refactor comapactcas index (#443)

- Bugfix: Fix scrub messing up payload and access time in disk cache bucket when compacting index
- Improvement: Split up disk cache bucket index into hash lookup and payload array to improve performance
- Improvement: Reserve space up front for compact binary output when saving cache bucket manifest to improve performance

1 files changed, 57 insertions, 16 deletions

diff --git a/src/zenstore/compactcas.cpp b/src/zenstore/compactcas.cpp
index ce2e53527..e6383c3a1 100644
--- a/src/zenstore/compactcas.cpp
+++ b/src/zenstore/compactcas.cpp
@@ -174,7 +174,8 @@ CasContainerStrategy::InsertChunk(const void* ChunkData, size_t ChunkSize, const
 		m_CasLog.Append(IndexEntry);
 		{
 			RwLock::ExclusiveLockScope _(m_LocationMapLock);
-			m_LocationMap.emplace(ChunkHash, DiskLocation);
+			m_LocationMap.emplace(ChunkHash, m_Locations.size());
+			m_Locations.push_back(DiskLocation);
 		}
 	});
 
@@ -201,7 +202,7 @@ CasContainerStrategy::FindChunk(const IoHash& ChunkHash)
 	{
 		return IoBuffer();
 	}
-	const BlockStoreLocation& Location = KeyIt->second.Get(m_PayloadAlignment);
+	const BlockStoreLocation& Location = m_Locations[KeyIt->second].Get(m_PayloadAlignment);
 
 	IoBuffer Chunk = m_BlockStore.TryGetChunk(Location);
 	return Chunk;
@@ -258,7 +259,7 @@ CasContainerStrategy::ScrubStorage(ScrubContext& Ctx)
 			for (const auto& Entry : m_LocationMap)
 			{
 				const IoHash&				  ChunkHash	   = Entry.first;
-				const BlockStoreDiskLocation& DiskLocation = Entry.second;
+				const BlockStoreDiskLocation& DiskLocation = m_Locations[Entry.second];
 				BlockStoreLocation			  Location	   = DiskLocation.Get(m_PayloadAlignment);
 
 				ChunkLocations.push_back(Location);
@@ -355,7 +356,7 @@ CasContainerStrategy::ScrubStorage(ScrubContext& Ctx)
 			std::vector<CasDiskIndexEntry> LogEntries;
 			LogEntries.reserve(BadKeys.size());
 			{
-				RwLock::ExclusiveLockScope __(m_LocationMapLock);
+				RwLock::ExclusiveLockScope IndexLock(m_LocationMapLock);
 				for (const IoHash& ChunkHash : BadKeys)
 				{
 					const auto KeyIt = m_LocationMap.find(ChunkHash);
@@ -364,9 +365,13 @@ CasContainerStrategy::ScrubStorage(ScrubContext& Ctx)
 						// Might have been GC'd
 						continue;
 					}
-					LogEntries.push_back({.Key = KeyIt->first, .Location = KeyIt->second, .Flags = CasDiskIndexEntry::kTombstone});
+					LogEntries.push_back(
+						{.Key = KeyIt->first, .Location = m_Locations[KeyIt->second], .Flags = CasDiskIndexEntry::kTombstone});
 					m_LocationMap.erase(KeyIt);
 				}
+
+				// Clean up m_Locations vectors
+				CompactIndex(IndexLock);
 			}
 			m_CasLog.Append(LogEntries);
 		}
@@ -423,8 +428,9 @@ CasContainerStrategy::CollectGarbage(GcContext& GcCtx)
 	uint64_t ReadBlockTimeUs		 = 0;
 	uint64_t ReadBlockLongestTimeUs	 = 0;
 
-	LocationMap_t					 LocationMap;
-	BlockStore::ReclaimSnapshotState BlockStoreState;
+	LocationMap_t						LocationMap;
+	std::vector<BlockStoreDiskLocation> Locations;
+	BlockStore::ReclaimSnapshotState	BlockStoreState;
 	{
 		ZEN_TRACE_CPU("CasContainer::CollectGarbage::State");
 
@@ -436,6 +442,7 @@ CasContainerStrategy::CollectGarbage(GcContext& GcCtx)
 			 WriteBlockLongestTimeUs = std::max(ElapsedUs, WriteBlockLongestTimeUs);
 		 });
 		LocationMap					 = m_LocationMap;
+		Locations					 = m_Locations;
 		BlockStoreState				 = m_BlockStore.GetReclaimSnapshotState();
 	}
 
@@ -459,7 +466,7 @@ CasContainerStrategy::CollectGarbage(GcContext& GcCtx)
 		ZEN_TRACE_CPU("CasContainer::CollectGarbage::Filter");
 		GcCtx.FilterCids(TotalChunkHashes, [&](const IoHash& ChunkHash, bool Keep) {
 			auto						  KeyIt		   = LocationMap.find(ChunkHash);
-			const BlockStoreDiskLocation& DiskLocation = KeyIt->second;
+			const BlockStoreDiskLocation& DiskLocation = Locations[KeyIt->second];
 			BlockStoreLocation			  Location	   = DiskLocation.Get(m_PayloadAlignment);
 			size_t						  ChunkIndex   = ChunkLocations.size();
 
@@ -499,7 +506,7 @@ CasContainerStrategy::CollectGarbage(GcContext& GcCtx)
 			for (const size_t ChunkIndex : RemovedChunks)
 			{
 				const IoHash&				  ChunkHash		  = ChunkIndexToChunkHash[ChunkIndex];
-				const BlockStoreDiskLocation& OldDiskLocation = LocationMap[ChunkHash];
+				const BlockStoreDiskLocation& OldDiskLocation = Locations[LocationMap[ChunkHash]];
 				LogEntries.push_back({.Key = ChunkHash, .Location = OldDiskLocation, .Flags = CasDiskIndexEntry::kTombstone});
 				DeletedChunks.push_back(ChunkHash);
 			}
@@ -521,15 +528,41 @@ CasContainerStrategy::CollectGarbage(GcContext& GcCtx)
 						m_LocationMap.erase(Entry.Key);
 						continue;
 					}
-					m_LocationMap[Entry.Key] = Entry.Location;
+					m_Locations[m_LocationMap[Entry.Key]] = Entry.Location;
 				}
 			}
 		},
 		[&GcCtx]() { return GcCtx.CollectSmallObjects(); });
 
+	if (!DeletedChunks.empty())
+	{
+		// Clean up m_Locations vectors
+		RwLock::ExclusiveLockScope IndexLock(m_LocationMapLock);
+		CompactIndex(IndexLock);
+	}
 	GcCtx.AddDeletedCids(DeletedChunks);
 }
 
+void
+CasContainerStrategy::CompactIndex(RwLock::ExclusiveLockScope&)
+{
+	ZEN_TRACE_CPU("CasContainer::CompactIndex");
+
+	size_t								EntryCount = m_LocationMap.size();
+	LocationMap_t						LocationMap;
+	std::vector<BlockStoreDiskLocation> Locations;
+	Locations.reserve(EntryCount);
+	LocationMap.reserve(EntryCount);
+	for (auto It : m_LocationMap)
+	{
+		size_t EntryIndex = Locations.size();
+		Locations.push_back(m_Locations[It.second]);
+		LocationMap.insert({It.first, EntryIndex});
+	}
+	m_LocationMap.swap(LocationMap);
+	m_Locations.swap(Locations);
+}
+
 GcStorageSize
 CasContainerStrategy::StorageSize() const
 {
@@ -592,7 +625,7 @@ CasContainerStrategy::MakeIndexSnapshot()
 			{
 				CasDiskIndexEntry& IndexEntry = Entries[EntryIndex++];
 				IndexEntry.Key				  = Entry.first;
-				IndexEntry.Location			  = Entry.second;
+				IndexEntry.Location			  = m_Locations[Entry.second];
 			}
 		}
 
@@ -673,7 +706,8 @@ CasContainerStrategy::ReadIndexFile(const std::filesystem::path& IndexPath, uint
 					ZEN_WARN("skipping invalid entry in '{}', reason: '{}'", IndexPath, InvalidEntryReason);
 					continue;
 				}
-				m_LocationMap[Entry.Key] = Entry.Location;
+				m_LocationMap[Entry.Key] = m_Locations.size();
+				m_Locations.push_back(Entry.Location);
 			}
 
 			OutVersion = CasDiskIndexHeader::CurrentVersion;
@@ -733,9 +767,11 @@ CasContainerStrategy::ReadLog(const std::filesystem::path& LogPath, uint64_t Ski
 					ZEN_WARN("skipping invalid entry in '{}', reason: '{}'", LogPath, InvalidEntryReason);
 					return;
 				}
-				m_LocationMap[Record.Key] = Record.Location;
+				m_LocationMap[Record.Key] = m_Locations.size();
+				m_Locations.push_back(Record.Location);
 			},
 			SkipEntryCount);
+
 		return LogEntryCount;
 	}
 	return 0;
@@ -749,6 +785,7 @@ CasContainerStrategy::OpenContainer(bool IsNewStore)
 	// Add .running file and delete on clean on close to detect bad termination
 
 	m_LocationMap.clear();
+	m_Locations.clear();
 
 	std::filesystem::path BasePath = GetBasePath(m_RootDirectory, m_ContainerBaseName);
 
@@ -797,11 +834,12 @@ CasContainerStrategy::OpenContainer(bool IsNewStore)
 	std::vector<CasDiskIndexEntry> BadEntries;
 	for (const auto& Entry : m_LocationMap)
 	{
-		const BlockStoreDiskLocation& DiskLocation = Entry.second;
-		auto						  BlockIt	   = BlockSizes.find(DiskLocation.GetBlockIndex());
+		const BlockStoreDiskLocation& DiskLocation = m_Locations[Entry.second];
+		uint32_t					  BlockIndex   = DiskLocation.GetBlockIndex();
+		auto						  BlockIt	   = BlockSizes.find(BlockIndex);
 		if (BlockIt == BlockSizes.end())
 		{
-			ZEN_WARN("Unknown block {} for entry {} in '{}'", DiskLocation.GetBlockIndex(), Entry.first.ToHexString(), BasePath);
+			ZEN_WARN("Unknown block {} for entry {} in '{}'", BlockIndex, Entry.first.ToHexString(), BasePath);
 		}
 		else
 		{
@@ -835,6 +873,9 @@ CasContainerStrategy::OpenContainer(bool IsNewStore)
 		{
 			m_LocationMap.erase(BadEntry.Key);
 		}
+
+		RwLock::ExclusiveLockScope IndexLock(m_LocationMapLock);
+		CompactIndex(IndexLock);
 	}
 
 	m_BlockStore.Prune(KnownLocations);