optimizations (#200)

* Use direct file read and direct buffer allocation for small IoBuffer materalization
* Reduce range of materialized data in CompositeBuffer reading
CompressedBuffer header reading often only need a small part and not the whole file
* reduce lock contention in IoBuffer::Materialize
* Reduce parsing of compressed headers
Validate header type at decompression
* faster CreateDirectories - start from leaf going up and recurse back
* optimized BufferHeader::IsValid
* Add ValidateCompressedHeader to use when we don't need the actual compressed data
Validate that we always get compressed data in CidStore::AddChunk
* changelog

5 files changed, 28 insertions, 19 deletions

diff --git a/zenstore/cidstore.cpp b/zenstore/cidstore.cpp
index 8b2797ce9..5a5116faf 100644
--- a/zenstore/cidstore.cpp
+++ b/zenstore/cidstore.cpp
@@ -23,14 +23,17 @@ struct CidStore::Impl
 
 	void Initialize(const CidStoreConfiguration& Config) { m_CasStore.Initialize(Config); }
 
-	CidStore::InsertResult AddChunk(const CompressedBuffer& ChunkData, CidStore::InsertMode Mode)
+	CidStore::InsertResult AddChunk(const IoBuffer& ChunkData, const IoHash& RawHash, CidStore::InsertMode Mode)
 	{
-		const IoHash DecompressedId = IoHash::FromBLAKE3(ChunkData.GetRawHash());
-		IoBuffer	 Payload		= ChunkData.GetCompressed().Flatten().AsIoBuffer();
-
+#ifndef NDEBUG
+		IoHash	 VerifyRawHash;
+		uint64_t _;
+		ZEN_ASSERT(CompressedBuffer::ValidateCompressedHeader(ChunkData, VerifyRawHash, _) && RawHash == VerifyRawHash);
+#endif	// NDEBUG
+		IoBuffer Payload(ChunkData);
 		Payload.SetContentType(ZenContentType::kCompressedBinary);
 
-		CasStore::InsertResult Result = m_CasStore.InsertChunk(Payload, DecompressedId, static_cast<CasStore::InsertMode>(Mode));
+		CasStore::InsertResult Result = m_CasStore.InsertChunk(Payload, RawHash, static_cast<CasStore::InsertMode>(Mode));
 
 		return {.New = Result.New};
 	}
@@ -78,9 +81,9 @@ CidStore::Initialize(const CidStoreConfiguration& Config)
 }
 
 CidStore::InsertResult
-CidStore::AddChunk(const CompressedBuffer& ChunkData, InsertMode Mode)
+CidStore::AddChunk(const IoBuffer& ChunkData, const IoHash& RawHash, InsertMode Mode)
 {
-	return m_Impl->AddChunk(ChunkData, Mode);
+	return m_Impl->AddChunk(ChunkData, RawHash, Mode);
 }
 
 IoBuffer
diff --git a/zenstore/compactcas.cpp b/zenstore/compactcas.cpp
index 7507a82f6..8679eb95e 100644
--- a/zenstore/compactcas.cpp
+++ b/zenstore/compactcas.cpp
@@ -269,9 +269,11 @@ CasContainerStrategy::Scrub(ScrubContext& Ctx)
 		}
 
 		IoBuffer Buffer(IoBuffer::Wrap, Data, Size);
-		if (CompressedBuffer Compressed = CompressedBuffer::FromCompressed(SharedBuffer(Buffer)); Compressed)
+		IoHash	 RawHash;
+		uint64_t RawSize;
+		if (CompressedBuffer::ValidateCompressedHeader(Buffer, RawHash, RawSize))
 		{
-			if (IoHash::FromBLAKE3(Compressed.GetRawHash()) != Hash)
+			if (RawHash != Hash)
 			{
 				// Hash mismatch
 				BadKeys.push_back(Hash);
@@ -295,10 +297,13 @@ CasContainerStrategy::Scrub(ScrubContext& Ctx)
 
 		const IoHash& Hash = ChunkIndexToChunkHash[ChunkIndex];
 		IoBuffer	  Buffer(IoBuffer::BorrowedFile, File.GetBasicFile().Handle(), Offset, Size);
+
+		IoHash	 RawHash;
+		uint64_t RawSize;
 		// TODO: Add API to verify compressed buffer without having to memorymap the whole file
-		if (CompressedBuffer Compressed = CompressedBuffer::FromCompressed(SharedBuffer(Buffer)); Compressed)
+		if (CompressedBuffer::ValidateCompressedHeader(Buffer, RawHash, RawSize))
 		{
-			if (IoHash::FromBLAKE3(Compressed.GetRawHash()) != Hash)
+			if (RawHash != Hash)
 			{
 				// Hash mismatch
 				BadKeys.push_back(Hash);
diff --git a/zenstore/filecas.cpp b/zenstore/filecas.cpp
index 38d7898cf..e67653e8a 100644
--- a/zenstore/filecas.cpp
+++ b/zenstore/filecas.cpp
@@ -519,9 +519,7 @@ FileCasStrategy::InsertChunk(const void* const ChunkData, const size_t ChunkSize
 	if (hRes == HRESULT_FROM_WIN32(ERROR_PATH_NOT_FOUND))
 	{
 		// Ensure parent directories exist and retry file creation
-
-		std::filesystem::create_directories(std::wstring_view(Name.ShardedPath.c_str(), Name.Shard2len));
-
+		CreateDirectories(std::wstring_view(Name.ShardedPath.c_str(), Name.Shard2len));
 		hRes = InternalCreateFile();
 	}
 
@@ -789,9 +787,11 @@ FileCasStrategy::Scrub(ScrubContext& Ctx)
 		ChunkBytes += Payload.FileSize();
 
 		IoBuffer Buffer(IoBuffer::BorrowedFile, Payload.Handle(), 0, Payload.FileSize());
-		if (CompressedBuffer Compressed = CompressedBuffer::FromCompressed(SharedBuffer(Buffer)); Compressed)
+		IoHash	 RawHash;
+		uint64_t RawSize;
+		if (CompressedBuffer::ValidateCompressedHeader(Buffer, RawHash, RawSize))
 		{
-			if (IoHash::FromBLAKE3(Compressed.GetRawHash()) != Hash)
+			if (RawHash != Hash)
 			{
 				// Hash mismatch
 				BadHashes.push_back(Hash);
diff --git a/zenstore/gc.cpp b/zenstore/gc.cpp
index 4094716ae..c50f59b64 100644
--- a/zenstore/gc.cpp
+++ b/zenstore/gc.cpp
@@ -917,7 +917,8 @@ TEST_CASE("gc.basic")
 	IoBuffer Chunk			 = CreateChunk(128);
 	auto	 CompressedChunk = Compress(Chunk);
 
-	const auto InsertResult = CidStore.AddChunk(CompressedChunk);
+	const auto InsertResult =
+		CidStore.AddChunk(CompressedChunk.GetCompressed().Flatten().AsIoBuffer(), IoHash::FromBLAKE3(CompressedChunk.DecodeRawHash()));
 	CHECK(InsertResult.New);
 
 	GcContext GcCtx(GcClock::Now() - std::chrono::hours(24));
@@ -926,7 +927,7 @@ TEST_CASE("gc.basic")
 	CidStore.Flush();
 	Gc.CollectGarbage(GcCtx);
 
-	CHECK(!CidStore.ContainsChunk(IoHash::FromBLAKE3(CompressedChunk.GetRawHash())));
+	CHECK(!CidStore.ContainsChunk(IoHash::FromBLAKE3(CompressedChunk.DecodeRawHash())));
 }
 
 TEST_CASE("gc.full")
diff --git a/zenstore/include/zenstore/cidstore.h b/zenstore/include/zenstore/cidstore.h
index e8984a83d..16ca78225 100644
--- a/zenstore/include/zenstore/cidstore.h
+++ b/zenstore/include/zenstore/cidstore.h
@@ -70,7 +70,7 @@ public:
 	};
 
 	void		 Initialize(const CidStoreConfiguration& Config);
-	InsertResult AddChunk(const CompressedBuffer& ChunkData, InsertMode Mode = InsertMode::kMayBeMovedInPlace);
+	InsertResult AddChunk(const IoBuffer& ChunkData, const IoHash& RawHash, InsertMode Mode = InsertMode::kMayBeMovedInPlace);
 	IoBuffer	 FindChunkByCid(const IoHash& DecompressedId);
 	bool		 ContainsChunk(const IoHash& DecompressedId);
 	void		 FilterChunks(HashKeySet& InOutChunks);