Caffeine使用一個ConcurrencyHashMap來保存所有數(shù)據(jù),那它的過期淘汰策略采用的什么的方式與數(shù)據(jù)結(jié)構(gòu)呢?其中寫過期是使用writeOrderDeque,這個比較簡單無需多說,而讀過期相對復(fù)雜很多,使用W-TinyLFU的結(jié)構(gòu)與算法。
網(wǎng)絡(luò)上有很多文章介紹W-TinyLFU結(jié)構(gòu)的,大家可以去查一下,這里主要是從源碼來分析,總的來說它使用了三個雙端隊列:accessOrderEdenDeque,accessOrderProbationDeque,accessOrderProtectedDeque,使用雙端隊列的原因是支持LRU算法比較方便。
accessOrderEdenDeque屬于eden區(qū),緩存1%的數(shù)據(jù),其余的99%緩存在main區(qū)。
accessOrderProbationDeque屬于main區(qū),緩存main內(nèi)數(shù)據(jù)的20%,這部分是屬于冷數(shù)據(jù),即將補淘汰。
accessOrderProtectedDeque屬于main區(qū),緩存main內(nèi)數(shù)據(jù)的80%,這部分是屬于熱數(shù)據(jù),是整個緩存的主存區(qū)。
我們先看一下淘汰方法入口:
void evictEntries() {
if (!evicts()) {
return;
}
//先從edn區(qū)淘汰
int candidates = evictFromEden();
//eden淘汰后的數(shù)據(jù)進入main區(qū),然后再從main區(qū)淘汰
evictFromMain(candidates);
}
accessOrderEdenDeque對應(yīng)W-TinyLFU的W(window),這里保存的是最新寫入數(shù)據(jù)的引用,它使用LRU淘汰,這里面的數(shù)據(jù)主要是應(yīng)對突發(fā)流量的問題,淘汰后的數(shù)據(jù)進入
accessOrderProbationDeque.代碼如下:
int evictFromEden() {
int candidates = 0;
Node<K, V> node = accessOrderEdenDeque().peek();
while (edenWeightedSize() > edenMaximum()) {
// The pending operations will adjust the size to reflect the correct weight
if (node == null) {
break;
}
Node<K, V> next = node.getNextInAccessOrder();
if (node.getWeight() != 0) {
node.makeMainProbation();
//先從eden區(qū)移除
accessOrderEdenDeque().remove(node);
//移除的數(shù)據(jù)加入到main區(qū)的probation隊列
accessOrderProbationDeque().add(node);
candidates++;
lazySetEdenWeightedSize(edenWeightedSize() - node.getPolicyWeight());
}
node = next;
}
return candidates;
}
數(shù)據(jù)進入probation隊列后,繼續(xù)執(zhí)行以下代碼:
void evictFromMain(int candidates) {
int victimQueue = PROBATION;
Node<K, V> victim = accessOrderProbationDeque().peekFirst();
Node<K, V> candidate = accessOrderProbationDeque().peekLast();
while (weightedSize() > maximum()) {
// Stop trying to evict candidates and always prefer the victim
if (candidates == 0) {
candidate = null;
}
// Try evicting from the protected and eden queues
if ((candidate == null) && (victim == null)) {
if (victimQueue == PROBATION) {
victim = accessOrderProtectedDeque().peekFirst();
victimQueue = PROTECTED;
continue;
} else if (victimQueue == PROTECTED) {
victim = accessOrderEdenDeque().peekFirst();
victimQueue = EDEN;
continue;
}
// The pending operations will adjust the size to reflect the correct weight
break;
}
// Skip over entries with zero weight
if ((victim != null) && (victim.getPolicyWeight() == 0)) {
victim = victim.getNextInAccessOrder();
continue;
} else if ((candidate != null) && (candidate.getPolicyWeight() == 0)) {
candidate = candidate.getPreviousInAccessOrder();
candidates--;
continue;
}
// Evict immediately if only one of the entries is present
if (victim == null) {
candidates--;
Node<K, V> evict = candidate;
candidate = candidate.getPreviousInAccessOrder();
evictEntry(evict, RemovalCause.SIZE, 0L);
continue;
} else if (candidate == null) {
Node<K, V> evict = victim;
victim = victim.getNextInAccessOrder();
evictEntry(evict, RemovalCause.SIZE, 0L);
continue;
}
// Evict immediately if an entry was collected
K victimKey = victim.getKey();
K candidateKey = candidate.getKey();
if (victimKey == null) {
Node<K, V> evict = victim;
victim = victim.getNextInAccessOrder();
evictEntry(evict, RemovalCause.COLLECTED, 0L);
continue;
} else if (candidateKey == null) {
candidates--;
Node<K, V> evict = candidate;
candidate = candidate.getPreviousInAccessOrder();
evictEntry(evict, RemovalCause.COLLECTED, 0L);
continue;
}
// Evict immediately if the candidate's weight exceeds the maximum
if (candidate.getPolicyWeight() > maximum()) {
candidates--;
Node<K, V> evict = candidate;
candidate = candidate.getPreviousInAccessOrder();
evictEntry(evict, RemovalCause.SIZE, 0L);
continue;
}
// Evict the entry with the lowest frequency
candidates--;
//最核心算法在這里:從probation的頭尾取出兩個node進行比較頻率,頻率更小者將被remove
if (admit(candidateKey, victimKey)) {
Node<K, V> evict = victim;
victim = victim.getNextInAccessOrder();
evictEntry(evict, RemovalCause.SIZE, 0L);
candidate = candidate.getPreviousInAccessOrder();
} else {
Node<K, V> evict = candidate;
candidate = candidate.getPreviousInAccessOrder();
evictEntry(evict, RemovalCause.SIZE, 0L);
}
}
}
上面的代碼邏輯是從probation的頭尾取出兩個node進行比較頻率,頻率更小者將被remove,其中尾部元素就是上一部分從eden中淘汰出來的元素,如果將兩步邏輯合并起來講是這樣的: 在eden隊列通過lru淘汰出來的”候選者“與probation隊列通過lru淘汰出來的“被驅(qū)逐者“進行頻率比較,失敗者將被從cache中真正移除。下面看一下它的比較邏輯admit:
boolean admit(K candidateKey, K victimKey) {
int victimFreq = frequencySketch().frequency(victimKey);
int candidateFreq = frequencySketch().frequency(candidateKey);
//如果候選者的頻率高就淘汰被驅(qū)逐者
if (candidateFreq > victimFreq) {
return true;
//如果被驅(qū)逐者比候選者的頻率高,并且候選者頻率小于等于5則淘汰者
} else if (candidateFreq <= 5) {
// The maximum frequency is 15 and halved to 7 after a reset to age the history. An attack
// exploits that a hot candidate is rejected in favor of a hot victim. The threshold of a warm
// candidate reduces the number of random acceptances to minimize the impact on the hit rate.
return false;
}
//隨機淘汰
int random = ThreadLocalRandom.current().nextInt();
return ((random & 127) == 0);
}
從frequencySketch取出候選者與被驅(qū)逐者的頻率,如果候選者的頻率高就淘汰被驅(qū)逐者,如果被驅(qū)逐者比候選者的頻率高,并且候選者頻率小于等于5則淘汰者,如果前面兩個條件都不滿足則隨機淘汰。
整個過程中你是不是發(fā)現(xiàn)protectedDeque并沒有什么作用,那它是怎么作為主存區(qū)來保存大部分數(shù)據(jù)的呢?
//onAccess方法觸發(fā)該方法
void reorderProbation(Node<K, V> node) {
if (!accessOrderProbationDeque().contains(node)) {
// Ignore stale accesses for an entry that is no longer present
return;
} else if (node.getPolicyWeight() > mainProtectedMaximum()) {
return;
}
long mainProtectedWeightedSize = mainProtectedWeightedSize() + node.getPolicyWeight();
//先從probation中移除
accessOrderProbationDeque().remove(node);
//加入到protected中
accessOrderProtectedDeque().add(node);
node.makeMainProtected();
long mainProtectedMaximum = mainProtectedMaximum();
//從protected中移除
while (mainProtectedWeightedSize > mainProtectedMaximum) {
Node<K, V> demoted = accessOrderProtectedDeque().pollFirst();
if (demoted == null) {
break;
}
demoted.makeMainProbation();
//加入到probation中
accessOrderProbationDeque().add(demoted);
mainProtectedWeightedSize -= node.getPolicyWeight();
}
lazySetMainProtectedWeightedSize(mainProtectedWeightedSize);
}
當(dāng)數(shù)據(jù)被訪問時并且該數(shù)據(jù)在probation中,這個數(shù)據(jù)就會移動到protected中去,同時通過lru從protected中淘汰一個數(shù)據(jù)進入到probation中。
這樣數(shù)據(jù)流轉(zhuǎn)的邏輯全部通了:新數(shù)據(jù)都會進入到eden中,通過lru淘汰到probation,并與probation中通過lru淘汰的數(shù)據(jù)進行使用頻率pk,如果勝利了就繼續(xù)留在probation中,如果失敗了就會被直接淘汰,當(dāng)這條數(shù)據(jù)被訪問了,則移動到protected。當(dāng)其它數(shù)據(jù)被訪問了,則它可能會從protected中通過lru淘汰到probation中。
TinyLFU
傳統(tǒng)LFU一般使用key-value形式來記錄每個key的頻率,優(yōu)點是數(shù)據(jù)結(jié)構(gòu)非常簡單,并且能跟緩存本身的數(shù)據(jù)結(jié)構(gòu)復(fù)用,增加一個屬性記錄頻率就行了,它的缺點也比較明顯就是頻率這個屬性會占用很大的空間,但如果改用壓縮方式存儲頻率呢? 頻率占用空間肯定可以減少,但會引出另外一個問題:怎么從壓縮后的數(shù)據(jù)里獲得對應(yīng)key的頻率呢?
TinyLFU的解決方案是類似位圖的方法,將key取hash值獲得它的位下標(biāo),然后用這個下標(biāo)來找頻率,但位圖只有0、1兩個值,那頻率明顯可能會非常大,這要怎么處理呢? 另外使用位圖需要預(yù)占非常大的空間,這個問題怎么解決呢?
TinyLFU根據(jù)最大數(shù)據(jù)量設(shè)置生成一個long數(shù)組,然后將頻率值保存在其中的四個long的4個bit位中(4個bit位不會大于15),取頻率值時則取四個中的最小一個。
Caffeine認為頻率大于15已經(jīng)很高了,是屬于熱數(shù)據(jù),所以它只需要4個bit位來保存,long有8個字節(jié)64位,這樣可以保存16個頻率。取hash值的后左移兩位,然后加上hash四次,這樣可以利用到16個中的13個,利用率挺高的,或許有更好的算法能將16個都利用到。
public void increment(@Nonnull E e) {
if (isNotInitialized()) {
return;
}
int hash = spread(e.hashCode());
int start = (hash & 3) << 2;
// Loop unrolling improves throughput by 5m ops/s
int index0 = indexOf(hash, 0); //indexOf也是一種hash方法,不過會通過tableMask來限制范圍
int index1 = indexOf(hash, 1);
int index2 = indexOf(hash, 2);
int index3 = indexOf(hash, 3);
boolean added = incrementAt(index0, start);
added |= incrementAt(index1, start + 1);
added |= incrementAt(index2, start + 2);
added |= incrementAt(index3, start + 3);
//當(dāng)數(shù)據(jù)寫入次數(shù)達到數(shù)據(jù)長度時就重置
if (added && (++size == sampleSize)) {
reset();
}
}
給對應(yīng)位置的bit位四位的Int值加1:
boolean incrementAt(int i, int j) {
int offset = j << 2;
long mask = (0xfL << offset);
//當(dāng)已達到15時,次數(shù)不再增加
if ((table[i] & mask) != mask) {
table[i] += (1L << offset);
return true;
}
return false;
}
獲得值的方法也是通過四次hash來獲得,然后取最小值:
public int frequency(@Nonnull E e) {
if (isNotInitialized()) {
return 0;
}
int hash = spread(e.hashCode());
int start = (hash & 3) << 2;
int frequency = Integer.MAX_VALUE;
//四次hash
for (int i = 0; i < 4; i++) {
int index = indexOf(hash, i);
//獲得bit位四位的Int值
int count = (int) ((table[index] >>> ((start + i) << 2)) & 0xfL);
//取最小值
frequency = Math.min(frequency, count);
}
return frequency;
}
當(dāng)數(shù)據(jù)寫入次數(shù)達到數(shù)據(jù)長度時就會將次數(shù)減半,一些冷數(shù)據(jù)在這個過程中將歸0,這樣會使hash沖突降低:
void reset() {
int count = 0;
for (int i = 0; i < table.length; i++) {
count += Long.bitCount(table[i] & ONE_MASK);
table[i] = (table[i] >>> 1) & RESET_MASK;
}
size = (size >>> 1) - (count >>> 2);
}
關(guān)注個人微信公眾號:肌肉碼農(nóng),回復(fù)“好友”討論問題。