optimized index snapshot reading/writing (#561)

the previous implementation of in-memory index snapshots serialise data to memory before writing to disk and vice versa when reading. This leads to some memory spikes which end up pushing useful data out of system cache and also cause stalls on I/O operations.

this change moves more code to a streaming serialisation approach which scales better from a memory usage perspective and also performs much better

1 files changed, 28 insertions, 28 deletions

diff --git a/src/zenserver/cache/cachedisklayer.h b/src/zenserver/cache/cachedisklayer.h
index 2986cedf8..8d015d127 100644
--- a/src/zenserver/cache/cachedisklayer.h
+++ b/src/zenserver/cache/cachedisklayer.h
@@ -189,7 +189,6 @@ public:
 	void SetAccessTime(std::string_view Bucket, const IoHash& HashKey, GcClock::TimePoint Time);
 #endif	// ZEN_WITH_TESTS
 
-private:
 	/** A cache bucket manages a single directory containing
 		metadata and data for that bucket
 	  */
@@ -225,21 +224,6 @@ private:
 #endif	// ZEN_WITH_TESTS
 
 	private:
-		GcManager&			  m_Gc;
-		std::atomic_uint64_t& m_OuterCacheMemoryUsage;
-		std::string			  m_BucketName;
-		std::filesystem::path m_BucketDir;
-		std::filesystem::path m_BlocksBasePath;
-		BucketConfiguration	  m_Configuration;
-		BlockStore			  m_BlockStore;
-		Oid					  m_BucketId;
-		std::atomic_bool	  m_IsFlushing{};
-
-		// These files are used to manage storage of small objects for this bucket
-
-		TCasLogFile<DiskIndexEntry> m_SlogFile;
-		uint64_t					m_LogFlushPosition = 0;
-
 #pragma pack(push)
 #pragma pack(1)
 		struct MetaDataIndex
@@ -309,6 +293,22 @@ private:
 
 		using IndexMap = tsl::robin_map<IoHash, PayloadIndex, IoHash::Hasher>;
 
+	private:
+		GcManager&			  m_Gc;
+		std::atomic_uint64_t& m_OuterCacheMemoryUsage;
+		std::string			  m_BucketName;
+		std::filesystem::path m_BucketDir;
+		std::filesystem::path m_BlocksBasePath;
+		BucketConfiguration	  m_Configuration;
+		BlockStore			  m_BlockStore;
+		Oid					  m_BucketId;
+		std::atomic_bool	  m_IsFlushing{};
+
+		// These files are used to manage storage of small objects for this bucket
+
+		TCasLogFile<DiskIndexEntry> m_SlogFile;
+		uint64_t					m_LogFlushPosition = 0;
+
 		std::atomic<uint64_t> m_DiskHitCount;
 		std::atomic<uint64_t> m_DiskMissCount;
 		std::atomic<uint64_t> m_DiskWriteCount;
@@ -342,19 +342,9 @@ private:
 		IoBuffer GetStandaloneCacheValue(ZenContentType ContentType, const IoHash& HashKey) const;
 		void	 PutInlineCacheValue(const IoHash& HashKey, const ZenCacheValue& Value, std::span<IoHash> References);
 		IoBuffer GetInlineCacheValue(const DiskLocation& Loc) const;
-		void	 MakeIndexSnapshot(const std::function<uint64_t()>& ClaimDiskReserveFunc = []() { return 0; });
-		uint64_t ReadIndexFile(const std::filesystem::path& IndexPath, uint32_t& OutVersion);
-		uint64_t ReadLog(const std::filesystem::path& LogPath, uint64_t LogPosition);
-		void	 OpenLog(const bool IsNew);
-		CbObject MakeManifest(IndexMap&&						 Index,
-							  std::vector<AccessTime>&&			 AccessTimes,
-							  const std::vector<BucketPayload>&	 Payloads,
-							  const std::vector<BucketMetaData>& MetaDatas);
-		void	 SaveManifest(
-				CbObject&&						 Manifest,
-				const std::function<uint64_t()>& ClaimDiskReserveFunc = []() { return 0; });
 		CacheValueDetails::ValueDetails GetValueDetails(const IoHash& Key, PayloadIndex Index) const;
-		void							CompactReferences(RwLock::ExclusiveLockScope&);
+
+		void		CompactReferences(RwLock::ExclusiveLockScope&);
 		void		SetReferences(RwLock::ExclusiveLockScope&, ReferenceIndex& FirstReferenceIndex, std::span<IoHash> References);
 		void		RemoveReferences(RwLock::ExclusiveLockScope&, ReferenceIndex& FirstReferenceIndex);
 		inline bool GetReferences(RwLock::SharedLockScope&, ReferenceIndex FirstReferenceIndex, std::vector<IoHash>& OutReferences) const
@@ -368,12 +358,20 @@ private:
 		ReferenceIndex AllocateReferenceEntry(RwLock::ExclusiveLockScope&, const IoHash& Key);
 		bool		   LockedGetReferences(ReferenceIndex FirstReferenceIndex, std::vector<IoHash>& OutReferences) const;
 		void		   ClearReferenceCache();
+
 		void		   SetMetaData(BucketPayload& Payload, const ZenCacheDiskLayer::CacheBucket::BucketMetaData& MetaData);
 		void		   RemoveMetaData(BucketPayload& Payload);
 		BucketMetaData GetMetaData(const BucketPayload& Payload) const;
 		void		   SetMemCachedData(BucketPayload& Payload, IoBuffer& MemCachedData);
 		size_t		   RemoveMemCachedData(BucketPayload& Payload);
 
+		void	 InitializeIndexFromDisk(bool IsNew);
+		uint64_t ReadIndexFile(const std::filesystem::path& IndexPath, uint32_t& OutVersion);
+		uint64_t ReadLog(const std::filesystem::path& LogPath, uint64_t LogPosition);
+
+		void SaveSnapshot(const std::function<uint64_t()>& ClaimDiskReserveFunc = []() { return 0; });
+		void WriteIndexSnapshot(const std::function<uint64_t()>& ClaimDiskReserveFunc = []() { return 0; });
+
 		void CompactState(std::vector<BucketPayload>&  Payloads,
 						  std::vector<AccessTime>&	   AccessTimes,
 						  std::vector<BucketMetaData>& MetaDatas,
@@ -404,8 +402,10 @@ private:
 
 		friend class DiskBucketReferenceChecker;
 		friend class DiskBucketStoreCompactor;
+		friend class BucketManifestSerializer;
 	};
 
+private:
 	inline void TryMemCacheTrim()
 	{
 		if (m_Configuration.MemCacheTargetFootprintBytes == 0)