mem cache perf improvements (#592)

- Improvement: Refactor memory cache for faster trimming and correct trim reporting
- Improvement: Added trace scopes for memory cache trimming

Adding a link back to the cache item payload on the memory cache item allows us to iterate over only the items cached in memory instead of over the entire index.
This also allows us to do efficient compact of the memory cache array when trimming.
It adds 4 bytes of overhead to each item cached in memory.

1 files changed, 36 insertions, 16 deletions

diff --git a/src/zenserver/cache/cachedisklayer.h b/src/zenserver/cache/cachedisklayer.h
index 277371f2c..6997a12e4 100644
--- a/src/zenserver/cache/cachedisklayer.h
+++ b/src/zenserver/cache/cachedisklayer.h
@@ -197,15 +197,15 @@ public:
 		CacheBucket(GcManager& Gc, std::atomic_uint64_t& OuterCacheMemoryUsage, std::string BucketName, const BucketConfiguration& Config);
 		~CacheBucket();
 
-		bool OpenOrCreate(std::filesystem::path BucketDir, bool AllowCreate = true);
-		bool Get(const IoHash& HashKey, ZenCacheValue& OutValue);
-		void Put(const IoHash& HashKey, const ZenCacheValue& Value, std::span<IoHash> References);
-		void MemCacheTrim(GcClock::TimePoint ExpireTime);
-		bool Drop();
-		void Flush();
-		void ScrubStorage(ScrubContext& Ctx);
-		void GatherReferences(GcContext& GcCtx);
-		void CollectGarbage(GcContext& GcCtx);
+		bool	 OpenOrCreate(std::filesystem::path BucketDir, bool AllowCreate = true);
+		bool	 Get(const IoHash& HashKey, ZenCacheValue& OutValue);
+		void	 Put(const IoHash& HashKey, const ZenCacheValue& Value, std::span<IoHash> References);
+		uint64_t MemCacheTrim(GcClock::TimePoint ExpireTime);
+		bool	 Drop();
+		void	 Flush();
+		void	 ScrubStorage(ScrubContext& Ctx);
+		void	 GatherReferences(GcContext& GcCtx);
+		void	 CollectGarbage(GcContext& GcCtx);
 
 		inline GcStorageSize StorageSize() const
 		{
@@ -218,7 +218,7 @@ public:
 		CacheValueDetails::BucketDetails GetValueDetails(RwLock::SharedLockScope& IndexLock, const std::string_view ValueFilter) const;
 		void EnumerateBucketContents(std::function<void(const IoHash& Key, const CacheValueDetails::ValueDetails& Details)>& Fn) const;
 
-		void GetUsageByAccess(GcClock::TimePoint TickStart, GcClock::Duration SectionLength, std::vector<uint64_t>& InOutUsageSlots);
+		void GetUsageByAccess(GcClock::TimePoint Now, GcClock::Duration MaxAge, std::vector<uint64_t>& InOutUsageSlots);
 #if ZEN_WITH_TESTS
 		void SetAccessTime(const IoHash& HashKey, GcClock::TimePoint Time);
 #endif	// ZEN_WITH_TESTS
@@ -286,6 +286,11 @@ public:
 
 			operator bool() const { return RawSize != 0 || RawHash != IoHash::Zero; };
 		};
+		struct MemCacheData
+		{
+			IoBuffer	 Payload;
+			PayloadIndex OwnerIndex;
+		};
 #pragma pack(pop)
 		static_assert(sizeof(BucketPayload) == 20u);
 		static_assert(sizeof(BucketMetaData) == 28u);
@@ -323,7 +328,7 @@ public:
 		std::vector<BucketPayload>	m_Payloads;
 		std::vector<BucketMetaData> m_MetaDatas;
 		std::vector<MetaDataIndex>	m_FreeMetaDatas;
-		std::vector<IoBuffer>		m_MemCachedPayloads;
+		std::vector<MemCacheData>	m_MemCachedPayloads;
 		std::vector<MemCachedIndex> m_FreeMemCachedPayloads;
 		std::vector<ReferenceIndex> m_FirstReferenceIndex;
 		std::vector<IoHash>			m_ReferenceHashes;
@@ -364,7 +369,7 @@ public:
 								   const ZenCacheDiskLayer::CacheBucket::BucketMetaData& MetaData);
 		void		   RemoveMetaData(RwLock::ExclusiveLockScope&, BucketPayload& Payload);
 		BucketMetaData GetMetaData(RwLock::SharedLockScope&, const BucketPayload& Payload) const;
-		void		   SetMemCachedData(RwLock::ExclusiveLockScope&, BucketPayload& Payload, IoBuffer& MemCachedData);
+		void		   SetMemCachedData(RwLock::ExclusiveLockScope&, PayloadIndex PayloadIndex, IoBuffer& MemCachedData);
 		size_t		   RemoveMemCachedData(RwLock::ExclusiveLockScope&, BucketPayload& Payload);
 
 		void	 InitializeIndexFromDisk(RwLock::ExclusiveLockScope&, bool IsNew);
@@ -390,7 +395,7 @@ public:
 						  std::vector<BucketPayload>&  Payloads,
 						  std::vector<AccessTime>&	   AccessTimes,
 						  std::vector<BucketMetaData>& MetaDatas,
-						  std::vector<IoBuffer>&	   MemCachedPayloads,
+						  std::vector<MemCacheData>&   MemCachedPayloads,
 						  std::vector<ReferenceIndex>& FirstReferenceIndex,
 						  IndexMap&					   Index,
 						  RwLock::ExclusiveLockScope&  IndexLock);
@@ -405,6 +410,10 @@ public:
 			m_MemCachedSize.fetch_sub(ValueSize, std::memory_order::relaxed);
 			m_OuterCacheMemoryUsage.fetch_sub(ValueSize, std::memory_order::relaxed);
 		}
+		static inline uint64_t EstimateMemCachePayloadMemory(uint64_t PayloadSize)
+		{
+			return sizeof(MemCacheData) + sizeof(IoBufferCore) + RoundUp(PayloadSize, 8u);
+		}
 
 		// These locks are here to avoid contention on file creation, therefore it's sufficient
 		// that we take the same lock for the same hash
@@ -436,10 +445,21 @@ private:
 		{
 			return;
 		}
+		if (m_IsMemCacheTrimming)
+		{
+			return;
+		}
+
+		const GcClock::Tick NowTick = GcClock::TickCount();
+		if (NowTick < m_NextAllowedTrimTick)
+		{
+			return;
+		}
+
 		MemCacheTrim();
 	}
-	void MemCacheTrim();
-	void MemCacheTrim(std::vector<CacheBucket*>& Buckets, GcClock::TimePoint ExpireTime);
+	void	 MemCacheTrim();
+	uint64_t MemCacheTrim(std::vector<CacheBucket*>& Buckets, GcClock::TimePoint ExpireTime);
 
 	GcManager&													  m_Gc;
 	JobQueue&													  m_JobQueue;
@@ -447,7 +467,7 @@ private:
 	Configuration												  m_Configuration;
 	std::atomic_uint64_t										  m_TotalMemCachedSize{};
 	std::atomic_bool											  m_IsMemCacheTrimming = false;
-	std::atomic<GcClock::Tick>									  m_LastTickMemCacheTrim;
+	std::atomic<GcClock::Tick>									  m_NextAllowedTrimTick;
 	mutable RwLock												  m_Lock;
 	std::unordered_map<std::string, std::unique_ptr<CacheBucket>> m_Buckets;
 	std::vector<std::unique_ptr<CacheBucket>>					  m_DroppedBuckets;