ConcurrentDictionary的源码看了很多遍，今天抽点时间整理一下，它的实现比Dictionary要复杂很多，至于线程安全我觉得比较简单，用的是lock的思想。首先我们来看看它的源码。
public class ConcurrentDictionary<TKey, TValue> : IDictionary<TKey, TValue>, IDictionary, IReadOnlyDictionary<TKey, TValue> { /// <summary> /// Tables that hold the internal state of the ConcurrentDictionary /// /// Wrapping the three tables in a single object allows us to atomically /// replace all tables at once. /// </summary> private class Tables { internal readonly Node[] m_buckets; // A singly-linked list for each bucket. internal readonly object[] m_locks; // A set of locks, each guarding a section of the table. internal volatile int[] m_countPerLock; // The number of elements guarded by each lock. internal readonly IEqualityComparer<TKey> m_comparer; // Key equality comparer internal Tables(Node[] buckets, object[] locks, int[] countPerLock, IEqualityComparer<TKey> comparer) { m_buckets = buckets; m_locks = locks; m_countPerLock = countPerLock; m_comparer = comparer; } } private const int DEFAULT_CONCURRENCY_MULTIPLIER = 4; private const int DEFAULT_CAPACITY = 31; private const int MAX_LOCK_NUMBER = 1024; // Whether TValue is a type that can be written atomically (i.e., with no danger of torn reads) private static readonly bool s_isValueWriteAtomic = IsValueWriteAtomic(); public ConcurrentDictionary() : this(DefaultConcurrencyLevel, DEFAULT_CAPACITY, true, EqualityComparer<TKey>.Default) public ConcurrentDictionary(int concurrencyLevel, int capacity) : this(concurrencyLevel, capacity, false, EqualityComparer<TKey>.Default) { } public ConcurrentDictionary(int concurrencyLevel, int capacity, IEqualityComparer<TKey> comparer) : this(concurrencyLevel, capacity, false, comparer){} internal ConcurrentDictionary(int concurrencyLevel, int capacity, bool growLockArray, IEqualityComparer<TKey> comparer) { if (concurrencyLevel < 1) { throw new ArgumentOutOfRangeException("concurrencyLevel", GetResource("ConcurrentDictionary_ConcurrencyLevelMustBePositive")); } if (capacity < 0) { throw new ArgumentOutOfRangeException("capacity", GetResource("ConcurrentDictionary_CapacityMustNotBeNegative")); } if (comparer == null) throw new ArgumentNullException("comparer"); // The capacity should be at least as large as the concurrency level. Otherwise, we would have locks that don't guard // any buckets. if (capacity < concurrencyLevel) { capacity = concurrencyLevel; } object[] locks = new object[concurrencyLevel]; for (int i = 0; i < locks.Length; i++) { locks[i] = new object(); } int[] countPerLock = new int[locks.Length]; Node[] buckets = new Node[capacity]; m_tables = new Tables(buckets, locks, countPerLock, comparer); m_growLockArray = growLockArray; m_budget = buckets.Length / locks.Length; } public TValue this[TKey key] { get { TValue value; if (!TryGetValue(key, out value)) { throw new KeyNotFoundException(); } return value; } set { if (key == null) throw new ArgumentNullException("key"); TValue dummy; TryAddInternal(key, value, true, true, out dummy); } } public bool TryGetValue(TKey key, out TValue value) { if (key == null) throw new ArgumentNullException("key"); int bucketNo, lockNoUnused; // We must capture the m_buckets field in a local variable. It is set to a new table on each table resize. Tables tables = m_tables; IEqualityComparer<TKey> comparer = tables.m_comparer; GetBucketAndLockNo(comparer.GetHashCode(key), out bucketNo, out lockNoUnused, tables.m_buckets.Length, tables.m_locks.Length); Node n = Volatile.Read<Node>(ref tables.m_buckets[bucketNo]); while (n != null) { if (comparer.Equals(n.m_key, key)) { value = n.m_value; return true; } n = n.m_next; } value = default(TValue); return false; } private bool TryAddInternal(TKey key, TValue value, bool updateIfExists, bool acquireLock, out TValue resultingValue) { while (true) { int bucketNo, lockNo; int hashcode; Tables tables = m_tables; IEqualityComparer<TKey> comparer = tables.m_comparer; hashcode = comparer.GetHashCode(key); GetBucketAndLockNo(hashcode, out bucketNo, out lockNo, tables.m_buckets.Length, tables.m_locks.Length); bool resizeDesired = false; bool lockTaken = false; try { if (acquireLock) Monitor.Enter(tables.m_locks[lockNo], ref lockTaken); // If the table just got resized, we may not be holding the right lock, and must retry. // This should be a rare occurence. if (tables != m_tables) { continue; } // Try to find this key in the bucket Node prev = null; for (Node node = tables.m_buckets[bucketNo]; node != null; node = node.m_next) { Assert((prev == null && node == tables.m_buckets[bucketNo]) || prev.m_next == node); if (comparer.Equals(node.m_key, key)) { // The key was found in the dictionary. If updates are allowed, update the value for that key. // We need to create a new node for the update, in order to support TValue types that cannot // be written atomically, since lock-free reads may be happening concurrently. if (updateIfExists) { if (s_isValueWriteAtomic) { node.m_value = value; } else { Node newNode = new Node(node.m_key, value, hashcode, node.m_next); if (prev == null) { tables.m_buckets[bucketNo] = newNode; } else { prev.m_next = newNode; } } resultingValue = value; } else { resultingValue = node.m_value; } return false; } prev = node; } // The key was not found in the bucket. Insert the key-value pair. Volatile.Write<Node>(ref tables.m_buckets[bucketNo], new Node(key, value, hashcode, tables.m_buckets[bucketNo])); checked { tables.m_countPerLock[lockNo]++; } if (tables.m_countPerLock[lockNo] > m_budget) { resizeDesired = true; } } finally { if (lockTaken) Monitor.Exit(tables.m_locks[lockNo]); } if (resizeDesired) { GrowTable(tables, tables.m_comparer, false, m_keyRehashCount); } resultingValue = value; return true; } } public bool TryRemove(TKey key, out TValue value) { if (key == null) throw new ArgumentNullException("key"); return TryRemoveInternal(key, out value, false, default(TValue)); } private bool TryRemoveInternal(TKey key, out TValue value, bool matchValue, TValue oldValue) { while (true) { Tables tables = m_tables; IEqualityComparer<TKey> comparer = tables.m_comparer; int bucketNo, lockNo; GetBucketAndLockNo(comparer.GetHashCode(key), out bucketNo, out lockNo, tables.m_buckets.Length, tables.m_locks.Length); lock (tables.m_locks[lockNo]) { // If the table just got resized, we may not be holding the right lock, and must retry. // This should be a rare occurence. if (tables != m_tables) { continue; } Node prev = null; for (Node curr = tables.m_buckets[bucketNo]; curr != null; curr = curr.m_next) { Assert((prev == null && curr == tables.m_buckets[bucketNo]) || prev.m_next == curr); if (comparer.Equals(curr.m_key, key)) { if (matchValue) { bool valuesMatch = EqualityComparer<TValue>.Default.Equals(oldValue, curr.m_value); if (!valuesMatch) { value = default(TValue); return false; } } if (prev == null) { Volatile.Write<Node>(ref tables.m_buckets[bucketNo], curr.m_next); } else { prev.m_next = curr.m_next; } value = curr.m_value; tables.m_countPerLock[lockNo]--; return true; } prev = curr; } } value = default(TValue); return false; } } private void GrowTable(Tables tables, IEqualityComparer<TKey> newComparer, bool regenerateHashKeys, int rehashCount) { int locksAcquired = 0; try { AcquireLocks(0, 1, ref locksAcquired); if (regenerateHashKeys && rehashCount == m_keyRehashCount) { tables = m_tables; } else { if (tables != m_tables) { return; } long approxCount = 0; for (int i = 0; i < tables.m_countPerLock.Length; i++) { approxCount += tables.m_countPerLock[i]; } if (approxCount < tables.m_buckets.Length / 4) { m_budget = 2 * m_budget; if (m_budget < 0) { m_budget = int.MaxValue; } return; } } int newLength = 0; bool maximizeTableSize = false; try { checked { newLength = tables.m_buckets.Length * 2 + 1; while (newLength % 3 == 0 || newLength % 5 == 0 || newLength % 7 == 0) { newLength += 2; } Assert(newLength % 2 != 0); if (newLength > Array.MaxArrayLength) { maximizeTableSize = true; } } } catch (OverflowException) { maximizeTableSize = true; } if (maximizeTableSize) { newLength = Array.MaxArrayLength; m_budget = int.MaxValue; } // Now acquire all other locks for the table AcquireLocks(1, tables.m_locks.Length, ref locksAcquired); object[] newLocks = tables.m_locks; // Add more locks if (m_growLockArray && tables.m_locks.Length < MAX_LOCK_NUMBER) { newLocks = new object[tables.m_locks.Length * 2]; Array.Copy(tables.m_locks, newLocks, tables.m_locks.Length); for (int i = tables.m_locks.Length; i < newLocks.Length; i++) { newLocks[i] = new object(); } } Node[] newBuckets = new Node[newLength]; int[] newCountPerLock = new int[newLocks.Length]; for (int i = 0; i < tables.m_buckets.Length; i++) { Node current = tables.m_buckets[i]; while (current != null) { Node next = current.m_next; int newBucketNo, newLockNo; int nodeHashCode = current.m_hashcode; if (regenerateHashKeys) { // Recompute the hash from the key nodeHashCode = newComparer.GetHashCode(current.m_key); } GetBucketAndLockNo(nodeHashCode, out newBucketNo, out newLockNo, newBuckets.Length, newLocks.Length); newBuckets[newBucketNo] = new Node(current.m_key, current.m_value, nodeHashCode, newBuckets[newBucketNo]); checked { newCountPerLock[newLockNo]++; } current = next; } } // If this resize regenerated the hashkeys, increment the count if (regenerateHashKeys) { // We use unchecked here because we don't want to throw an exception if // an overflow happens unchecked { m_keyRehashCount++; } } // Adjust the budget m_budget = Math.Max(1, newBuckets.Length / newLocks.Length); // Replace tables with the new versions m_tables = new Tables(newBuckets, newLocks, newCountPerLock, newComparer); } finally { // Release all locks that we took earlier ReleaseLocks(0, locksAcquired); } } private void AcquireLocks(int fromInclusive, int toExclusive, ref int locksAcquired) { Assert(fromInclusive <= toExclusive); object[] locks = m_tables.m_locks; for (int i = fromInclusive; i < toExclusive; i++) { bool lockTaken = false; try { Monitor.Enter(locks[i], ref lockTaken); } finally { if (lockTaken) { locksAcquired++; } } } } private void GetBucketAndLockNo(int hashcode, out int bucketNo, out int lockNo, int bucketCount, int lockCount) { bucketNo = (hashcode & 0x7fffffff) % bucketCount; lockNo = bucketNo % lockCount; Assert(bucketNo >= 0 && bucketNo < bucketCount); Assert(lockNo >= 0 && lockNo < lockCount); } private static int DefaultConcurrencyLevel { get { return DEFAULT_CONCURRENCY_MULTIPLIER * PlatformHelper.ProcessorCount; } } private class Node { internal TKey m_key; internal TValue m_value; internal volatile Node m_next; internal int m_hashcode; internal Node(TKey key, TValue value, int hashcode, Node next) { m_key = key; m_value = value; m_next = next; m_hashcode = hashcode; } } } public static class Volatile { [ResourceExposure(ResourceScope.None)] [ReliabilityContract(Consistency.WillNotCorruptState, Cer.Success)] [SecuritySafeCritical] //the intrinsic implementation of this method contains unverifiable code public static T Read<T>(ref T location) where T : class { var value = location; Thread.MemoryBarrier(); return value; } [ResourceExposure(ResourceScope.None)] [ReliabilityContract(Consistency.WillNotCorruptState, Cer.Success)] [SecuritySafeCritical] //the intrinsic implementation of this method contains unverifiable code public static void Write<T>(ref T location, T value) where T : class { Thread.MemoryBarrier(); location = value; } }
ConcurrentDictionary的构造函数依然有int capacity参数，该参数是控制ConcurrentDictionary里面的初始节点数组的大小【Node[] buckets = new Node[capacity] 和m_tables = new Tables(buckets, locks, countPerLock, comparer);】，同时构造函数中多了一个int concurrencyLevel参数，控制并行度【object[] locks = new object[concurrencyLevel]; for (int i = 0; i < locks.Length; i++){ locks[i] = new object(); }】。如果指定了int capacity参数，很多时候参数bool growLockArray为false【m_growLockArray = growLockArray;】表示ConcurrentDictionary在扩容的时候，object[] locks 这个锁的对象数组不扩容，可以理解为锁的粒度变大了，先前4个key公用一个lock对象，现在可能8个key对应一个对象；m_budget = buckets.Length / locks.Length中的m_budget 可以理解为一个lock对象被多少个key共享。
现在我们来看看TryGetValue获取值，这个方法非常简单，应为读取时不需要加锁的，所以首先根据key计算其哈希值，再找到对应的哈希桶，读取哈希桶的数据【Node n = Volatile.Read<Node>(ref tables.m_buckets[bucketNo])】；一个哈希桶的数据可能有多个【 while (n != null){if (comparer.Equals(n.m_key, key)){ value = n.m_value; return true; } n = n.m_next;}】，所以从这里可以看出来每个 哈希桶里面是一个Node链表数据结构。
接下来我们看看比较复杂的TryAddInternal方法，优先需要根据key来确定哈希桶，无论是添加还是修改 都需要锁定对象，所以这里用的是Monitor.Enter(tables.m_locks[lockNo], ref lockTaken); 在最后在释放锁 Monitor.Exit(tables.m_locks[lockNo]);，如果是添加元素那么直接给里面的哈希桶赋值 Volatile.Write<Node>(ref tables.m_buckets[bucketNo], new Node(key, value, hashcode, tables.m_buckets[bucketNo]));注意Node的构造函数，tables.m_buckets[bucketNo])将是新节点的m_next值，也就是添加的新节点永远是哈希桶链表的第一个节点，这里，赋值后对应的lock对象的计数器需要加1【tables.m_countPerLock[lockNo]++;】，如果每个计数器达到预计达阀值就需要扩容了【if (tables.m_countPerLock[lockNo] > m_budget){ resizeDesired = true;}】，那么修改也是首先找到对应的node节点【如果添加的key所在哈希桶里面存在数据】，如果value是可以直接修改的话，那么我们直接修改【 if (s_isValueWriteAtomic) { node.m_value = value;}】，不是的话那我们就克隆一个节点 替换掉原先的节点【Node newNode = new Node(node.m_key, value, hashcode, node.m_next); if (prev == null){ tables.m_buckets[bucketNo] = newNode; } else{ prev.m_next = newNode;}】，如果是桶的第一个节点那么替换比较简单，否者就修改先前节点的m_next 属性。
接下来我们来看看哈希桶的扩容GrowTable，这个方法比较复杂，我就没怎么仔细研读了，首先是多线程我们需要考虑线程安全，说白了就是加锁 AcquireLocks(0, 1, ref locksAcquired)，哈希桶扩容基本是按照2倍来扩容的【 newLength = tables.m_buckets.Length * 2 + 1; while (newLength % 3 == 0 || newLength % 5 == 0 || newLength % 7 == 0){ newLength += 2; }】，在正真扩容前我们需要锁定所有对象【AcquireLocks(1, tables.m_locks.Length, ref locksAcquired);】，扩容首先需要扩容锁的对象数组
if (m_growLockArray && tables.m_locks.Length < MAX_LOCK_NUMBER) { newLocks = new object[tables.m_locks.Length * 2]; Array.Copy(tables.m_locks, newLocks, tables.m_locks.Length); for (int i = tables.m_locks.Length; i < newLocks.Length; i++) { newLocks[i] = new object(); } }
然后在是哈希桶扩容，这里扩容可以理解为克隆原先的节点到新的数组中 旧的位置上【newBuckets[newBucketNo] = new Node(current.m_key, current.m_value, nodeHashCode, newBuckets[newBucketNo]);】
Node[] newBuckets = new Node[newLength]; int[] newCountPerLock = new int[newLocks.Length]; for (int i = 0; i < tables.m_buckets.Length; i++) { Node current = tables.m_buckets[i]; while (current != null) { Node next = current.m_next; int newBucketNo, newLockNo; int nodeHashCode = current.m_hashcode; if (regenerateHashKeys) { // Recompute the hash from the key nodeHashCode = newComparer.GetHashCode(current.m_key); } GetBucketAndLockNo(nodeHashCode, out newBucketNo, out newLockNo, newBuckets.Length, newLocks.Length); newBuckets[newBucketNo] = new Node(current.m_key, current.m_value, nodeHashCode, newBuckets[newBucketNo]); checked { newCountPerLock[newLockNo]++; } current = next; } }
看来扩容，最后来看看移除元素，首先需要根据key来计算哈希桶的位置【GetBucketAndLockNo(comparer.GetHashCode(key), out bucketNo, out lockNo, tables.m_buckets.Length, tables.m_locks.Length)】，然后锁住对应的对象【 lock (tables.m_locks[lockNo])】，在哈希桶里面获取遍历链表查找对应的key，如果是桶的第一个节点则直接写 Volatile.Write<Node>(ref tables.m_buckets[bucketNo], curr.m_next)，否者修改链表prev.m_next = curr.m_next，最后该lock对象的计数器需要减1【tables.m_countPerLock[lockNo]--】。
-----------------------------在一次面试的时候 被问到Count属性， 我们来看看Count的实现吧：
private void AcquireAllLocks(ref int locksAcquired) { // First, acquire lock 0 AcquireLocks(0, 1, ref locksAcquired); // Now that we have lock 0, the m_locks array will not change (i.e., grow), // and so we can safely read m_locks.Length. AcquireLocks(1, m_tables.m_locks.Length, ref locksAcquired); Assert(locksAcquired == m_tables.m_locks.Length); } private void AcquireLocks(int fromInclusive, int toExclusive, ref int locksAcquired) { Assert(fromInclusive <= toExclusive); object[] locks = m_tables.m_locks; for (int i = fromInclusive; i < toExclusive; i++) { bool lockTaken = false; try { Monitor.Enter(locks[i], ref lockTaken); } finally { if (lockTaken) { locksAcquired++; } } } } private int GetCountInternal() { int count = 0; // Compute the count, we allow overflow for (int i = 0; i < m_tables.m_countPerLock.Length; i++) { count += m_tables.m_countPerLock[i]; } return count; } private void ReleaseLocks(int fromInclusive, int toExclusive) { Assert(fromInclusive <= toExclusive); for (int i = fromInclusive; i < toExclusive; i++) { Monitor.Exit(m_tables.m_locks[i]); } }
看到这里Count是需要获取m_tables.m_locks每一个对象的锁， ConcurrentDictionary的性能比lock+Dictionary 的性能高出的主要原因就是锁的粒度变小了， 但是这个count需要获取多个对象的锁， 所以相对耗时，同样GetKeys(),GetValues(),ToArray()，IsEmpty也是和Count一样，需要获取所有的锁.