[C#] Allow recovery without index checkpoint (#328)

* Allow recovery without index checkpoint, by simply using an empty hash table and replaying the log from the begin address.

* Remove NullDevice as default device in LogSettings. Make sure users do not instantiate FASTER without the necessary devices, in order to avoid unexpected behavior.

cs/src/core/Allocator/AllocatorBase.cs

@@ -484,6 +484,10 @@ public unsafe abstract partial class AllocatorBase<Key, Value> : IDisposable
         /// <param name="flushCallback"></param>
         public AllocatorBase(LogSettings settings, IFasterEqualityComparer<Key> comparer, Action<long, long> evictCallback, LightEpoch epoch, Action<CommitInfo> flushCallback)
         {
+            if (settings.LogDevice == null)
+            {
+                throw new FasterException("LogSettings.LogDevice needs to be specified (e.g., use Devices.CreateLogDevice, AzureStorageDevice, or NullDevice)");
+            }
             if (evictCallback != null)
             {
                 ReadCache = true;

cs/src/core/Allocator/GenericAllocator.cs

@@ -42,6 +42,11 @@ public unsafe sealed class GenericAllocator<Key, Value> : AllocatorBase<Key, Val
         public GenericAllocator(LogSettings settings, SerializerSettings<Key, Value> serializerSettings, IFasterEqualityComparer<Key> comparer, Action<long, long> evictCallback = null, LightEpoch epoch = null, Action<CommitInfo> flushCallback = null)
             : base(settings, comparer, evictCallback, epoch, flushCallback)
         {
+            if (settings.ObjectLogDevice == null)
+            {
+                throw new FasterException("LogSettings.ObjectLogDevice needs to be specified (e.g., use Devices.CreateLogDevice, AzureStorageDevice, or NullDevice)");
+            }
+
             SerializerSettings = serializerSettings;
 
             if ((!keyBlittable) && (settings.LogDevice as NullDevice == null) && ((SerializerSettings == null) || (SerializerSettings.keySerializer == null)))

cs/src/core/Device/Devices.cs

@@ -26,9 +26,6 @@ public static class Devices
         /// <returns>Device instance</returns>
         public static IDevice CreateLogDevice(string logPath, bool preallocateFile = false, bool deleteOnClose = false, long capacity = CAPACITY_UNSPECIFIED, bool recoverDevice = false)
         {
-            if (string.IsNullOrWhiteSpace(logPath))
-                return new NullDevice();
-
             IDevice logDevice;
 
 #if DOTNETCORE

cs/src/core/Index/Common/LogSettings.cs

@@ -32,12 +32,12 @@ public class LogSettings
         /// <summary>
         /// Device used for main hybrid log
         /// </summary>
-        public IDevice LogDevice = new NullDevice();
+        public IDevice LogDevice;
 
         /// <summary>
         /// Device used for serialized heap objects in hybrid log
         /// </summary>
-        public IDevice ObjectLogDevice = new NullDevice();
+        public IDevice ObjectLogDevice;
 
         /// <summary>
         /// Size of a segment (group of pages), in bits

cs/src/core/Index/FASTER/FASTER.cs

@@ -144,6 +144,7 @@ public partial class FasterKV<Key, Value> : FasterBase,
                         readcache = new VariableLengthBlittableAllocator<Key, Value>(
                             new LogSettings
                             {
+                                LogDevice = new NullDevice(),
                                 PageSizeBits = logSettings.ReadCacheSettings.PageSizeBits,
                                 MemorySizeBits = logSettings.ReadCacheSettings.MemorySizeBits,
                                 SegmentSizeBits = logSettings.ReadCacheSettings.MemorySizeBits,
@@ -162,6 +163,7 @@ public partial class FasterKV<Key, Value> : FasterBase,
                         readcache = new BlittableAllocator<Key, Value>(
                             new LogSettings
                             {
+                                LogDevice = new NullDevice(),
                                 PageSizeBits = logSettings.ReadCacheSettings.PageSizeBits,
                                 MemorySizeBits = logSettings.ReadCacheSettings.MemorySizeBits,
                                 SegmentSizeBits = logSettings.ReadCacheSettings.MemorySizeBits,
@@ -183,6 +185,8 @@ public partial class FasterKV<Key, Value> : FasterBase,
                     readcache = new GenericAllocator<Key, Value>(
                         new LogSettings
                         {
+                            LogDevice = new NullDevice(),
+                            ObjectLogDevice = new NullDevice(),
                             PageSizeBits = logSettings.ReadCacheSettings.PageSizeBits,
                             MemorySizeBits = logSettings.ReadCacheSettings.MemorySizeBits,
                             SegmentSizeBits = logSettings.ReadCacheSettings.MemorySizeBits,

cs/src/core/Index/FASTER/FASTERIterator.cs

@@ -92,7 +92,7 @@ public FasterKVIterator(FasterKV<Key, Value> fht, Functions functions, Compactio
             this.cf = cf;
             enumerationPhase = 0;
             fhtSession = fht.NewSession<Empty, Empty, Empty, Functions>(functions);
-            tempKv = new FasterKV<Key, Value>(fht.IndexSize, new LogSettings { MutableFraction = 1 }, comparer: fht.Comparer, variableLengthStructSettings: variableLengthStructSettings);
+            tempKv = new FasterKV<Key, Value>(fht.IndexSize, new LogSettings { LogDevice = new NullDevice(), ObjectLogDevice = new NullDevice(), MutableFraction = 1 }, comparer: fht.Comparer, variableLengthStructSettings: variableLengthStructSettings);
             tempKvSession = tempKv.NewSession<Empty, Empty, Empty, Functions>(functions);
             iter1 = fht.Log.Scan(fht.Log.BeginAddress, untilAddress);
         }

cs/src/core/Index/FASTER/LogAccessor.cs

@@ -253,7 +253,7 @@ public void Compact<CompactionFunctions>(CompactionFunctions compactionFunctions
             var originalUntilAddress = untilAddress;
 
             using (var fhtSession = fht.NewSession<Empty, Empty, Empty, Functions>(functions))
-            using (var tempKv = new FasterKV<Key, Value>(fht.IndexSize, new LogSettings(), comparer: fht.Comparer, variableLengthStructSettings: variableLengthStructSettings))
+            using (var tempKv = new FasterKV<Key, Value>(fht.IndexSize, new LogSettings { LogDevice = new NullDevice(), ObjectLogDevice = new NullDevice() }, comparer: fht.Comparer, variableLengthStructSettings: variableLengthStructSettings))
             using (var tempKvSession = tempKv.NewSession<Empty, Empty, Empty, Functions>(functions))
             {
                 using (var iter1 = fht.Log.Scan(fht.Log.BeginAddress, untilAddress))

cs/src/core/Index/Recovery/IndexRecovery.cs

@@ -27,7 +27,9 @@ internal void RecoverFuzzyIndex(IndexCheckpointInfo info)
         {
             var token = info.info.token;
             var ht_version = resizeInfo.version;
-            Debug.Assert(state[ht_version].size == info.info.table_size);
+
+            if (state[ht_version].size != info.info.table_size)
+                throw new FasterException($"Incompatible hash table size during recovery; allocated {state[ht_version].size} buckets, recovering {info.info.table_size} buckets");
 
             // Create devices to read from using Async API
             info.main_ht_device = checkpointManager.GetIndexDevice(token);

cs/src/core/Index/Recovery/Recovery.cs

@@ -98,14 +98,14 @@ private void InternalRecoverFromLatestCheckpoints(int numPagesToPreload)
                 }
 
                 Debug.WriteLine("Index Checkpoint: {0}", indexToken);
+                recoveredICInfo.info.DebugPrint();
                 break;
             }
 
             if (recoveredICInfo.IsDefault())
-                throw new FasterException("Unable to find valid index token");
-
-            recoveredICInfo.info.DebugPrint();
-
+            {
+                Debug.WriteLine("No index checkpoint found, recovering from beginning of log");
+            }
 
             InternalRecover(recoveredICInfo, recoveredHLCInfo, numPagesToPreload);
         }
@@ -125,18 +125,33 @@ private void InternalRecover(Guid indexToken, Guid hybridLogToken, int numPagesT
 
 
             // Recovery appropriate context information
-            var recoveredICInfo = new IndexCheckpointInfo();
-            recoveredICInfo.Recover(indexToken, checkpointManager);
-            recoveredICInfo.info.DebugPrint();
-
             var recoveredHLCInfo = new HybridLogCheckpointInfo();
             recoveredHLCInfo.Recover(hybridLogToken, checkpointManager);
             recoveredHLCInfo.info.DebugPrint();
 
-            // Check if the two checkpoints are compatible for recovery
-            if (!IsCompatible(recoveredICInfo.info, recoveredHLCInfo.info))
+            IndexCheckpointInfo recoveredICInfo;
+            try
             {
-                throw new FasterException("Cannot recover from (" + indexToken.ToString() + "," + hybridLogToken.ToString() + ") checkpoint pair!\n");
+                recoveredICInfo = new IndexCheckpointInfo();
+                recoveredICInfo.Recover(indexToken, checkpointManager);
+                recoveredICInfo.info.DebugPrint();
+            }
+            catch
+            {
+                recoveredICInfo = default;
+            }
+
+            if (recoveredICInfo.IsDefault())
+            {
+                Debug.WriteLine("Invalid index checkpoint token, recovering from beginning of log");
+            }
+            else
+            {
+                // Check if the two checkpoints are compatible for recovery
+                if (!IsCompatible(recoveredICInfo.info, recoveredHLCInfo.info))
+                {
+                    throw new FasterException("Cannot recover from (" + indexToken.ToString() + "," + hybridLogToken.ToString() + ") checkpoint pair!\n");
+                }
             }
 
             InternalRecover(recoveredICInfo, recoveredHLCInfo, numPagesToPreload);
@@ -153,7 +168,11 @@ private void InternalRecover(IndexCheckpointInfo recoveredICInfo, HybridLogCheck
             systemState.version = (v + 1);
 
             // Recover fuzzy index from checkpoint
-            RecoverFuzzyIndex(recoveredICInfo);
+            if (recoveredICInfo.IsDefault())
+                recoveredICInfo.info.startLogicalAddress = recoveredHLCInfo.info.beginAddress;
+            else
+                RecoverFuzzyIndex(recoveredICInfo);
+
 
             // Recover segment offsets for object log
             if (recoveredHLCInfo.info.objectLogSegmentOffsets != null)