[C#] Recovery fixes (#421)

* [C#] Fix recovery of hash tables, speed up snapshot recovery. Use snapshot as default checkpoint for benchmarks. * avoid unnecessary hlog read. * ensure tail page allocated after recovery
microsoft · Mar 8, 2021 · 1eca35c · 1eca35c
1 parent 624a566
commit 1eca35c
Show file tree

Hide file tree

Showing 10 changed files with 61 additions and 47 deletions.
diff --git a/cs/benchmark/FasterYcsbBenchmark.cs b/cs/benchmark/FasterYcsbBenchmark.cs
@@ -116,10 +116,10 @@ public FASTER_YcsbBenchmark(int threadCount_, int numaStyle_, string distributio
 
             if (kSmallMemoryLog)
                 store = new FasterKV<Key, Value>
-                    (kMaxKey / 2, new LogSettings { LogDevice = device, PreallocateLog = true, PageSizeBits = 22, SegmentSizeBits = 26, MemorySizeBits = 26 }, new CheckpointSettings { CheckPointType = CheckpointType.FoldOver, CheckpointDir = path });
+                    (kMaxKey / 2, new LogSettings { LogDevice = device, PreallocateLog = true, PageSizeBits = 22, SegmentSizeBits = 26, MemorySizeBits = 26 }, new CheckpointSettings { CheckPointType = CheckpointType.Snapshot, CheckpointDir = path });
             else
                 store = new FasterKV<Key, Value>
-                    (kMaxKey / 2, new LogSettings { LogDevice = device, PreallocateLog = true }, new CheckpointSettings { CheckPointType = CheckpointType.FoldOver, CheckpointDir = path });
+                    (kMaxKey / 2, new LogSettings { LogDevice = device, PreallocateLog = true }, new CheckpointSettings { CheckPointType = CheckpointType.Snapshot, CheckpointDir = path });
         }
 
         private void RunYcsb(int thread_idx)

diff --git a/cs/src/core/Allocator/AllocatorBase.cs b/cs/src/core/Allocator/AllocatorBase.cs
@@ -1176,12 +1176,15 @@ public void RecoveryReset(long tailAddress, long headAddress, long beginAddress,
             TailPageOffset.Page = (int)tailPage;
             TailPageOffset.Offset = (int)offsetInPage;
 
-            // allocate next page as well - this is an invariant in the allocator!
+            // Allocate current page if necessary
             var pageIndex = (TailPageOffset.Page % BufferSize);
+            if (!IsAllocated(pageIndex))
+                AllocatePage(pageIndex);
+
+            // Allocate next page as well - this is an invariant in the allocator!
             var nextPageIndex = (pageIndex + 1) % BufferSize;
-            if (tailAddress > 0)
-                if (!IsAllocated(nextPageIndex))
-                    AllocatePage(nextPageIndex);
+            if (!IsAllocated(nextPageIndex))
+                AllocatePage(nextPageIndex);
 
             BeginAddress = beginAddress;
             HeadAddress = headAddress;

diff --git a/cs/src/core/Allocator/BlittableAllocator.cs b/cs/src/core/Allocator/BlittableAllocator.cs
@@ -130,7 +130,7 @@ internal override void AllocatePage(int index)
 
             handles[index] = GCHandle.Alloc(tmp, GCHandleType.Pinned);
             long p = (long)handles[index].AddrOfPinnedObject();
-            pointers[index] = (p + (sectorSize - 1)) & ~(sectorSize - 1);
+            pointers[index] = (p + (sectorSize - 1)) & ~((long)sectorSize - 1);
             values[index] = tmp;
         }
 

diff --git a/cs/src/core/Allocator/BlittableFrame.cs b/cs/src/core/Allocator/BlittableFrame.cs
@@ -35,7 +35,7 @@ public void Allocate(int index)
 
             handles[index] = GCHandle.Alloc(tmp, GCHandleType.Pinned);
             long p = (long)handles[index].AddrOfPinnedObject();
-            pointers[index] = (p + (sectorSize - 1)) & ~(sectorSize - 1);
+            pointers[index] = (p + (sectorSize - 1)) & ~((long)sectorSize - 1);
             frame[index] = tmp;
         }
 

diff --git a/cs/src/core/Allocator/GenericAllocator.cs b/cs/src/core/Allocator/GenericAllocator.cs
@@ -581,7 +581,7 @@ private void AsyncReadPageWithObjectsCallback<TContext>(uint errorCode, uint num
             if (size > int.MaxValue)
                 throw new FasterException("Unable to read object page, total size greater than 2GB: " + size);
 
-            var alignedLength = (size + (sectorSize - 1)) & ~(sectorSize - 1);
+            var alignedLength = (size + (sectorSize - 1)) & ~((long)sectorSize - 1);
             var objBuffer = bufferPool.Get((int)alignedLength);
             result.freeBuffer2 = objBuffer;
 

diff --git a/cs/src/core/Allocator/VarLenBlittableAllocator.cs b/cs/src/core/Allocator/VarLenBlittableAllocator.cs
@@ -235,7 +235,7 @@ internal override void AllocatePage(int index)
 
             handles[index] = GCHandle.Alloc(tmp, GCHandleType.Pinned);
             long p = (long)handles[index].AddrOfPinnedObject();
-            pointers[index] = (p + (sectorSize - 1)) & ~(sectorSize - 1);
+            pointers[index] = (p + (sectorSize - 1)) & ~((long)sectorSize - 1);
             values[index] = tmp;
         }
 

diff --git a/cs/src/core/Index/Recovery/IndexCheckpoint.cs b/cs/src/core/Index/Recovery/IndexCheckpoint.cs
@@ -22,7 +22,7 @@ internal unsafe void TakeIndexFuzzyCheckpoint()
             BeginMainIndexCheckpoint(ht_version, _indexCheckpoint.main_ht_device, out ulong ht_num_bytes_written, UseReadCache, SkipReadCacheBucket);
 
             var sectorSize = _indexCheckpoint.main_ht_device.SectorSize;
-            var alignedIndexSize = (uint)((ht_num_bytes_written + (sectorSize - 1)) & ~(sectorSize - 1));
+            var alignedIndexSize = (ht_num_bytes_written + (sectorSize - 1)) & ~((ulong)sectorSize - 1);
             overflowBucketsAllocator.BeginCheckpoint(_indexCheckpoint.main_ht_device, alignedIndexSize, out ulong ofb_num_bytes_written, UseReadCache, SkipReadCacheBucket, epoch);
             _indexCheckpoint.info.num_ht_bytes = ht_num_bytes_written;
             _indexCheckpoint.info.num_ofb_bytes = ofb_num_bytes_written;
@@ -37,7 +37,7 @@ internal void TakeIndexFuzzyCheckpoint(int ht_version, IDevice device,
         {
             BeginMainIndexCheckpoint(ht_version, device, out numBytesWritten);
             var sectorSize = device.SectorSize;
-            var alignedIndexSize = (uint)((numBytesWritten + (sectorSize - 1)) & ~(sectorSize - 1));
+            var alignedIndexSize = (numBytesWritten + (sectorSize - 1)) & ~((ulong)sectorSize - 1);
             overflowBucketsAllocator.BeginCheckpoint(ofbdevice, alignedIndexSize, out ofbnumBytesWritten);
             num_ofb_buckets = overflowBucketsAllocator.GetMaxValidAddress();
         }

diff --git a/cs/src/core/Index/Recovery/IndexRecovery.cs b/cs/src/core/Index/Recovery/IndexRecovery.cs
@@ -18,7 +18,7 @@ public partial class FasterBase
         // Derived class exposed API
         internal void RecoverFuzzyIndex(IndexCheckpointInfo info)
         {
-            uint alignedIndexSize = InitializeMainIndexRecovery(ref info, isAsync: false);
+            ulong alignedIndexSize = InitializeMainIndexRecovery(ref info, isAsync: false);
             overflowBucketsAllocator.Recover(info.main_ht_device, alignedIndexSize, info.info.num_buckets, info.info.num_ofb_bytes);
 
             // Wait until reading is complete
@@ -28,13 +28,13 @@ internal void RecoverFuzzyIndex(IndexCheckpointInfo info)
 
         internal async ValueTask RecoverFuzzyIndexAsync(IndexCheckpointInfo info, CancellationToken cancellationToken)
         {
-            uint alignedIndexSize = InitializeMainIndexRecovery(ref info, isAsync: true);
+            ulong alignedIndexSize = InitializeMainIndexRecovery(ref info, isAsync: true);
             await this.recoveryCountdown.WaitAsync(cancellationToken);
             await overflowBucketsAllocator.RecoverAsync(info.main_ht_device, alignedIndexSize, info.info.num_buckets, info.info.num_ofb_bytes, cancellationToken);
             FinalizeMainIndexRecovery(info);
         }
 
-        private uint InitializeMainIndexRecovery(ref IndexCheckpointInfo info, bool isAsync)
+        private ulong InitializeMainIndexRecovery(ref IndexCheckpointInfo info, bool isAsync)
         {
             var token = info.info.token;
             var ht_version = resizeInfo.version;
@@ -53,7 +53,7 @@ private uint InitializeMainIndexRecovery(ref IndexCheckpointInfo info, bool isAs
             BeginMainIndexRecovery(ht_version, info.main_ht_device, info.info.num_ht_bytes, isAsync);
 
 
-            var alignedIndexSize = (uint)((info.info.num_ht_bytes + (sectorSize - 1)) & ~(sectorSize - 1));
+            var alignedIndexSize = (info.info.num_ht_bytes + (sectorSize - 1)) & ~((ulong)sectorSize - 1);
             return alignedIndexSize;
         }
 
@@ -71,7 +71,7 @@ internal void RecoverFuzzyIndex(int ht_version, IDevice device, ulong num_ht_byt
         {
             BeginMainIndexRecovery(ht_version, device, num_ht_bytes);
             var sectorSize = device.SectorSize;
-            var alignedIndexSize = (uint)((num_ht_bytes + (sectorSize - 1)) & ~(sectorSize - 1));
+            var alignedIndexSize = (num_ht_bytes + (sectorSize - 1)) & ~((ulong)sectorSize - 1);
             overflowBucketsAllocator.Recover(ofbdevice, alignedIndexSize, num_buckets, num_ofb_bytes);
         }
 
@@ -81,7 +81,7 @@ internal async ValueTask RecoverFuzzyIndexAsync(int ht_version, IDevice device,
             BeginMainIndexRecovery(ht_version, device, num_ht_bytes, isAsync: true);
             await this.recoveryCountdown.WaitAsync(cancellationToken);
             var sectorSize = device.SectorSize;
-            var alignedIndexSize = (uint)((num_ht_bytes + (sectorSize - 1)) & ~(sectorSize - 1));
+            var alignedIndexSize = (num_ht_bytes + (sectorSize - 1)) & ~((ulong)sectorSize - 1);
             await overflowBucketsAllocator.RecoverAsync(ofbdevice, alignedIndexSize, num_buckets, num_ofb_bytes, cancellationToken);
         }