/* *
* slotsmgrttagslot host port timeout slot
* */
void
slotsmgrttagslotCommand(redisClient *c) {
sds host = c->argv[1]->ptr;
sds port = c->argv[2]->ptr;
int timeout, slot;
if (parse_timeout(c, c->argv[3], &timeout) != 0) {
return;
}
if (parse_slot(c, c->argv[4], &slot) != 0) {
return;
}
dict *d = c->db->hash_slots[slot];
int succ = 0;
do {
const dictEntry *de = dictGetRandomKey(d);
if (de == NULL) {
break;
}
sds skey = dictGetKey(de);
robj *key = createStringObject(skey, sdslen(skey));
succ = slotsmgrttag_command(c, host, port, timeout, key);
decrRefCount(key);
if (succ < 0) {
return;
}
} while (0);
addReplyMultiBulkLen(c, 2);
addReplyLongLong(c, succ);
addReplyLongLong(c, dictSize(d));
}
/* Return random element from a non empty set.
* The returned element can be a int64_t value if the set is encoded
* as an "intset" blob of integers, or a redis object if the set
* is a regular set.
*
* The caller provides both pointers to be populated with the right
* object. The return value of the function is the object->encoding
* field of the object and is used by the caller to check if the
* int64_t pointer or the redis object pointere was populated.
*
* When an object is returned (the set was a real set) the ref count
* of the object is not incremented so this function can be considered
* copy on write friendly. */
int setTypeRandomElement(robj *setobj, robj **objele, int64_t *llele) {
if (setobj->encoding == REDIS_ENCODING_HT) {
dictEntry *de = dictGetRandomKey(setobj->ptr);
*objele = dictGetEntryKey(de);
} else if (setobj->encoding == REDIS_ENCODING_INTSET) {
*llele = intsetRandom(setobj->ptr);
} else {
redisPanic("Unknown set encoding");
}
return setobj->encoding;
}
const sds dbGetRandSet(void)
{
dictEntry *entry = NULL;
sds result = NULL;
lockRead(sets);
if (NULL != (entry = dictGetRandomKey(sets)))
{
result = (sds) entry->key;
}
unlockRead(sets);
return result;
}
/* Return random element from a non empty set.
* The returned element can be a int64_t value if the set is encoded
* as an "intset" blob of integers, or a redis object if the set
* is a regular set.
*
* The caller provides both pointers to be populated with the right
* object. The return value of the function is the object->encoding
* field of the object and is used by the caller to check if the
* int64_t pointer or the redis object pointer was populated.
*
* Note that both the objele and llele pointers should be passed and cannot
* be NULL since the function will try to defensively populate the non
* used field with values which are easy to trap if misused.
*
* When an object is returned (the set was a real set) the ref count
* of the object is not incremented so this function can be considered
* copy on write friendly. */
int setTypeRandomElement(robj *setobj, robj **objele, int64_t *llele) {
if (setobj->encoding == REDIS_ENCODING_HT) {
dictEntry *de = dictGetRandomKey(setobj->ptr);
*objele = dictGetKey(de);
*llele = -123456789; /* Not needed. Defensive. */
} else if (setobj->encoding == REDIS_ENCODING_INTSET) {
*llele = intsetRandom(setobj->ptr);
*objele = NULL; /* Not needed. Defensive. */
} else {
redisPanic("Unknown set encoding");
}
return setobj->encoding;
}
/* Try to share an object against the shared objects pool */
static robj *tryObjectSharing(robj *o) {
struct dictEntry *de;
unsigned long c;
if (o == NULL || server.shareobjects == 0) return o;
de = dictFind(server.sharingpool,o);
if (de) {
robj *shared = dictGetEntryKey(de);
c = ((unsigned long) dictGetEntryVal(de))+1;
dictGetEntryVal(de) = (void*) c;
incrRefCount(shared);
decrRefCount(o);
return shared;
} else {
/* Here we are using a stream algorihtm: Every time an object is
* shared we increment its count, everytime there is a miss we
* recrement the counter of a random object. If this object reaches
* zero we remove the object and put the current object instead. */
if (dictSize(server.sharingpool) >=
server.sharingpoolsize) {
de = dictGetRandomKey(server.sharingpool);
c = ((unsigned long) dictGetEntryVal(de))-1;
dictGetEntryVal(de) = (void*) c;
if (c == 0) {
dictDelete(server.sharingpool,de->key);
}
} else {
c = 0; /* If the pool is empty we want to add this object */
}
if (c == 0) {
int retval;
retval = dictAdd(server.sharingpool,o,(void*)1);
redisAssert(retval == DICT_OK);
incrRefCount(o);
}
return o;
}
}
/* Return a random key, in form of a Redis object.
* If there are no keys, NULL is returned.
*随机key从数据库中
* The function makes sure to return keys not already expired. */
robj *dbRandomKey(redisDb *db) {
struct dictEntry *de;
while(1) {
sds key;
robj *keyobj;
de = dictGetRandomKey(db->dict);
if (de == NULL) return NULL;
key = dictGetKey(de);
keyobj = createStringObject(key,sdslen(key));
if (dictFind(db->expires,key)) {//判断键是否过期
if (expireIfNeeded(db,keyobj)) {//如果过期
decrRefCount(keyobj);//减少key的引用计数
continue; /* search for another key. This expired. */
}
}
return keyobj;
}
}
/* Return a random key, in form of a Redis object.
* If there are no keys, NULL is returned.
*
* The function makes sure to return keys not already expired. */
robj *dbRandomKey(redisDb *db) {
struct dictEntry *de;
while(1) {
sds key;
robj *keyobj;
de = dictGetRandomKey(db->dict);
if (de == NULL) return NULL;
key = dictGetEntryKey(de);
keyobj = createStringObject(key,sdslen(key),sdslogiclock(key),sdsversion(key));
if (dictFind(db->expires,key)) {
if (expireIfNeeded(db,keyobj)) {
decrRefCount(keyobj);
continue; /* search for another key. This expired. */
}
}
return keyobj;
}
}
void srandmemberWithCountCommand(redisClient *c) {
long l;
unsigned long count, size;
int uniq = 1;
robj *set, *ele;
int64_t llele;
int encoding;
dict *d;
if (getLongFromObjectOrReply(c,c->argv[2],&l,NULL) != REDIS_OK) return;
if (l >= 0) {
count = (unsigned) l;
} else {
/* A negative count means: return the same elements multiple times
* (i.e. don't remove the extracted element after every extraction). */
count = -l;
uniq = 0;
}
if ((set = lookupKeyReadOrReply(c,c->argv[1],shared.emptymultibulk))
== NULL || checkType(c,set,REDIS_SET)) return;
size = setTypeSize(set);
/* If count is zero, serve it ASAP to avoid special cases later. */
if (count == 0) {
addReply(c,shared.emptymultibulk);
return;
}
/* CASE 1: The count was negative, so the extraction method is just:
* "return N random elements" sampling the whole set every time.
* This case is trivial and can be served without auxiliary data
* structures. */
if (!uniq) {
addReplyMultiBulkLen(c,count);
while(count--) {
encoding = setTypeRandomElement(set,&ele,&llele);
if (encoding == REDIS_ENCODING_INTSET) {
addReplyBulkLongLong(c,llele);
} else {
addReplyBulk(c,ele);
}
}
return;
}
/* CASE 2:
* The number of requested elements is greater than the number of
* elements inside the set: simply return the whole set. */
if (count >= size) {
sunionDiffGenericCommand(c,c->argv+1,1,NULL,REDIS_OP_UNION);
return;
}
/* For CASE 3 and CASE 4 we need an auxiliary dictionary. */
d = dictCreate(&setDictType,NULL);
/* CASE 3:
* The number of elements inside the set is not greater than
* SRANDMEMBER_SUB_STRATEGY_MUL times the number of requested elements.
* In this case we create a set from scratch with all the elements, and
* subtract random elements to reach the requested number of elements.
*
* This is done because if the number of requsted elements is just
* a bit less than the number of elements in the set, the natural approach
* used into CASE 3 is highly inefficient. */
if (count*SRANDMEMBER_SUB_STRATEGY_MUL > size) {
setTypeIterator *si;
/* Add all the elements into the temporary dictionary. */
si = setTypeInitIterator(set);
while((encoding = setTypeNext(si,&ele,&llele)) != -1) {
int retval = DICT_ERR;
if (encoding == REDIS_ENCODING_INTSET) {
retval = dictAdd(d,createStringObjectFromLongLong(llele),NULL);
} else if (ele->encoding == REDIS_ENCODING_RAW) {
retval = dictAdd(d,dupStringObject(ele),NULL);
} else if (ele->encoding == REDIS_ENCODING_INT) {
retval = dictAdd(d,
createStringObjectFromLongLong((long)ele->ptr),NULL);
}
redisAssert(retval == DICT_OK);
}
setTypeReleaseIterator(si);
redisAssert(dictSize(d) == size);
/* Remove random elements to reach the right count. */
while(size > count) {
dictEntry *de;
de = dictGetRandomKey(d);
dictDelete(d,dictGetKey(de));
size--;
}
}
/* CASE 4: We have a big set compared to the requested number of elements.
* In this case we can simply get random elements from the set and add
* to the temporary set, trying to eventually get enough unique elements
* to reach the specified count. */
else {
unsigned long added = 0;
while(added < count) {
encoding = setTypeRandomElement(set,&ele,&llele);
if (encoding == REDIS_ENCODING_INTSET) {
ele = createStringObjectFromLongLong(llele);
} else if (ele->encoding == REDIS_ENCODING_RAW) {
ele = dupStringObject(ele);
} else if (ele->encoding == REDIS_ENCODING_INT) {
ele = createStringObjectFromLongLong((long)ele->ptr);
}
/* Try to add the object to the dictionary. If it already exists
* free it, otherwise increment the number of objects we have
* in the result dictionary. */
if (dictAdd(d,ele,NULL) == DICT_OK)
added++;
else
decrRefCount(ele);
}
}
/* CASE 3 & 4: send the result to the user. */
{
dictIterator *di;
dictEntry *de;
addReplyMultiBulkLen(c,count);
di = dictGetIterator(d);
while((de = dictNext(di)) != NULL)
addReplyBulk(c,dictGetKey(de));
dictReleaseIterator(di);
dictRelease(d);
}
}
int freeMemoryIfNeeded(void) {
size_t mem_reported, mem_used, mem_tofree, mem_freed;
int slaves = listLength(server.slaves);
mstime_t latency, eviction_latency;
long long delta;
/* Check if we are over the memory usage limit. If we are not, no need
* to subtract the slaves output buffers. We can just return ASAP. */
mem_reported = zmalloc_used_memory();
if (mem_reported <= server.maxmemory) return C_OK;
/* Remove the size of slaves output buffers and AOF buffer from the
* count of used memory. */
mem_used = mem_reported;
if (slaves) {
listIter li;
listNode *ln;
listRewind(server.slaves,&li);
while((ln = listNext(&li))) {
client *slave = listNodeValue(ln);
unsigned long obuf_bytes = getClientOutputBufferMemoryUsage(slave);
if (obuf_bytes > mem_used)
mem_used = 0;
else
mem_used -= obuf_bytes;
}
}
if (server.aof_state != AOF_OFF) {
mem_used -= sdslen(server.aof_buf);
mem_used -= aofRewriteBufferSize();
}
/* Check if we are still over the memory limit. */
if (mem_used <= server.maxmemory) return C_OK;
/* Compute how much memory we need to free. */
mem_tofree = mem_used - server.maxmemory;
mem_freed = 0;
if (server.maxmemory_policy == MAXMEMORY_NO_EVICTION)
goto cant_free; /* We need to free memory, but policy forbids. */
latencyStartMonitor(latency);
while (mem_freed < mem_tofree) {
int j, k, i, keys_freed = 0;
static int next_db = 0;
sds bestkey = NULL;
int bestdbid;
redisDb *db;
dict *dict;
dictEntry *de;
if (server.maxmemory_policy == MAXMEMORY_ALLKEYS_LRU ||
server.maxmemory_policy == MAXMEMORY_VOLATILE_LRU)
{
struct evictionPoolEntry *pool = EvictionPoolLRU;
while(bestkey == NULL) {
unsigned long total_keys = 0, keys;
/* We don't want to make local-db choices when expiring keys,
* so to start populate the eviction pool sampling keys from
* every DB. */
for (i = 0; i < server.dbnum; i++) {
db = server.db+i;
dict = (server.maxmemory_policy == MAXMEMORY_ALLKEYS_LRU) ?
db->dict : db->expires;
if ((keys = dictSize(dict)) != 0) {
evictionPoolPopulate(i, dict, db->dict, pool);
total_keys += keys;
}
}
if (!total_keys) break; /* No keys to evict. */
/* Go backward from best to worst element to evict. */
for (k = EVPOOL_SIZE-1; k >= 0; k--) {
if (pool[k].key == NULL) continue;
bestdbid = pool[k].dbid;
if (server.maxmemory_policy == MAXMEMORY_ALLKEYS_LRU) {
de = dictFind(server.db[pool[k].dbid].dict,
pool[k].key);
} else {
de = dictFind(server.db[pool[k].dbid].expires,
pool[k].key);
}
/* Remove the entry from the pool. */
if (pool[k].key != pool[k].cached)
sdsfree(pool[k].key);
pool[k].key = NULL;
pool[k].idle = 0;
/* If the key exists, is our pick. Otherwise it is
* a ghost and we need to try the next element. */
if (de) {
bestkey = dictGetKey(de);
break;
} else {
/* Ghost... Iterate again. */
}
}
}
}
/* volatile-random and allkeys-random policy */
else if (server.maxmemory_policy == MAXMEMORY_ALLKEYS_RANDOM ||
server.maxmemory_policy == MAXMEMORY_VOLATILE_RANDOM)
{
/* When evicting a random key, we try to evict a key for
* each DB, so we use the static 'next_db' variable to
* incrementally visit all DBs. */
for (i = 0; i < server.dbnum; i++) {
j = (++next_db) % server.dbnum;
db = server.db+j;
dict = (server.maxmemory_policy == MAXMEMORY_ALLKEYS_RANDOM) ?
db->dict : db->expires;
if (dictSize(dict) != 0) {
de = dictGetRandomKey(dict);
bestkey = dictGetKey(de);
bestdbid = j;
break;
}
}
}
/* volatile-ttl */
else if (server.maxmemory_policy == MAXMEMORY_VOLATILE_TTL) {
long bestttl = 0; /* Initialized to avoid warning. */
/* In this policy we scan a single DB per iteration (visiting
* a different DB per call), expiring the key with the smallest
* TTL among the few sampled.
*
* Note that this algorithm makes local-DB choices, and should
* use a pool and code more similr to the one used in the
* LRU eviction policies in the future. */
for (i = 0; i < server.dbnum; i++) {
j = (++next_db) % server.dbnum;
db = server.db+j;
dict = db->expires;
if (dictSize(dict) != 0) {
for (k = 0; k < server.maxmemory_samples; k++) {
sds thiskey;
long thisttl;
de = dictGetRandomKey(dict);
thiskey = dictGetKey(de);
thisttl = (long) dictGetVal(de);
/* Keys expiring sooner (smaller unix timestamp) are
* better candidates for deletion */
if (bestkey == NULL || thisttl < bestttl) {
bestkey = thiskey;
bestttl = thisttl;
bestdbid = j;
}
}
}
}
}
/* Finally remove the selected key. */
if (bestkey) {
db = server.db+bestdbid;
robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
propagateExpire(db,keyobj,server.lazyfree_lazy_eviction);
/* We compute the amount of memory freed by db*Delete() alone.
* It is possible that actually the memory needed to propagate
* the DEL in AOF and replication link is greater than the one
* we are freeing removing the key, but we can't account for
* that otherwise we would never exit the loop.
*
* AOF and Output buffer memory will be freed eventually so
* we only care about memory used by the key space. */
delta = (long long) zmalloc_used_memory();
latencyStartMonitor(eviction_latency);
if (server.lazyfree_lazy_eviction)
dbAsyncDelete(db,keyobj);
else
dbSyncDelete(db,keyobj);
latencyEndMonitor(eviction_latency);
latencyAddSampleIfNeeded("eviction-del",eviction_latency);
latencyRemoveNestedEvent(latency,eviction_latency);
delta -= (long long) zmalloc_used_memory();
mem_freed += delta;
server.stat_evictedkeys++;
notifyKeyspaceEvent(NOTIFY_EVICTED, "evicted",
keyobj, db->id);
decrRefCount(keyobj);
keys_freed++;
/* When the memory to free starts to be big enough, we may
* start spending so much time here that is impossible to
* deliver data to the slaves fast enough, so we force the
* transmission here inside the loop. */
if (slaves) flushSlavesOutputBuffers();
}
if (!keys_freed) {
latencyEndMonitor(latency);
latencyAddSampleIfNeeded("eviction-cycle",latency);
goto cant_free; /* nothing to free... */
}
}
latencyEndMonitor(latency);
latencyAddSampleIfNeeded("eviction-cycle",latency);
return C_OK;
cant_free:
/* We are here if we are not able to reclaim memory. There is only one
* last thing we can try: check if the lazyfree thread has jobs in queue
* and wait... */
while(bioPendingJobsOfType(BIO_LAZY_FREE)) {
if (((mem_reported - zmalloc_used_memory()) + mem_freed) >= mem_tofree)
break;
usleep(1000);
}
return C_ERR;
}
/* Try to swap an object that's a good candidate for swapping.
* Returns REDIS_OK if the object was swapped, REDIS_ERR if it's not possible
* to swap any object at all.
*
* If 'usethreaded' is true, Redis will try to swap the object in background
* using I/O threads. */
int vmSwapOneObject(int usethreads) {
int j, i;
struct dictEntry *best = NULL;
double best_swappability = 0;
redisDb *best_db = NULL;
robj *val;
sds key;
for (j = 0; j < server.dbnum; j++) {
redisDb *db = server.db+j;
/* Why maxtries is set to 100?
* Because this way (usually) we'll find 1 object even if just 1% - 2%
* are swappable objects */
int maxtries = 100;
if (dictSize(db->dict) == 0) continue;
for (i = 0; i < 5; i++) {
dictEntry *de;
double swappability;
if (maxtries) maxtries--;
de = dictGetRandomKey(db->dict);
val = dictGetEntryVal(de);
/* Only swap objects that are currently in memory.
*
* Also don't swap shared objects: not a good idea in general and
* we need to ensure that the main thread does not touch the
* object while the I/O thread is using it, but we can't
* control other keys without adding additional mutex. */
if (val->storage != REDIS_VM_MEMORY || val->refcount != 1) {
if (maxtries) i--; /* don't count this try */
continue;
}
swappability = computeObjectSwappability(val);
if (!best || swappability > best_swappability) {
best = de;
best_swappability = swappability;
best_db = db;
}
}
}
if (best == NULL) return REDIS_ERR;
key = dictGetEntryKey(best);
val = dictGetEntryVal(best);
redisLog(REDIS_DEBUG,"Key with best swappability: %s, %f",
key, best_swappability);
/* Swap it */
if (usethreads) {
robj *keyobj = createStringObject(key,sdslen(key));
vmSwapObjectThreaded(keyobj,val,best_db);
decrRefCount(keyobj);
return REDIS_OK;
} else {
vmpointer *vp;
if ((vp = vmSwapObjectBlocking(val)) != NULL) {
dictGetEntryVal(best) = vp;
return REDIS_OK;
} else {
return REDIS_ERR;
}
}
}
/* How a good candidate is this object for swapping?
* The better candidate it is, the greater the returned value.
*
* Currently we try to perform a fast estimation of the object size in
* memory, and combine it with aging informations.
*
* Basically swappability = idle-time * log(estimated size)
*
* Bigger objects are preferred over smaller objects, but not
* proportionally, this is why we use the logarithm. This algorithm is
* just a first try and will probably be tuned later. */
double computeObjectSwappability(robj *o) {
/* actual age can be >= minage, but not < minage. As we use wrapping
* 21 bit clocks with minutes resolution for the LRU. */
time_t minage = estimateObjectIdleTime(o);
#ifdef _WIN32
ssize_t asize = 0, elesize;
#else
long asize = 0, elesize;
#endif
robj *ele;
list *l;
listNode *ln;
dict *d;
struct dictEntry *de;
if (minage <= 0) return 0;
switch(o->type) {
case REDIS_STRING:
if (o->encoding != REDIS_ENCODING_RAW) {
asize = sizeof(*o);
} else {
#ifdef _WIN32
asize = sdslen(o->ptr)+sizeof(*o)+sizeof(size_t)*2;
#else
asize = sdslen(o->ptr)+sizeof(*o)+sizeof(long)*2;
#endif
}
break;
case REDIS_LIST:
if (o->encoding == REDIS_ENCODING_ZIPLIST) {
asize = sizeof(*o)+ziplistBlobLen(o->ptr);
} else {
l = o->ptr;
ln = listFirst(l);
asize = sizeof(list);
if (ln) {
ele = ln->value;
elesize = (ele->encoding == REDIS_ENCODING_RAW) ?
(sizeof(*o)+sdslen(ele->ptr)) : sizeof(*o);
asize += (sizeof(listNode)+elesize)*listLength(l);
}
}
break;
case REDIS_SET:
if (o->encoding == REDIS_ENCODING_INTSET) {
intset *is = o->ptr;
asize = sizeof(*is)+is->encoding*is->length;
} else {
d = o->ptr;
asize = sizeof(dict)+(sizeof(struct dictEntry*)*dictSlots(d));
if (dictSize(d)) {
de = dictGetRandomKey(d);
ele = dictGetEntryKey(de);
elesize = (ele->encoding == REDIS_ENCODING_RAW) ?
(sizeof(*o)+sdslen(ele->ptr)) : sizeof(*o);
asize += (sizeof(struct dictEntry)+elesize)*dictSize(d);
}
}
break;
case REDIS_ZSET:
if (o->encoding == REDIS_ENCODING_ZIPLIST) {
asize = sizeof(*o)+(ziplistBlobLen(o->ptr) / 2);
} else {
d = ((zset*)o->ptr)->dict;
asize = sizeof(zset)+(sizeof(struct dictEntry*)*dictSlots(d));
if (dictSize(d)) {
de = dictGetRandomKey(d);
ele = dictGetEntryKey(de);
elesize = (ele->encoding == REDIS_ENCODING_RAW) ?
(sizeof(*o)+sdslen(ele->ptr)) : sizeof(*o);
asize += (sizeof(struct dictEntry)+elesize)*dictSize(d);
asize += sizeof(zskiplistNode)*dictSize(d);
}
}
break;
case REDIS_HASH:
if (o->encoding == REDIS_ENCODING_ZIPMAP) {
unsigned char *p = zipmapRewind((unsigned char*)o->ptr);
unsigned int len = zipmapLen((unsigned char*)o->ptr);
unsigned int klen, vlen;
unsigned char *key, *val;
if ((p = zipmapNext(p,&key,&klen,&val,&vlen)) == NULL) {
klen = 0;
vlen = 0;
}
asize = len*(klen+vlen+3);
} else if (o->encoding == REDIS_ENCODING_HT) {
d = o->ptr;
asize = sizeof(dict)+(sizeof(struct dictEntry*)*dictSlots(d));
if (dictSize(d)) {
de = dictGetRandomKey(d);
ele = dictGetEntryKey(de);
elesize = (ele->encoding == REDIS_ENCODING_RAW) ?
(sizeof(*o)+sdslen(ele->ptr)) : sizeof(*o);
ele = dictGetEntryVal(de);
elesize = (ele->encoding == REDIS_ENCODING_RAW) ?
(sizeof(*o)+sdslen(ele->ptr)) : sizeof(*o);
asize += (sizeof(struct dictEntry)+elesize)*dictSize(d);
}
}
break;
}
return (double)minage*log(1+(double)asize);
}
int freeMemoryIfNeeded(void) {
size_t mem_reported, mem_used, mem_tofree, mem_freed;
mstime_t latency, eviction_latency;
long long delta;
int slaves = listLength(server.slaves);
/* When clients are paused the dataset should be static not just from the
* POV of clients not being able to write, but also from the POV of
* expires and evictions of keys not being performed. */
if (clientsArePaused()) return C_OK;
/* Check if we are over the memory usage limit. If we are not, no need
* to subtract the slaves output buffers. We can just return ASAP. */
mem_reported = zmalloc_used_memory();
if (mem_reported <= server.maxmemory) return C_OK;
/* Remove the size of slaves output buffers and AOF buffer from the
* count of used memory. */
mem_used = mem_reported;
size_t overhead = freeMemoryGetNotCountedMemory();
mem_used = (mem_used > overhead) ? mem_used-overhead : 0;
/* Check if we are still over the memory limit. */
if (mem_used <= server.maxmemory) return C_OK;
/* Compute how much memory we need to free. */
mem_tofree = mem_used - server.maxmemory;
mem_freed = 0;
if (server.maxmemory_policy == MAXMEMORY_NO_EVICTION)
goto cant_free; /* We need to free memory, but policy forbids. */
latencyStartMonitor(latency);
while (mem_freed < mem_tofree) {
int j, k, i, keys_freed = 0;
static int next_db = 0;
sds bestkey = NULL;
int bestdbid;
redisDb *db;
dict *dict;
dictEntry *de;
if (server.maxmemory_policy & (MAXMEMORY_FLAG_LRU|MAXMEMORY_FLAG_LFU) ||
server.maxmemory_policy == MAXMEMORY_VOLATILE_TTL)
{
struct evictionPoolEntry *pool = EvictionPoolLRU;
while(bestkey == NULL) {
unsigned long total_keys = 0, keys;
/* We don't want to make local-db choices when expiring keys,
* so to start populate the eviction pool sampling keys from
* every DB. */
for (i = 0; i < server.dbnum; i++) {
db = server.db+i;
dict = (server.maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS) ?
db->dict : db->expires;
if ((keys = dictSize(dict)) != 0) {
evictionPoolPopulate(i, dict, db->dict, pool);
total_keys += keys;
}
}
if (!total_keys) break; /* No keys to evict. */
/* Go backward from best to worst element to evict. */
for (k = EVPOOL_SIZE-1; k >= 0; k--) {
if (pool[k].key == NULL) continue;
bestdbid = pool[k].dbid;
if (server.maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS) {
de = dictFind(server.db[pool[k].dbid].dict,
pool[k].key);
} else {
de = dictFind(server.db[pool[k].dbid].expires,
pool[k].key);
}
/* Remove the entry from the pool. */
if (pool[k].key != pool[k].cached)
sdsfree(pool[k].key);
pool[k].key = NULL;
pool[k].idle = 0;
/* If the key exists, is our pick. Otherwise it is
* a ghost and we need to try the next element. */
if (de) {
bestkey = dictGetKey(de);
break;
} else {
/* Ghost... Iterate again. */
}
}
}
}
/* volatile-random and allkeys-random policy */
else if (server.maxmemory_policy == MAXMEMORY_ALLKEYS_RANDOM ||
server.maxmemory_policy == MAXMEMORY_VOLATILE_RANDOM)
{
/* When evicting a random key, we try to evict a key for
* each DB, so we use the static 'next_db' variable to
* incrementally visit all DBs. */
for (i = 0; i < server.dbnum; i++) {
j = (++next_db) % server.dbnum;
db = server.db+j;
dict = (server.maxmemory_policy == MAXMEMORY_ALLKEYS_RANDOM) ?
db->dict : db->expires;
if (dictSize(dict) != 0) {
de = dictGetRandomKey(dict);
bestkey = dictGetKey(de);
bestdbid = j;
break;
}
}
}
/* Finally remove the selected key. */
if (bestkey) {
db = server.db+bestdbid;
robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
propagateExpire(db,keyobj,server.lazyfree_lazy_eviction);
/* We compute the amount of memory freed by db*Delete() alone.
* It is possible that actually the memory needed to propagate
* the DEL in AOF and replication link is greater than the one
* we are freeing removing the key, but we can't account for
* that otherwise we would never exit the loop.
*
* AOF and Output buffer memory will be freed eventually so
* we only care about memory used by the key space. */
delta = (long long) zmalloc_used_memory();
latencyStartMonitor(eviction_latency);
if (server.lazyfree_lazy_eviction)
dbAsyncDelete(db,keyobj);
else
dbSyncDelete(db,keyobj);
latencyEndMonitor(eviction_latency);
latencyAddSampleIfNeeded("eviction-del",eviction_latency);
latencyRemoveNestedEvent(latency,eviction_latency);
delta -= (long long) zmalloc_used_memory();
mem_freed += delta;
server.stat_evictedkeys++;
notifyKeyspaceEvent(NOTIFY_EVICTED, "evicted",
keyobj, db->id);
decrRefCount(keyobj);
keys_freed++;
/* When the memory to free starts to be big enough, we may
* start spending so much time here that is impossible to
* deliver data to the slaves fast enough, so we force the
* transmission here inside the loop. */
if (slaves) flushSlavesOutputBuffers();
/* Normally our stop condition is the ability to release
* a fixed, pre-computed amount of memory. However when we
* are deleting objects in another thread, it's better to
* check, from time to time, if we already reached our target
* memory, since the "mem_freed" amount is computed only
* across the dbAsyncDelete() call, while the thread can
* release the memory all the time. */
if (server.lazyfree_lazy_eviction && !(keys_freed % 16)) {
overhead = freeMemoryGetNotCountedMemory();
mem_used = zmalloc_used_memory();
mem_used = (mem_used > overhead) ? mem_used-overhead : 0;
if (mem_used <= server.maxmemory) {
mem_freed = mem_tofree;
}
}
}
if (!keys_freed) {
latencyEndMonitor(latency);
latencyAddSampleIfNeeded("eviction-cycle",latency);
goto cant_free; /* nothing to free... */
}
}
latencyEndMonitor(latency);
latencyAddSampleIfNeeded("eviction-cycle",latency);
return C_OK;
cant_free:
/* We are here if we are not able to reclaim memory. There is only one
* last thing we can try: check if the lazyfree thread has jobs in queue
* and wait... */
while(bioPendingJobsOfType(BIO_LAZY_FREE)) {
if (((mem_reported - zmalloc_used_memory()) + mem_freed) >= mem_tofree)
break;
usleep(1000);
}
return C_ERR;
}
