Merge branch 'devel' of github.com:arangodb/arangodb into devel

Documentation/Books/Users/HttpReplications/ReplicationApplier.mdpp

@@ -16,4 +16,8 @@ configuration of an ArangoDB database's replication applier.
 @startDocuBlock JSF_put_api_replication_applier_stop
 
 <!-- arangod/RestHandler/RestReplicationHandler.cpp -->
-@startDocuBlock JSF_get_api_replication_applier_state
+@startDocuBlock JSF_get_api_replication_applier_state
+
+<!-- arangod/RestHandler/RestReplicationHandler.cpp -->
+@startDocuBlock JSF_put_api_replication_makeSlave
+

arangod/VocBase/collection.cpp

@@ -37,6 +37,7 @@
 #include "Basics/JsonHelper.h"
 #include "Basics/logging.h"
 #include "Basics/tri-strings.h"
+#include "Basics/memory-map.h"
 #include "VocBase/document-collection.h"
 #include "VocBase/server.h"
 #include "VocBase/vocbase.h"
@@ -738,6 +739,11 @@ static bool IterateDatafilesVector (const TRI_vector_pointer_t* const files,
     if (! TRI_IterateDatafile(datafile, iterator, data)) {
       return false;
     }
+
+    if (datafile->isPhysical(datafile) && datafile->_isSealed) {
+      TRI_MMFileAdvise(datafile->_data, datafile->_maximalSize,
+                       TRI_MADVISE_RANDOM);
+    }
   }
 
   return true;

arangod/VocBase/compactor.cpp

@@ -37,6 +37,7 @@
 #include "Basics/files.h"
 #include "Basics/logging.h"
 #include "Basics/tri-strings.h"
+#include "Basics/memory-map.h"
 #include "Utils/transactions.h"
 #include "VocBase/document-collection.h"
 #include "VocBase/server.h"
@@ -763,6 +764,12 @@ static compaction_initial_context_t InitCompaction (TRI_document_collection_t* d
     compaction_info_t* compaction = static_cast<compaction_info_t*>(TRI_AtVector(compactions, i));
     TRI_datafile_t* df = compaction->_datafile;
 
+    // We will sequentially scan the logfile for collection:
+    if (df->isPhysical(df)) {
+      TRI_MMFileAdvise(df->_data, df->_maximalSize, TRI_MADVISE_SEQUENTIAL);
+      TRI_MMFileAdvise(df->_data, df->_maximalSize, TRI_MADVISE_WILLNEED);
+    }
+
     if (i == 0) {
       // extract and store fid
       context._fid = compaction->_datafile->_fid;
@@ -772,6 +779,10 @@ static compaction_initial_context_t InitCompaction (TRI_document_collection_t* d
 
     bool ok = TRI_IterateDatafile(df, CalculateSize, &context);
 
+    if (df->isPhysical(df)) {
+      TRI_MMFileAdvise(df->_data, df->_maximalSize, TRI_MADVISE_RANDOM);
+    }
+
     if (! ok) {
       context._failed = true;
       break;

arangod/VocBase/datafile.cpp

@@ -1404,6 +1404,10 @@ TRI_datafile_t* TRI_CreatePhysicalDatafile (char const* filename,
                fid,
                static_cast<char*>(data));
 
+  // Advise OS that sequential access is going to happen:
+  TRI_MMFileAdvise(datafile->_data, datafile->_maximalSize,
+                   TRI_MADVISE_SEQUENTIAL);
+
   return datafile;
 }
 
@@ -1893,6 +1897,12 @@ TRI_datafile_t* TRI_OpenDatafile (char const* filename,
     TRI_ProtectMMFile(datafile->_data, datafile->_maximalSize, PROT_READ | PROT_WRITE, datafile->_fd, &datafile->_mmHandle);
   }
 
+  // Advise on sequential use:
+  TRI_MMFileAdvise(datafile->_data, datafile->_maximalSize,
+                   TRI_MADVISE_SEQUENTIAL);
+  TRI_MMFileAdvise(datafile->_data, datafile->_maximalSize,
+                   TRI_MADVISE_WILLNEED);
+
   return datafile;
 }
 
@@ -2061,6 +2071,12 @@ int TRI_SealDatafile (TRI_datafile_t* datafile) {
     return datafile->_lastError;
   }
 
+  if (datafile->isPhysical(datafile)) {
+    // From now on we predict random access (until collection or compaction):
+    TRI_MMFileAdvise(datafile->_data, datafile->_maximalSize,
+                     TRI_MADVISE_RANDOM);
+  }
+
   return TRI_ERROR_NO_ERROR;
 }
 

arangod/VocBase/datafile.h

@@ -731,19 +731,19 @@ int TRI_SealDatafile (TRI_datafile_t* datafile) TRI_WARN_UNUSED_RESULT;
 bool TRI_RenameDatafile (TRI_datafile_t* datafile, char const* filename);
 
 ////////////////////////////////////////////////////////////////////////////////
-/// @brief truncates a datafile and seals it
+/// @brief truncates a datafile and seals it, only called by arango-dfdd
 ////////////////////////////////////////////////////////////////////////////////
 
 int TRI_TruncateDatafile (char const* path, TRI_voc_size_t position);
 
 ////////////////////////////////////////////////////////////////////////////////
-/// @brief try to repair a datafile
+/// @brief try to repair a datafile, only called by arango-dfdd
 ////////////////////////////////////////////////////////////////////////////////
 
 bool TRI_TryRepairDatafile (char const* path);
 
 ////////////////////////////////////////////////////////////////////////////////
-/// @brief returns information about the datafile
+/// @brief returns information about the datafile, only called by arango-dfdd
 ////////////////////////////////////////////////////////////////////////////////
 
 TRI_df_scan_t TRI_ScanDatafile (char const* path);

arangod/Wal/CollectorThread.cpp

@@ -32,6 +32,7 @@
 #include "Basics/MutexLocker.h"
 #include "Basics/hashes.h"
 #include "Basics/logging.h"
+#include "Basics/memory-map.h"
 #include "Basics/ConditionLocker.h"
 #include "Basics/Exceptions.h"
 #include "Utils/CollectionGuard.h"
@@ -739,6 +740,12 @@ int CollectorThread::collect (Logfile* logfile) {
 
   TRI_ASSERT(df != nullptr);
 
+  // We will sequentially scan the logfile for collection:
+  TRI_MMFileAdvise(df->_data, df->_maximalSize, TRI_MADVISE_SEQUENTIAL);
+  TRI_MMFileAdvise(df->_data, df->_maximalSize, TRI_MADVISE_WILLNEED);
+  TRI_DEFER (TRI_MMFileAdvise(df->_data, df->_maximalSize,
+                              TRI_MADVISE_RANDOM));
+
   // create a state for the collector, beginning with the list of failed transactions
   CollectorState state;
   state.failedTransactions = _logfileManager->getFailedTransactions();

arangod/Wal/LogfileManager.cpp

@@ -39,6 +39,7 @@
 #include "Basics/ReadLocker.h"
 #include "Basics/StringUtils.h"
 #include "Basics/WriteLocker.h"
+#include "Basics/memory-map.h"
 #include "VocBase/server.h"
 #include "Wal/AllocatorThread.h"
 #include "Wal/CollectorThread.h"
@@ -2064,6 +2065,12 @@ int LogfileManager::inspectLogfiles () {
               (unsigned long long) logfile->df()->_tickMin, 
               (unsigned long long) logfile->df()->_tickMax);
 
+    if (logfile->status() == Logfile::StatusType::SEALED) {
+      // If it is sealed, switch to random access:
+      TRI_datafile_t* df = logfile->df();
+      TRI_MMFileAdvise(df->_data, df->_maximalSize, TRI_MADVISE_RANDOM);
+    }
+
     {
       MUTEX_LOCKER(_idLock);
       if (logfile->status() == Logfile::StatusType::SEALED &&

arangod/Wal/RecoverState.cpp

@@ -32,6 +32,7 @@
 #include "Basics/conversions.h"
 #include "Basics/files.h"
 #include "Basics/Exceptions.h"
+#include "Basics/memory-map.h"
 #include "VocBase/collection.h"
 #include "VocBase/replication-applier.h"
 #include "VocBase/VocShaper.h"
@@ -1355,11 +1356,21 @@ int RecoverState::replayLogfile (Logfile* logfile,
   LOG_INFO("replaying WAL logfile '%s' (%d of %d)", 
            logfile->filename().c_str(), number + 1, n);
 
+  // Advise on sequential use:
+  TRI_MMFileAdvise(logfile->df()->_data, logfile->df()->_maximalSize,
+                   TRI_MADVISE_SEQUENTIAL);
+  TRI_MMFileAdvise(logfile->df()->_data, logfile->df()->_maximalSize,
+                   TRI_MADVISE_WILLNEED);
+
   if (! TRI_IterateDatafile(logfile->df(), &RecoverState::ReplayMarker, static_cast<void*>(this))) {
     LOG_WARNING("WAL inspection failed when scanning logfile '%s'", logfile->filename().c_str());
     return TRI_ERROR_ARANGO_RECOVERY;
   }
 
+  // Advise on random access use:
+  TRI_MMFileAdvise(logfile->df()->_data, logfile->df()->_maximalSize,
+                   TRI_MADVISE_RANDOM);
+
   return TRI_ERROR_NO_ERROR;
 }
 

lib/Basics/AssocMulti.h

@@ -41,6 +41,7 @@
 #include "Basics/logging.h"
 #include "Basics/Mutex.h"
 #include "Basics/MutexLocker.h"
+#include "Basics/memory-map.h"
 
 namespace triagens {
   namespace basics {
@@ -243,6 +244,17 @@ namespace triagens {
               // may fail...
               b._table = new EntryType[b._nrAlloc];
 
+#ifdef __linux__
+              if (b._nrAlloc > 1000000) {
+                uintptr_t mem = reinterpret_cast<uintptr_t>(b._table);
+                uintptr_t pageSize = getpagesize();
+                mem = (mem / pageSize) * pageSize;
+                void* memptr = reinterpret_cast<void*>(mem);
+                TRI_MMFileAdvise(memptr, b._nrAlloc * sizeof(EntryType),
+                                 TRI_MADVISE_RANDOM);
+              }
+#endif
+
               for (IndexType i = 0; i < b._nrAlloc; i++) {
                 invalidateEntry(b, i);
               }
@@ -1032,8 +1044,8 @@ namespace triagens {
 /// @brief resize the array
 ////////////////////////////////////////////////////////////////////////////////
 
-        int resize (IndexType size) throw() {
-          size /= static_cast<IndexType>(_buckets.size());
+        int resize (size_t size) throw() {
+          size /= _buckets.size();
           for (auto& b : _buckets) {
             if (2 * (2 * size + 1) < 3 * b._nrUsed) {
               return TRI_ERROR_BAD_PARAMETER;
@@ -1103,7 +1115,7 @@ namespace triagens {
 /// @brief resize the array, internal method
 ////////////////////////////////////////////////////////////////////////////////
 
-        void resizeInternal (Bucket& b, IndexType size) {
+        void resizeInternal (Bucket& b, size_t size) {
           LOG_ACTION("index-resize %s, target size: %llu",
                      _contextCallback().c_str(),
                      (unsigned long long) size);
@@ -1112,9 +1124,20 @@ namespace triagens {
           EntryType* oldTable = b._table;
           IndexType oldAlloc = b._nrAlloc;
 
-          b._nrAlloc = static_cast<IndexType>(TRI_NearPrime(size));
+          b._nrAlloc = static_cast<IndexType>(TRI_NearPrime(static_cast<uint64_t>(size)));
           try {
             b._table = new EntryType[b._nrAlloc];
+#ifdef __linux__
+            if (b._nrAlloc > 1000000) {
+              uintptr_t mem = reinterpret_cast<uintptr_t>(b._table);
+              uintptr_t pageSize = getpagesize();
+              mem = (mem / pageSize) * pageSize;
+              void* memptr = reinterpret_cast<void*>(mem);
+              TRI_MMFileAdvise(memptr, b._nrAlloc * sizeof(EntryType),
+                               TRI_MADVISE_RANDOM);
+            }
+#endif
+
             IndexType i;
             for (i = 0; i < b._nrAlloc; i++) {
               invalidateEntry(b, i);

lib/Basics/AttributeNameParser.h

@@ -47,14 +47,20 @@ namespace triagens {
 ////////////////////////////////////////////////////////////////////////////////
 
     struct AttributeName {
-      std::string const name;
-      bool const shouldExpand;
+      std::string name;
+      bool shouldExpand;
 
       AttributeName (std::string const& pName)
-        : name(pName), shouldExpand(false) { };
+        : name(pName), shouldExpand(false) { 
+      }
 
       AttributeName (std::string const& pName, bool pExpand)
-        : name(pName), shouldExpand(pExpand) { };
+        : name(pName), shouldExpand(pExpand) { 
+      }
+
+      AttributeName (const AttributeName& b) 
+        : name(b.name), shouldExpand(b.shouldExpand) {
+      }
 
       bool operator== (const AttributeName& b) const {
         return name == b.name && shouldExpand == b.shouldExpand;

lib/Basics/memory-map-posix.cpp

@@ -151,6 +151,30 @@ int TRI_ProtectMMFile (void* memoryAddress,
   return TRI_ERROR_SYS_ERROR;
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief gives hints about upcoming sequential memory usage
+////////////////////////////////////////////////////////////////////////////////
+
+int TRI_MMFileAdvise (void* memoryAddress, size_t numOfBytes, int advice) {
+#ifdef __linux__
+  LOG_DEBUG("Doing madvise %d for %lu length %lu", advice,
+            (uint64_t) memoryAddress, numOfBytes);
+  char buffer[256];
+  int res = madvise(memoryAddress, numOfBytes, advice);
+  if (res == 0) {
+    return TRI_ERROR_NO_ERROR;
+  }
+  else {
+    char* p = strerror_r(errno, buffer, 256);
+    LOG_INFO("madvise %d for %lu length %lu failed with: %s ",
+             advice, (uint64_t) memoryAddress, numOfBytes, p);
+    return TRI_ERROR_INTERNAL;
+  }
+#else
+  return TRI_ERROR_NO_ERROR;
+#endif
+}
+
 #endif
 
 // -----------------------------------------------------------------------------

lib/Basics/memory-map-posix.h

@@ -49,6 +49,22 @@
 #define TRI_MMAP_ANONYMOUS MAP_ANON
 #endif
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief constants for TRI_MMFileAdvise
+////////////////////////////////////////////////////////////////////////////////
+
+#ifdef __linux__
+#define TRI_MADVISE_SEQUENTIAL MADV_SEQUENTIAL
+#define TRI_MADVISE_RANDOM     MADV_RANDOM
+#define TRI_MADVISE_WILLNEED   MADV_WILLNEED
+#define TRI_MADVISE_DONTNEED   MADV_DONTNEED
+#else
+#define TRI_MADVISE_SEQUENTIAL 0
+#define TRI_MADVISE_RANDOM     0
+#define TRI_MADVISE_WILLNEED   0
+#define TRI_MADVISE_DONTNEED   0
+#endif
+
 #endif
 
 #endif

lib/Basics/memory-map-win32.cpp

@@ -335,6 +335,15 @@ int TRI_ProtectMMFile (void* memoryAddress,
   return TRI_ERROR_NO_ERROR;
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief gives hints about upcoming sequential memory usage
+////////////////////////////////////////////////////////////////////////////////
+
+int TRI_MMFileAdvise (void*, size_t, int) {
+  // Not on Windows
+  return TRI_ERROR_NO_ERROR;
+}
+
 
 #endif
 

lib/Basics/memory-map-win32.h

@@ -77,6 +77,15 @@
 
 #endif
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief constants for TRI_MMFileAdvise, not on Windows
+////////////////////////////////////////////////////////////////////////////////
+
+#define TRI_MADVISE_SEQUENTIAL 0
+#define TRI_MADVISE_RANDOM     0
+#define TRI_MADVISE_WILLNEED   0
+#define TRI_MADVISE_DONTNEED   0
+
 #endif
 
 // -----------------------------------------------------------------------------

lib/Basics/memory-map.h

@@ -103,6 +103,12 @@ int TRI_ProtectMMFile (void* memoryAddress,
                        int fileDescriptor,
                        void** mmHandle);
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief gives hints about upcoming memory usage
+////////////////////////////////////////////////////////////////////////////////
+
+int TRI_MMFileAdvise (void* memoryAddress, size_t numOfBytes, int advice);
+
 #endif
 
 // -----------------------------------------------------------------------------