don't read chunks into memory during cache batch fetch unless we may cache them in memory (#188)

* Don't read chunks into memory during cache batch fetch unless we may cache them in memory

2 files changed, 14 insertions, 13 deletions

diff --git a/src/zenstore/blockstore.cpp b/src/zenstore/blockstore.cpp
index 239b9e56b..8a6d9c18d 100644
--- a/src/zenstore/blockstore.cpp
+++ b/src/zenstore/blockstore.cpp
@@ -254,7 +254,6 @@ BlockStoreFile::GetMetaPath() const
 ////////////////////////////////////////////////////////
 
 constexpr uint64_t DefaultIterateSmallChunkWindowSize = 2 * 1024 * 1024;
-constexpr uint64_t IterateSmallChunkMaxGapSize		  = 4 * 1024;
 
 BlockStore::BlockStore()
 {
@@ -718,26 +717,27 @@ BlockStore::IterateBlock(std::span<const BlockStoreLocation>   ChunkLocations,
 		return true;
 	}
 
-	uint64_t IterateSmallChunkWindowSize = Max(DefaultIterateSmallChunkWindowSize, LargeSizeLimit);
-	if (LargeSizeLimit == 0u)
-	{
-		LargeSizeLimit = IterateSmallChunkWindowSize;
-	}
-	else
+	if (LargeSizeLimit == 0)
 	{
-		IterateSmallChunkWindowSize =
-			Min((LargeSizeLimit + IterateSmallChunkMaxGapSize) * ChunkLocations.size(), IterateSmallChunkWindowSize);
+		LargeSizeLimit = DefaultIterateSmallChunkWindowSize;
 	}
 
+	uint64_t IterateSmallChunkWindowSize = Max(DefaultIterateSmallChunkWindowSize, LargeSizeLimit);
+
+	const uint64_t IterateSmallChunkMaxGapSize = Max(2048u, IterateSmallChunkWindowSize / 512u);
+
+	IterateSmallChunkWindowSize = Min((LargeSizeLimit + IterateSmallChunkMaxGapSize) * ChunkLocations.size(), IterateSmallChunkWindowSize);
+
 	uint32_t			BlockIndex = ChunkLocations[InChunkIndexes[0]].BlockIndex;
 	std::vector<size_t> ChunkIndexes(InChunkIndexes.begin(), InChunkIndexes.end());
 	std::sort(ChunkIndexes.begin(), ChunkIndexes.end(), [&](size_t IndexA, size_t IndexB) -> bool {
 		return ChunkLocations[IndexA].Offset < ChunkLocations[IndexB].Offset;
 	});
 
-	auto GetNextRange = [LargeSizeLimit, IterateSmallChunkWindowSize, &ChunkLocations](uint64_t				   BlockFileSize,
-																					   std::span<const size_t> ChunkIndexes,
-																					   size_t				   StartIndexOffset) -> size_t {
+	auto GetNextRange = [LargeSizeLimit,
+						 IterateSmallChunkWindowSize,
+						 IterateSmallChunkMaxGapSize,
+						 &ChunkLocations](uint64_t BlockFileSize, std::span<const size_t> ChunkIndexes, size_t StartIndexOffset) -> size_t {
 		size_t					  ChunkCount	= 0;
 		size_t					  StartIndex	= ChunkIndexes[StartIndexOffset];
 		const BlockStoreLocation& StartLocation = ChunkLocations[StartIndex];
diff --git a/src/zenstore/cache/cachedisklayer.cpp b/src/zenstore/cache/cachedisklayer.cpp
index 8a2de34e2..9161905d7 100644
--- a/src/zenstore/cache/cachedisklayer.cpp
+++ b/src/zenstore/cache/cachedisklayer.cpp
@@ -1503,7 +1503,8 @@ ZenCacheDiskLayer::CacheBucket::EndGetBatch(GetBatchHandle* Batch) noexcept
 			{
 				ZEN_TRACE_CPU("Z$::Bucket::EndGetBatch::ReadInline");
 				m_BlockStore.IterateChunks(InlineBlockLocations, [&](uint32_t, std::span<const size_t> ChunkIndexes) -> bool {
-					const uint64_t LargeChunkSizeLimit = Max(m_Configuration.MemCacheSizeThreshold, 32u * 1024u);
+					// Only read into memory the IoBuffers we could potentially add to memcache
+					const uint64_t LargeChunkSizeLimit = Max(m_Configuration.MemCacheSizeThreshold, 1u * 1024u);
 					m_BlockStore.IterateBlock(
 						InlineBlockLocations,
 						ChunkIndexes,