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

1 files changed, 25 insertions, 22 deletions

diff --git a/zenserver/cache/structuredcachestore.cpp b/zenserver/cache/structuredcachestore.cpp
index 1f48aaebe..75f845cbf 100644
--- a/zenserver/cache/structuredcachestore.cpp
+++ b/zenserver/cache/structuredcachestore.cpp
@@ -491,7 +491,7 @@ ZenCacheMemoryLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 
 	std::vector<IoHash> BadHashes;
 
-	auto ValidateEntry = [](ZenContentType ContentType, IoBuffer Buffer) {
+	auto ValidateEntry = [](const IoHash& Hash, ZenContentType ContentType, IoBuffer Buffer) {
 		if (ContentType == ZenContentType::kCbObject)
 		{
 			CbValidateError Error = ValidateCompactBinary(Buffer, CbValidateMode::All);
@@ -499,7 +499,13 @@ ZenCacheMemoryLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 		}
 		if (ContentType == ZenContentType::kCompressedBinary)
 		{
-			if (CompressedBuffer Compressed = CompressedBuffer::FromCompressed(SharedBuffer(Buffer)); !Compressed)
+			IoHash	 RawHash;
+			uint64_t RawSize;
+			if (!CompressedBuffer::ValidateCompressedHeader(Buffer, RawHash, RawSize))
+			{
+				return false;
+			}
+			if (Hash != RawHash)
 			{
 				return false;
 			}
@@ -509,7 +515,7 @@ ZenCacheMemoryLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 
 	for (auto& Kv : m_CacheMap)
 	{
-		if (!ValidateEntry(Kv.second.Payload.GetContentType(), Kv.second.Payload))
+		if (!ValidateEntry(Kv.first, Kv.second.Payload.GetContentType(), Kv.second.Payload))
 		{
 			BadHashes.push_back(Kv.first);
 		}
@@ -1021,7 +1027,7 @@ ZenCacheDiskLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 	std::vector<BlockStoreLocation> ChunkLocations;
 	std::vector<IoHash>				ChunkIndexToChunkHash;
 
-	auto ValidateEntry = [](ZenContentType ContentType, IoBuffer Buffer) {
+	auto ValidateEntry = [](const IoHash& Hash, ZenContentType ContentType, IoBuffer Buffer) {
 		if (ContentType == ZenContentType::kCbObject)
 		{
 			CbValidateError Error = ValidateCompactBinary(Buffer, CbValidateMode::All);
@@ -1029,7 +1035,13 @@ ZenCacheDiskLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 		}
 		if (ContentType == ZenContentType::kCompressedBinary)
 		{
-			if (CompressedBuffer Compressed = CompressedBuffer::FromCompressed(SharedBuffer(Buffer)); !Compressed)
+			IoHash	 RawHash;
+			uint64_t RawSize;
+			if (!CompressedBuffer::ValidateCompressedHeader(Buffer, RawHash, RawSize))
+			{
+				return false;
+			}
+			if (RawHash != Hash)
 			{
 				return false;
 			}
@@ -1077,7 +1089,7 @@ ZenCacheDiskLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 				BadKeys.push_back(HashKey);
 				continue;
 			}
-			if (!ValidateEntry(Loc.GetContentType(), Value.Value))
+			if (!ValidateEntry(HashKey, Loc.GetContentType(), Value.Value))
 			{
 				BadKeys.push_back(HashKey);
 				continue;
@@ -1108,7 +1120,7 @@ ZenCacheDiskLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 			return;
 		}
 		ZenContentType ContentType = m_Index.at(Hash).Location.GetContentType();
-		if (!ValidateEntry(ContentType, Buffer))
+		if (!ValidateEntry(Hash, ContentType, Buffer))
 		{
 			BadKeys.push_back(Hash);
 			return;
@@ -1127,7 +1139,7 @@ ZenCacheDiskLayer::CacheBucket::Scrub(ScrubContext& Ctx)
 			return;
 		}
 		ZenContentType ContentType = m_Index.at(Hash).Location.GetContentType();
-		if (!ValidateEntry(ContentType, Buffer))
+		if (!ValidateEntry(Hash, ContentType, Buffer))
 		{
 			BadKeys.push_back(Hash);
 			return;
@@ -1678,18 +1690,8 @@ ZenCacheDiskLayer::CacheBucket::PutStandaloneCacheValue(const IoHash& HashKey, c
 	DataFile.MoveTemporaryIntoPlace(FsPath, Ec);
 	if (Ec)
 	{
-		std::filesystem::path ParentPath = FsPath.parent_path();
-		if (!std::filesystem::is_directory(ParentPath))
-		{
-			Ec.clear();
-			std::filesystem::create_directories(ParentPath, Ec);
-			if (Ec)
-			{
-				throw std::system_error(
-					Ec,
-					fmt::format("Failed to create parent directory '{}' for file '{}' for put in '{}'", ParentPath, FsPath, m_BucketDir));
-			}
-		}
+		CreateDirectories(FsPath.parent_path());
+		Ec.clear();
 
 		// Try again
 		DataFile.MoveTemporaryIntoPlace(FsPath, Ec);
@@ -3100,7 +3102,8 @@ TEST_CASE("z$.scrub")
 			{
 				IoBuffer		 AttachmentData			  = CreateBinaryCacheValue(AttachmentSizes[Index]);
 				CompressedBuffer CompressedAttachmentData = CompressedBuffer::Compress(SharedBuffer(AttachmentData));
-				Record.AddBinaryAttachment(fmt::format("attachment-{}", Index), IoHash::FromBLAKE3(CompressedAttachmentData.GetRawHash()));
+				Record.AddBinaryAttachment(fmt::format("attachment-{}", Index),
+										   IoHash::FromBLAKE3(CompressedAttachmentData.DecodeRawHash()));
 				Result.Attachments[Index] = CompressedAttachmentData;
 			}
 			Result.Record = Record.Save().GetBuffer().AsIoBuffer();
@@ -3142,7 +3145,7 @@ TEST_CASE("z$.scrub")
 				Zcs.Put("mybucket", Cid, {.Value = Record.Record});
 				for (const CompressedBuffer& Attachment : Record.Attachments)
 				{
-					CidStore.AddChunk(Attachment);
+					CidStore.AddChunk(Attachment.GetCompressed().Flatten().AsIoBuffer(), IoHash::FromBLAKE3(Attachment.DecodeRawHash()));
 				}
 			}
 		};