mmap tarfiles directly when mmapping is enabled, instead of copying data into separate mmapped block

Co-authored-by: Kajari Ghosh <ghoshkaj@gmail.com>

src/engine/datafacade/mmap_memory_allocator.cpp

@@ -1,5 +1,6 @@
 #include "engine/datafacade/mmap_memory_allocator.hpp"
 
+#include "storage/block.hpp"
 #include "storage/io.hpp"
 #include "storage/serialization.hpp"
 #include "storage/storage.hpp"
@@ -7,7 +8,7 @@
 #include "util/log.hpp"
 #include "util/mmap_file.hpp"
 
-#include "boost/assert.hpp"
+#include <boost/assert.hpp>
 
 namespace osrm
 {
@@ -16,46 +17,50 @@ namespace engine
 namespace datafacade
 {
 
-MMapMemoryAllocator::MMapMemoryAllocator(const storage::StorageConfig &config,
-                                         const boost::filesystem::path &memory_file)
+MMapMemoryAllocator::MMapMemoryAllocator(const storage::StorageConfig &config)
 {
     storage::Storage storage(config);
+    std::vector<storage::SharedDataIndex::AllocatedRegion> allocated_regions;
 
-    if (!boost::filesystem::exists(memory_file))
     {
-        storage::DataLayout initial_layout;
-        storage.PopulateStaticLayout(initial_layout);
-        storage.PopulateUpdatableLayout(initial_layout);
+        std::unique_ptr<storage::BaseDataLayout> fake_layout =
+            std::make_unique<storage::TarDataLayout>();
 
-        auto data_size = initial_layout.GetSizeOfLayout();
+        // Convert the boost::filesystem::path object into a plain string
+        // that's stored as a member of this allocator object
+        rtree_filename = storage.PopulateLayoutWithRTree(*fake_layout);
 
-        storage::io::BufferWriter writer;
-        storage::serialization::write(writer, initial_layout);
-        auto encoded_layout = writer.GetBuffer();
-
-        auto total_size = data_size + encoded_layout.size();
-
-        mapped_memory = util::mmapFile<char>(memory_file, mapped_memory_file, total_size);
-
-        std::copy(encoded_layout.begin(), encoded_layout.end(), mapped_memory.data());
-
-        index = storage::SharedDataIndex(
-            {{mapped_memory.data() + encoded_layout.size(), std::move(initial_layout)}});
-
-        storage.PopulateStaticData(index);
-        storage.PopulateUpdatableData(index);
+        // Now, we add one more AllocatedRegion, with it's start address as the start
+        // of the rtree_filename string we've saved.  In the fake_layout, we've
+        // stated that the data is at offset 0, which is where the string starts
+        // at it's own memory address.
+        // The syntax &(rtree_filename[0]) gets the memory address of the first char.
+        // We can't use the convenient `.data()` or `.c_str()` methods, because
+        // prior to C++17 (which we're not using), those return a `const char *`,
+        // which isn't compatible with the `char *` that AllocatedRegion expects
+        // for it's memory_ptr
+        allocated_regions.push_back({&(rtree_filename[0]), std::move(fake_layout)});
     }
-    else
+
+    auto files = storage.GetStaticFiles();
+    auto updatable_files = storage.GetUpdatableFiles();
+    files.insert(files.end(), updatable_files.begin(), updatable_files.end());
+
+    for (const auto &file : files)
     {
-        mapped_memory = util::mmapFile<char>(memory_file, mapped_memory_file);
-
-        storage::DataLayout layout;
-        storage::io::BufferReader reader(mapped_memory.data(), mapped_memory.size());
-        storage::serialization::read(reader, layout);
-        auto layout_size = reader.GetPosition();
-
-        index = storage::SharedDataIndex({{mapped_memory.data() + layout_size, std::move(layout)}});
+        if (boost::filesystem::exists(file.second))
+        {
+            std::unique_ptr<storage::BaseDataLayout> layout =
+                std::make_unique<storage::TarDataLayout>();
+            boost::iostreams::mapped_file mapped_memory_file;
+            util::mmapFile<char>(file.second, mapped_memory_file);
+            mapped_memory_files.push_back(std::move(mapped_memory_file));
+            storage::populateLayoutFromFile(file.second, *layout);
+            allocated_regions.push_back({mapped_memory_file.data(), std::move(layout)});
+        }
     }
+
+    index = storage::SharedDataIndex{std::move(allocated_regions)};
 }
 
 MMapMemoryAllocator::~MMapMemoryAllocator() {}

src/engine/datafacade/process_memory_allocator.cpp

@@ -15,14 +15,20 @@ ProcessMemoryAllocator::ProcessMemoryAllocator(const storage::StorageConfig &con
     storage::Storage storage(config);
 
     // Calculate the layout/size of the memory block
-    storage::DataLayout layout;
-    storage.PopulateStaticLayout(layout);
-    storage.PopulateUpdatableLayout(layout);
+    auto static_files = storage.GetStaticFiles();
+    auto updatable_files = storage.GetUpdatableFiles();
+    std::unique_ptr<storage::BaseDataLayout> layout =
+        std::make_unique<storage::ContiguousDataLayout>();
+    storage.PopulateLayoutWithRTree(*layout);
+    storage.PopulateLayout(*layout, static_files);
+    storage.PopulateLayout(*layout, updatable_files);
 
     // Allocate the memory block, then load data from files into it
-    internal_memory = std::make_unique<char[]>(layout.GetSizeOfLayout());
+    internal_memory = std::make_unique<char[]>(layout->GetSizeOfLayout());
 
-    index = storage::SharedDataIndex({{internal_memory.get(), std::move(layout)}});
+    std::vector<storage::SharedDataIndex::AllocatedRegion> regions;
+    regions.push_back({internal_memory.get(), std::move(layout)});
+    index = {std::move(regions)};
 
     storage.PopulateStaticData(index);
     storage.PopulateUpdatableData(index);

src/engine/datafacade/shared_memory_allocator.cpp

@@ -25,8 +25,9 @@ SharedMemoryAllocator::SharedMemoryAllocator(
         auto mem = storage::makeSharedMemory(shm_key);
 
         storage::io::BufferReader reader(reinterpret_cast<char *>(mem->Ptr()), mem->Size());
-        storage::DataLayout layout;
-        storage::serialization::read(reader, layout);
+        std::unique_ptr<storage::BaseDataLayout> layout =
+            std::make_unique<storage::ContiguousDataLayout>();
+        storage::serialization::read(reader, *layout);
         auto layout_size = reader.GetPosition();
 
         regions.push_back({reinterpret_cast<char *>(mem->Ptr()) + layout_size, std::move(layout)});