " and cGenerations = " << cGenerations;
+            std::string dump = GetDumpName(1, convertString<std::string,std::wstring>(L"gc_gen"));
+            if(WriteDumpHelper(dump) == false)
+            {
+                SendCatastrophicFailureStatus();
+                return E_FAIL;
+            }
         }
-
-        gcGenStarted = true;
     }
     else
     {
-        LOG(TRACE) << "CorProfiler::GarbageCollectionStarted: Invalid trigger data, trigger = " << triggerType << " cGenerations " << cGenerations;
+        LOG(TRACE) << "CorProfiler::GarbageCollectionStarted: Trigger = " << triggerType << " cGenerations " << cGenerations << " gcGeneration " << gcGeneration << " threshold size " << gcMemoryThresholdMonitorList.size();
     }
 
     LOG(TRACE) << "CorProfiler::GarbageCollectionStarted: Exit";
@@ -861,12 +889,24 @@ HRESULT STDMETHODCALLTYPE CorProfiler::GarbageCollectionFinished()
 {
     LOG(TRACE) << "CorProfiler::GarbageCollectionFinished: Enter";
 
-    if(gen2Collection == true)
+    if(gcGenStarted == true && gcMemoryThresholdMonitorList.size() > 0)
     {
-        // During a GC threshold trigger, we only want to check heap sizes and thresholds after a gen2 collection
-        gen2Collection = false;
-        uint64_t heapSize = GetGCHeapSize();
-        LOG(TRACE) << "CorProfiler::GarbageCollectionFinished: Total heap size " << heapSize;
+        // During a GC threshold trigger, we only want to check heap sizes and thresholds after the gen collection
+        uint64_t heapSize = 0;
+        gcGenStarted = false;
+
+        if(gcGeneration == CUMULATIVE_GC_SIZE)
+        {
+            // If a generation was not explicitly specified on the command line (for example: -gcm 10,20,30) we want to get _all_ the memory of
+            // the managed heap and hence we add up all generations, LOH and POH.
+            heapSize += GetGCHeapSize(CUMULATIVE_GC_SIZE);
+            LOG(TRACE) << "CorProfiler::GarbageCollectionFinished: Cumulative heap size " << heapSize;
+        }
+        else
+        {
+            heapSize = GetGCHeapSize(gcGeneration);
+            LOG(TRACE) << "CorProfiler::GarbageCollectionFinished: Generation  " << gcGeneration << " heap size " << heapSize;
+        }
 
         if(currentThresholdIndex < gcMemoryThresholdMonitorList.size() && heapSize >= gcMemoryThresholdMonitorList[currentThresholdIndex])
         {
@@ -894,7 +934,7 @@ HRESULT STDMETHODCALLTYPE CorProfiler::GarbageCollectionFinished()
             }
         }
     }
-    else if(gcGenStarted == true)
+    else if(gcGenStarted == true && gcMemoryThresholdMonitorList.size() == 0)
     {
         // GC Generation dump
         LOG(TRACE) << "CorProfiler::GarbageCollectionFinished: Dump on generation: " << gcGeneration;

src/Monitor.c

@@ -1381,7 +1381,7 @@ char* GetClientData(struct ProcDumpConfiguration *self, char* fullDumpPath)
             return NULL;
         }
 
-        clientData = GetClientDataString(Exception, fullDumpPath, exceptionFilter);
+        clientData = GetClientDataHelper(Exception, fullDumpPath, "%s", exceptionFilter);
         if(clientData == NULL)
         {
             Trace("GetClientData: Failed to get client data (-e).");
@@ -1390,25 +1390,25 @@ char* GetClientData(struct ProcDumpConfiguration *self, char* fullDumpPath)
     }
     else if (self->bMonitoringGCMemory)
     {
-        // GC Memory trigger (-gcm);<fullpathtodumplocation>;<pidofprocdump>;Threshold1;Threshold2,...
+        // GC Memory trigger (-gcm);<fullpathtodumplocation>;<pidofprocdump>;Generation:Threshold1;Threshold2,...
         thresholds = GetThresholds(self);
         if(thresholds == NULL)
         {
             Trace("GetClientData: Failed to get thresholds.");
             return NULL;
         }
 
-        clientData = GetClientDataString(GCThreshold, fullDumpPath, thresholds);
+        clientData = GetClientDataHelper(GCThreshold, fullDumpPath, "%d;%s", self->DumpGCGeneration == -1 ? CUMULATIVE_GC_SIZE : self->DumpGCGeneration, thresholds);
         if(clientData == NULL)
         {
             Trace("GetClientData: Failed to get client data (-gcm).");
             return NULL;
         }
     }
-    else if(self->DumpGCGeneration)
+    else if(self->DumpGCGeneration != -1 && self->MemoryThreshold == NULL)
     {
         // GC Generation (-gcgen);<fullpathtodumplocation>;<pidofprocdump>;GCGeneration
-        clientData = GetClientDataInt(GCGeneration, fullDumpPath, self->DumpGCGeneration);
+        clientData = GetClientDataHelper(GCGeneration, fullDumpPath, "%d", self->DumpGCGeneration);
         if(clientData == NULL)
         {
             Trace("GetClientData: Failed to get client data (-gcgen).");
@@ -1427,51 +1427,36 @@ char* GetClientData(struct ProcDumpConfiguration *self, char* fullDumpPath)
 
 //-------------------------------------------------------------------------------------
 //
-// GetClientDataString
+// GetClientDataHelper
 //
-// Helper that fetches client data where the value is an 'char*'.
+// Helper that fetches client data based on format specified.
 //
 //-------------------------------------------------------------------------------------
-char* GetClientDataString(enum TriggerType triggerType, char* path, char* value)
+char* GetClientDataHelper(enum TriggerType triggerType, char* path, const char* format, ...)
 {
     unsigned int clientDataSize = 0;
+    unsigned int clientDataPrefixSize = 0;
     char* clientData = NULL;
 
-    clientDataSize = snprintf(NULL, 0, "%d;%s;%d;%s", triggerType, path, getpid(), value) + 1;
-    clientData = malloc(clientDataSize);
-    if(clientData == NULL)
-    {
-        Trace("GetClientDataString: Failed to allocate memory for client data.");
-        return NULL;
-    }
-
-    sprintf(clientData, "%d;%s;%d;%s", triggerType, path, getpid(), value);
-
-    return clientData;
-}
-
-//-------------------------------------------------------------------------------------
-//
-// GetClientDataInt
-//
-// Helper that fetches client data where the value is an 'int'.
-//
-//-------------------------------------------------------------------------------------
-char* GetClientDataInt(enum TriggerType triggerType, char* path, int value)
-{
-    unsigned int clientDataSize = 0;
-    char* clientData = NULL;
+    va_list args;
+    va_start(args, format);
 
-    clientDataSize = snprintf(NULL, 0, "%d;%s;%d;%d", triggerType, path, getpid(), value) + 1;
+    clientDataPrefixSize = snprintf(NULL, 0, "%d;%s;%d;", triggerType, path, getpid());
+    va_list args_copy;
+    va_copy(args_copy, args);
+    clientDataSize = clientDataPrefixSize + vsnprintf(NULL, 0, format, args_copy) + 1;
+    va_end(args_copy);
     clientData = malloc(clientDataSize);
     if(clientData == NULL)
     {
-        Trace("GetClientDataInt: Failed to allocate memory for client data.");
+        Trace("GetClientDataHelper: Failed to allocate memory for client data.");
         return NULL;
     }
 
-    sprintf(clientData, "%d;%s;%d;%d", triggerType, path, getpid(), value);
+    sprintf(clientData, "%d;%s;%d;", triggerType, path, getpid());
+    vsprintf(clientData+clientDataPrefixSize, format, args);
 
+    va_end(args);
     return clientData;
 }