/** wrapper around assoc_find which does the lazy expiration/deletion logic */
item *do_item_get_notedeleted(const char *key, const size_t nkey, bool *delete_locked) {
item *it = assoc_find(key, nkey);
int was_found = 0;
if (delete_locked) *delete_locked = false;
if (settings.verbose > 2) {
if (it == NULL) {
fprintf(stderr, "> NOT FOUND %s", key);
} else {
fprintf(stderr, "> FOUND KEY %s", ITEM_key(it));
was_found++;
}
}
if (it != NULL && (it->it_flags & ITEM_DELETED)) {
/* it's flagged as delete-locked. let's see if that condition
is past due, and the 5-second delete_timer just hasn't
gotten to it yet... */
if (!item_delete_lock_over(it)) {
if (delete_locked) *delete_locked = true;
it = NULL;
}
}
if (it == NULL && was_found) {
fprintf(stderr, " -nuked by delete lock");
was_found--;
}
if (it != NULL && settings.oldest_live != 0 && settings.oldest_live <= current_time &&
it->time <= settings.oldest_live) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it == NULL && was_found) {
fprintf(stderr, " -nuked by flush");
was_found--;
}
if (it != NULL && it->exptime != 0 && it->exptime <= current_time) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it == NULL && was_found) {
fprintf(stderr, " -nuked by expire");
was_found--;
}
if (it != NULL) {
it->refcount++;
DEBUG_REFCNT(it, '+');
}
if (settings.verbose > 2)
fprintf(stderr, "\n");
return it;
}
/** wrapper around assoc_find which does the lazy expiration logic */
hash_item *do_item_get(struct default_engine *engine,
const hash_key *key) {
rel_time_t current_time = engine->server.core->get_current_time();
hash_item *it = assoc_find(engine,
crc32c(hash_key_get_key(key),
hash_key_get_key_len(key), 0),
key);
int was_found = 0;
if (engine->config.verbose > 2) {
EXTENSION_LOGGER_DESCRIPTOR *logger;
logger = (void*)engine->server.extension->get_extension(EXTENSION_LOGGER);
if (it == NULL) {
logger->log(EXTENSION_LOG_DEBUG, NULL,
"> NOT FOUND in bucket %d, %s",
hash_key_get_bucket_index(key),
hash_key_get_client_key(key));
} else {
logger->log(EXTENSION_LOG_DEBUG, NULL,
"> FOUND KEY in bucket %d, %s",
hash_key_get_bucket_index(item_get_key(it)),
hash_key_get_client_key(item_get_key(it)));
was_found++;
}
}
if (it != NULL && engine->config.oldest_live != 0 &&
engine->config.oldest_live <= current_time &&
it->time <= engine->config.oldest_live) {
do_item_unlink(engine, it); /* MTSAFE - items.lock held */
it = NULL;
}
if (it == NULL && was_found) {
EXTENSION_LOGGER_DESCRIPTOR *logger;
logger = (void*)engine->server.extension->get_extension(EXTENSION_LOGGER);
logger->log(EXTENSION_LOG_DEBUG, NULL, " -nuked by flush");
was_found--;
}
if (it != NULL && it->exptime != 0 && it->exptime <= current_time) {
do_item_unlink(engine, it); /* MTSAFE - items.lock held */
it = NULL;
}
if (it == NULL && was_found) {
EXTENSION_LOGGER_DESCRIPTOR *logger;
logger = (void*)engine->server.extension->get_extension(EXTENSION_LOGGER);
logger->log(EXTENSION_LOG_DEBUG, NULL, " -nuked by expire");
was_found--;
}
if (it != NULL) {
it->refcount++;
DEBUG_REFCNT(it, '+');
do_item_update(engine, it);
}
return it;
}
/** wrapper around assoc_find which does the lazy expiration/deletion logic */
item *do_item_get_notedeleted(const char *key, const size_t nkey, bool *delete_locked) {
item *it = assoc_find(key, nkey);
if (delete_locked) *delete_locked = false;
if (it != NULL && (it->it_flags & ITEM_DELETED)) {
/* it's flagged as delete-locked. let's see if that condition
is past due, and the 5-second delete_timer just hasn't
gotten to it yet... */
if (!item_delete_lock_over(it)) {
if (delete_locked) *delete_locked = true;
it = NULL;
}
}
if (it != NULL && settings.oldest_live != 0 && settings.oldest_live <= current_time &&
it->time <= settings.oldest_live) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it != NULL && it->exptime != 0 && it->exptime <= current_time) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it != NULL) {
it->refcount++;
DEBUG_REFCNT(it, '+');
}
return it;
}
/** wrapper around assoc_find which does the lazy expiration logic */
item *do_item_get(const char *key, const size_t nkey, const uint32_t hv) {
mutex_lock(&cache_lock);
item *it = assoc_find(key, nkey, hv);
if (it != NULL) {
refcount_incr(&it->refcount);
/* Optimization for slab reassignment. prevents popular items from
* jamming in busy wait. Can only do this here to satisfy lock order
* of item_lock, cache_lock, slabs_lock. */
if (slab_rebalance_signal &&
((void *)it >= slab_rebal.slab_start && (void *)it < slab_rebal.slab_end)) {
do_item_unlink_nolock(it, hv);
do_item_remove(it);
it = NULL;
}
}
pthread_mutex_unlock(&cache_lock);
int was_found = 0;
if (settings.verbose > 2) {
if (it == NULL) {
fprintf(stderr, "> NOT FOUND %s", key);
} else {
fprintf(stderr, "> FOUND KEY %s", ITEM_key(it));
was_found++;
}
}
if (it != NULL) {
if (settings.oldest_live != 0 && settings.oldest_live <= current_time &&
it->time <= settings.oldest_live) {
do_item_unlink(it, hv);
do_item_remove(it);
it = NULL;
if (was_found) {
fprintf(stderr, " -nuked by flush");
}
} else if (it->exptime != 0 && it->exptime <= current_time) {
if (it->exptime + 10 < current_time) {
do_item_unlink(it, hv);
do_item_remove(it);
if (was_found) {
fprintf(stderr, " -nuked by expire");
}
} else {
/* re-active just expired items, to anti miss-storm */
it->exptime = current_time + 10;
}
it = NULL;
} else {
it->it_flags |= ITEM_FETCHED;
DEBUG_REFCNT(it, '+');
}
}
if (settings.verbose > 2)
fprintf(stderr, "\n");
return it;
}
static hash_item *do_item_get(struct demo_engine *engine,
const char *key, const size_t nkey,
bool LRU_reposition)
{
rel_time_t current_time = engine->server.core->get_current_time();
hash_item *it = dm_assoc_find(engine, engine->server.core->hash(key, nkey, 0),
key, nkey);
if (it != NULL) {
if (do_item_isvalid(engine, it, current_time)==false) {
do_item_unlink(engine, it, ITEM_UNLINK_INVALID);
it = NULL;
}
}
if (it != NULL) {
ITEM_REFCOUNT_INCR(it);
DEBUG_REFCNT(it, '+');
}
if (engine->config.verbose > 2) {
if (it == NULL) {
logger->log(EXTENSION_LOG_INFO, NULL, "> NOT FOUND %s\n",
key);
} else {
logger->log(EXTENSION_LOG_INFO, NULL, "> FOUND KEY %s\n",
(const char*)dm_item_get_key(it));
}
}
return it;
}
/*
* Unlinks an item from the LRU and hashtable.
*/
void item_unlink(item *item) {
uint32_t hv;
hv = hash(ITEM_key(item), item->nkey, 0);
item_lock(hv);
do_item_unlink(item, hv);
item_unlock(hv);
}
/*
* adds a delta value to a numeric item.
*
* c connection requesting the operation
* it item to adjust
* incr true to increment value, false to decrement
* delta amount to adjust value by
* @param ritem The resulting item after adding the delta. Only valid if
* ENGINE_SUCCESS is returned. Caller is responsible for calling
* do_item_release() on this when finished with it.
*
* returns a response code to send back to the client.
*/
static ENGINE_ERROR_CODE do_add_delta(struct default_engine *engine,
hash_item *it, const bool incr,
const int64_t delta, item** ritem,
uint64_t *result, const void *cookie) {
const char *ptr;
uint64_t value;
char buf[80];
int res;
if (it->nbytes >= (sizeof(buf) - 1)) {
return ENGINE_EINVAL;
}
ptr = item_get_data(it);
memcpy(buf, ptr, it->nbytes);
buf[it->nbytes] = '\0';
if (!safe_strtoull(buf, &value)) {
return ENGINE_EINVAL;
}
if (incr) {
value += delta;
} else {
if ((uint64_t)delta > value) {
value = 0;
} else {
value -= delta;
}
}
*result = value;
res = snprintf(buf, sizeof(buf), "%" PRIu64, value);
if (res < 0 || res >= sizeof(buf)) {
return ENGINE_EINVAL;
}
if (it->refcount == 1 && res <= (int)it->nbytes) {
/* we can do inline replacement */
memcpy(item_get_data(it), buf, res);
memset(item_get_data(it) + res, ' ', it->nbytes - res);
item_set_cas(NULL, NULL, it, get_cas_id());
*ritem = it;
} else {
hash_item *new_it = do_item_alloc(engine, item_get_key(it),
it->flags,
it->exptime, res,
cookie, it->datatype);
if (new_it == NULL) {
do_item_unlink(engine, it);
return ENGINE_ENOMEM;
}
memcpy(item_get_data(new_it), buf, res);
do_item_replace(engine, it, new_it);
*ritem = new_it;
}
return ENGINE_SUCCESS;
}
/*
* Flushes expired items after a flush_all call
*/
void item_flush_expired(struct default_engine *engine) {
cb_mutex_enter(&engine->items.lock);
rel_time_t now = engine->server.core->get_current_time();
if (now > engine->config.oldest_live) {
engine->config.oldest_live = now - 1;
}
for (int ii = 0; ii < POWER_LARGEST; ii++) {
hash_item *iter, *next;
/*
* The LRU is sorted in decreasing time order, and an item's
* timestamp is never newer than its last access time, so we
* only need to walk back until we hit an item older than the
* oldest_live time.
* The oldest_live checking will auto-expire the remaining items.
*/
for (iter = engine->items.heads[ii]; iter != NULL; iter = next) {
if (iter->time >= engine->config.oldest_live) {
next = iter->next;
if ((iter->iflag & ITEM_SLABBED) == 0) {
do_item_unlink(engine, iter);
}
} else {
/* We've hit the first old item. Continue to the next queue. */
break;
}
}
}
cb_mutex_exit(&engine->items.lock);
}
static ENGINE_ERROR_CODE item_scrub(struct default_engine *engine,
hash_item *item,
void *cookie) {
rel_time_t current_time = engine->server.core->get_current_time();
(void)cookie;
engine->scrubber.visited++;
/*
scrubber is used for generic bucket deletion and scrub_cmd
all expired or orphaned items are unlinked
*/
if (engine->scrubber.force_delete && item->refcount > 0) {
// warn that someone isn't releasing items before deleting their bucket.
EXTENSION_LOGGER_DESCRIPTOR* logger;
logger = (void*)engine->server.extension->get_extension(EXTENSION_LOGGER);
logger->log(EXTENSION_LOG_WARNING, NULL,
"Bucket (%d) deletion is removing an item with refcount %d",
engine->bucket_id,
item->refcount);
}
if (engine->scrubber.force_delete || (item->refcount == 0 &&
(item->exptime != 0 && item->exptime < current_time))) {
do_item_unlink(engine, item);
engine->scrubber.cleaned++;
}
return ENGINE_SUCCESS;
}
int do_item_replace(item *it, item *new_it) {
MEMCACHED_ITEM_REPLACE(ITEM_key(it), it->nkey, it->nbytes,
ITEM_key(new_it), new_it->nkey, new_it->nbytes);
assert((it->it_flags & ITEM_SLABBED) == 0);
do_item_unlink(it);
return do_item_link(new_it);
}
int do_item_replace(item *it, item *new_it, const uint32_t hv) {
new_it->it_flags &= ~ITEM_FAKE_MISSED;
MEMCACHED_ITEM_REPLACE(ITEM_key(it), it->nkey, it->nbytes,
ITEM_key(new_it), new_it->nkey, new_it->nbytes);
assert((it->it_flags & ITEM_SLABBED) == 0);
do_item_unlink(it, hv);
return do_item_link(new_it, hv);
}
/** wrapper around assoc_find which does the lazy expiration logic */
item *do_item_get(const char *key, const size_t nkey) {
item *it = assoc_find(key, nkey);
if (it != NULL && settings.oldest_live != 0 && settings.oldest_live <= current_time &&
it->time <= settings.oldest_live) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it != NULL && it->exptime != 0 && it->exptime <= current_time) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it != NULL) {
it->refcount++;
DEBUG_REFCNT(it, '+');
}
return it;
}
void LRU_list::item_unlink(base_item* item) {
uint32_t hv;
hv = HashTable::hash(item->data, item->nkey);
//哈希局部锁
hashtable.hash_lock(hv);
do_item_unlink(item, hv);
hashtable.hash_unlock(hv);
}
int do_item_replace(item *it, item *new_it, const uint32_t hv) {
syslog(LOG_INFO, "[%s:%s:%d]", __FILE__, __func__, __LINE__);
MEMCACHED_ITEM_REPLACE(ITEM_key(it), it->nkey, it->nbytes,
ITEM_key(new_it), new_it->nkey, new_it->nbytes);
assert((it->it_flags & ITEM_SLABBED) == 0);
do_item_unlink(it, hv);
return do_item_link(new_it, hv);
}
int do_item_replace(struct default_engine *engine,
hash_item *it, hash_item *new_it) {
MEMCACHED_ITEM_REPLACE(item_get_key(it), it->nkey, it->nbytes,
item_get_key(new_it), new_it->nkey, new_it->nbytes);
assert((it->iflag & ITEM_SLABBED) == 0);
do_item_unlink(engine, it);
return do_item_link(engine, new_it);
}
//用新的item替换老的item
int do_item_replace(item *it, item *new_it, const uint32_t hv)
{
MEMCACHED_ITEM_REPLACE(ITEM_key(it), it->nkey, it->nbytes,
ITEM_key(new_it), new_it->nkey, new_it->nbytes);
assert((it->it_flags & ITEM_SLABBED) == 0);//判断it是已经分配过的，如果未分配，则断言失败
do_item_unlink(it, hv);//删除原来的item
return do_item_link(new_it, hv);//重新添加新的item
}
/** wrapper around assoc_find which does the lazy expiration logic */
item *do_item_get(const char *key, const size_t nkey) {
item *it = assoc_find(key, nkey);
int was_found = 0;
if (settings.verbose > 2) {
if (it == NULL) {
fprintf(stderr, "> NOT FOUND %s", key);
} else {
fprintf(stderr, "> FOUND KEY %s", ITEM_key(it));
was_found++;
}
}
if (it != NULL && settings.oldest_live != 0 && settings.oldest_live <= current_time &&
it->time <= settings.oldest_live) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it == NULL && was_found) {
fprintf(stderr, " -nuked by flush");
was_found--;
}
if (it != NULL && it->exptime != 0 && it->exptime <= current_time) {
do_item_unlink(it); /* MTSAFE - cache_lock held */
it = NULL;
}
if (it == NULL && was_found) {
fprintf(stderr, " -nuked by expire");
was_found--;
}
if (it != NULL) {
it->refcount++;
DEBUG_REFCNT(it, '+');
}
if (settings.verbose > 2)
fprintf(stderr, "\n");
return it;
}
static ENGINE_ERROR_CODE do_item_dcp_step(struct default_engine *engine,
struct dcp_connection *connection,
const void *cookie,
struct dcp_message_producers *producers)
{
ENGINE_ERROR_CODE ret = ENGINE_DISCONNECT;
while (connection->it == NULL) {
if (!do_item_walk_cursor(engine, &connection->cursor, 1,
item_dcp_iterfunc, connection, &ret)) {
/* find next slab class to look at.. */
bool linked = false;
int ii;
for (ii = connection->cursor.slabs_clsid + 1; ii < POWER_LARGEST && !linked; ++ii) {
if (engine->items.heads[ii] != NULL) {
/* add the item at the tail */
do_item_link_cursor(engine, &connection->cursor, ii);
linked = true;
}
}
if (!linked) {
break;
}
}
}
if (connection->it != NULL) {
rel_time_t current_time = engine->server.core->get_current_time();
rel_time_t exptime = connection->it->exptime;
if (exptime != 0 && exptime < current_time) {
const hash_key* key = item_get_key(connection->it);
ret = producers->expiration(cookie, connection->opaque,
hash_key_get_client_key(key),
hash_key_get_client_key_len(key),
item_get_cas(connection->it),
0, 0, 0, NULL, 0);
if (ret == ENGINE_SUCCESS) {
do_item_unlink(engine, connection->it);
do_item_release(engine, connection->it);
}
} else {
ret = producers->mutation(cookie, connection->opaque,
connection->it, 0, 0, 0, 0, NULL, 0, 0);
}
if (ret == ENGINE_SUCCESS) {
connection->it = NULL;
}
} else {
return ENGINE_DISCONNECT;
}
return ret;
}
static void do_item_replace(struct demo_engine *engine,
hash_item *it, hash_item *new_it)
{
MEMCACHED_ITEM_REPLACE(dm_item_get_key(it), it->nkey, it->nbytes,
dm_item_get_key(new_it), new_it->nkey, new_it->nbytes);
do_item_unlink(engine, it, ITEM_UNLINK_REPLACE);
/* Cache item replacement does not drop the prefix item even if it's empty.
* So, the below do_item_link function always return SUCCESS.
*/
(void)do_item_link(engine, new_it);
}
static ENGINE_ERROR_CODE item_scrub(struct default_engine *engine,
hash_item *item,
void *cookie) {
(void)cookie;
engine->scrubber.visited++;
rel_time_t current_time = engine->server.core->get_current_time();
if (item->refcount == 0 &&
(item->exptime != 0 && item->exptime < current_time)) {
do_item_unlink(engine, item);
engine->scrubber.cleaned++;
}
return ENGINE_SUCCESS;
}
void do_item_stats(ADD_STAT add_stats, void *c) {
int i;
for (i = 0; i < LARGEST_ID; i++) {
if (tails[i] != NULL) {
const char *fmt = "items:%d:%s";
char key_str[STAT_KEY_LEN];
char val_str[STAT_VAL_LEN];
int klen = 0, vlen = 0;
int search = 50;
while (search > 0 &&
tails[i] != NULL &&
((settings.oldest_live != 0 && /* Item flushd */
settings.oldest_live <= current_time &&
tails[i]->time <= settings.oldest_live) ||
(tails[i]->exptime != 0 && /* and not expired */
tails[i]->exptime < current_time))) {
--search;
if (tails[i]->refcount == 0) {
do_item_unlink(tails[i]);
} else {
break;
}
}
if (tails[i] == NULL) {
/* We removed all of the items in this slab class */
continue;
}
APPEND_NUM_FMT_STAT(fmt, i, "number", "%u", sizes[i]);
APPEND_NUM_FMT_STAT(fmt, i, "age", "%u", tails[i]->time);
APPEND_NUM_FMT_STAT(fmt, i, "evicted",
"%u", itemstats[i].evicted);
APPEND_NUM_FMT_STAT(fmt, i, "evicted_nonzero",
"%u", itemstats[i].evicted_nonzero);
APPEND_NUM_FMT_STAT(fmt, i, "evicted_time",
"%u", itemstats[i].evicted_time);
APPEND_NUM_FMT_STAT(fmt, i, "outofmemory",
"%u", itemstats[i].outofmemory);
APPEND_NUM_FMT_STAT(fmt, i, "tailrepairs",
"%u", itemstats[i].tailrepairs);;
APPEND_NUM_FMT_STAT(fmt, i, "reclaimed",
"%u", itemstats[i].reclaimed);;
}
}
/* getting here means both ascii and binary terminators fit */
add_stats(NULL, 0, NULL, 0, c);
}
int item_test() {
int maxi = 0;
//test set.
for(int i = 0; i < 10; i++) {
char key[1024];
memset(key, 0, 1024);
sprintf(key, "charlie_%d", i);
const size_t nkey = strlen(key) + 1;
const int flags = 0;
const time_t exptime = 0;
const int nbytes = 1024;
uint32_t cur_hv = jenkins_hash((void *)key, nkey);
item *it = do_item_alloc((const char *)key, nkey, flags, exptime, nbytes, cur_hv);
if(it == NULL) {
fprintf(stderr, "\033[31malloc fail\033[0m");
maxi = i;
break;
}
char val[1024];
sprintf(val, "%d", i);
memcpy(ITEM_data(it), (void *)&val, strlen(val)+1);
}
//test get.
for(int i = 0; i < 10; ++i) {
char key[1024];
memset(key, 0, 1024);
sprintf(key, "charlie_%d", i);
const size_t nkey = strlen(key) + 1;
uint32_t cur_hv = jenkins_hash((void *)key, nkey);
item *it = assoc_find(key, nkey, cur_hv);
if(it == NULL) {
fprintf(stderr, "\033[31mget fail\033[0m");
return -1;
}
int val = 0;
memcpy((void *)&val, ITEM_data(it), sizeof(val));
if(i&0x1) {
fprintf(stdout, "del key:%s value:%d\n", ITEM_key(it), val);
do_item_unlink(it, cur_hv);
lru_traverse(NULL);
}
}
return 0;
}
static void do_item_stats(struct default_engine *engine,
ADD_STAT add_stats, const void *c) {
int i;
rel_time_t current_time = engine->server.core->get_current_time();
for (i = 0; i < POWER_LARGEST; i++) {
if (engine->items.tails[i] != NULL) {
int search = search_items;
while (search > 0 &&
engine->items.tails[i] != NULL &&
((engine->config.oldest_live != 0 && /* Item flushd */
engine->config.oldest_live <= current_time &&
engine->items.tails[i]->time <= engine->config.oldest_live) ||
(engine->items.tails[i]->exptime != 0 && /* and not expired */
engine->items.tails[i]->exptime < current_time))) {
--search;
if (engine->items.tails[i]->refcount == 0) {
do_item_unlink(engine, engine->items.tails[i]);
} else {
break;
}
}
if (engine->items.tails[i] == NULL) {
/* We removed all of the items in this slab class */
continue;
}
const char *prefix = "items";
add_statistics(c, add_stats, prefix, i, "number", "%u",
engine->items.sizes[i]);
add_statistics(c, add_stats, prefix, i, "age", "%u",
engine->items.tails[i]->time);
add_statistics(c, add_stats, prefix, i, "evicted",
"%u", engine->items.itemstats[i].evicted);
add_statistics(c, add_stats, prefix, i, "evicted_nonzero",
"%u", engine->items.itemstats[i].evicted_nonzero);
add_statistics(c, add_stats, prefix, i, "evicted_time",
"%u", engine->items.itemstats[i].evicted_time);
add_statistics(c, add_stats, prefix, i, "outofmemory",
"%u", engine->items.itemstats[i].outofmemory);
add_statistics(c, add_stats, prefix, i, "tailrepairs",
"%u", engine->items.itemstats[i].tailrepairs);;
add_statistics(c, add_stats, prefix, i, "reclaimed",
"%u", engine->items.itemstats[i].reclaimed);;
}
}
}
static ENGINE_ERROR_CODE do_item_delete(struct demo_engine *engine,
const void* key, const size_t nkey,
uint64_t cas)
{
ENGINE_ERROR_CODE ret;
hash_item *it = do_item_get(engine, key, nkey, true);
if (it == NULL) {
ret = ENGINE_KEY_ENOENT;
} else {
if (cas == 0 || cas == dm_item_get_cas(it)) {
do_item_unlink(engine, it, ITEM_UNLINK_NORMAL);
ret = ENGINE_SUCCESS;
} else {
ret = ENGINE_KEY_EEXISTS;
}
do_item_release(engine, it);
}
return ret;
}
/* expires items that are more recent than the oldest_live setting. */
void do_item_flush_expired(void) {
int i;
item *iter, *next;
if (settings.oldest_live == 0)
return;
for (i = 0; i < LARGEST_ID; i++) {
/* The LRU is sorted in decreasing time order, and an item's timestamp
* is never newer than its last access time, so we only need to walk
* back until we hit an item older than the oldest_live time.
* The oldest_live checking will auto-expire the remaining items.
*/
for (iter = heads[i]; iter != NULL; iter = next) {
if (iter->time >= settings.oldest_live) {
next = iter->next;
if ((iter->it_flags & ITEM_SLABBED) == 0) {
do_item_unlink(iter);
}
} else {
/* We've hit the first old item. Continue to the next queue. */
break;
}
}
}
}
/*@null@*/
item *do_item_alloc(char *key, const size_t nkey, const int flags, const rel_time_t exptime, const int nbytes) {
uint8_t nsuffix;
item *it;
char suffix[40];
size_t ntotal = item_make_header(nkey + 1, flags, nbytes, suffix, &nsuffix);
unsigned int id = slabs_clsid(ntotal);
if (id == 0)
return 0;
it = slabs_alloc(ntotal);
if (it == 0) {
int tries = 50;
item *search;
/* If requested to not push old items out of cache when memory runs out,
* we're out of luck at this point...
*/
if (settings.evict_to_free == 0) return NULL;
/*
* try to get one off the right LRU
* don't necessariuly unlink the tail because it may be locked: refcount>0
* search up from tail an item with refcount==0 and unlink it; give up after 50
* tries
*/
if (id > LARGEST_ID) return NULL;
if (tails[id] == 0) return NULL;
for (search = tails[id]; tries > 0 && search != NULL; tries--, search=search->prev) {
if (search->refcount == 0) {
if (search->exptime == 0 || search->exptime > current_time) {
STATS_LOCK();
stats.evictions++;
STATS_UNLOCK();
}
do_item_unlink(search);
break;
}
}
it = slabs_alloc(ntotal);
if (it == 0) return NULL;
}
assert(it->slabs_clsid == 0);
it->slabs_clsid = id;
assert(it != heads[it->slabs_clsid]);
it->next = it->prev = it->h_next = 0;
it->refcount = 1; /* the caller will have a reference */
DEBUG_REFCNT(it, '*');
it->it_flags = 0;
it->nkey = nkey;
it->nbytes = nbytes;
strcpy(ITEM_key(it), key);
it->exptime = exptime;
memcpy(ITEM_suffix(it), suffix, (size_t)nsuffix);
it->nsuffix = nsuffix;
return it;
}
int do_item_replace(item *it, item *new_it) {
assert((it->it_flags & ITEM_SLABBED) == 0);
do_item_unlink(it);
return do_item_link(new_it);
}
/*@null@*/
item *do_item_alloc(char *key, const size_t nkey, const int flags, const rel_time_t exptime, const int nbytes) {
uint8_t nsuffix;
item *it = NULL;
char suffix[40];
size_t ntotal = item_make_header(nkey + 1, flags, nbytes, suffix, &nsuffix);
if (settings.use_cas) {
ntotal += sizeof(uint64_t);
}
unsigned int id = slabs_clsid(ntotal);
if (id == 0)
return 0;
/* do a quick check if we have any expired items in the tail.. */
int tries = 50;
item *search;
for (search = tails[id];
tries > 0 && search != NULL;
tries--, search=search->prev) {
if (search->refcount == 0 &&
(search->exptime != 0 && search->exptime < current_time)) {
it = search;
/* I don't want to actually free the object, just steal
* the item to avoid to grab the slab mutex twice ;-)
*/
it->refcount = 1;
do_item_unlink(it);
/* Initialize the item block: */
it->slabs_clsid = 0;
it->refcount = 0;
break;
}
}
if (it == NULL && (it = slabs_alloc(ntotal, id)) == NULL) {
/*
** Could not find an expired item at the tail, and memory allocation
** failed. Try to evict some items!
*/
tries = 50;
/* If requested to not push old items out of cache when memory runs out,
* we're out of luck at this point...
*/
if (settings.evict_to_free == 0) {
itemstats[id].outofmemory++;
return NULL;
}
/*
* try to get one off the right LRU
* don't necessariuly unlink the tail because it may be locked: refcount>0
* search up from tail an item with refcount==0 and unlink it; give up after 50
* tries
*/
if (tails[id] == 0) {
itemstats[id].outofmemory++;
return NULL;
}
for (search = tails[id]; tries > 0 && search != NULL; tries--, search=search->prev) {
if (search->refcount == 0) {
if (search->exptime == 0 || search->exptime > current_time) {
itemstats[id].evicted++;
itemstats[id].evicted_time = current_time - search->time;
STATS_LOCK();
stats.evictions++;
STATS_UNLOCK();
}
do_item_unlink(search);
break;
}
}
it = slabs_alloc(ntotal, id);
if (it == 0) {
itemstats[id].outofmemory++;
/* Last ditch effort. There is a very rare bug which causes
* refcount leaks. We've fixed most of them, but it still happens,
* and it may happen in the future.
* We can reasonably assume no item can stay locked for more than
* three hours, so if we find one in the tail which is that old,
* free it anyway.
*/
tries = 50;
for (search = tails[id]; tries > 0 && search != NULL; tries--, search=search->prev) {
if (search->refcount != 0 && search->time + TAIL_REPAIR_TIME < current_time) {
itemstats[id].tailrepairs++;
search->refcount = 0;
do_item_unlink(search);
break;
}
}
it = slabs_alloc(ntotal, id);
if (it == 0) {
return NULL;
}
}
}
assert(it->slabs_clsid == 0);
it->slabs_clsid = id;
assert(it != heads[it->slabs_clsid]);
it->next = it->prev = it->h_next = 0;
it->refcount = 1; /* the caller will have a reference */
DEBUG_REFCNT(it, '*');
it->it_flags = settings.use_cas ? ITEM_CAS : 0;
it->nkey = nkey;
it->nbytes = nbytes;
memcpy(ITEM_key(it), key, nkey);
it->exptime = exptime;
memcpy(ITEM_suffix(it), suffix, (size_t)nsuffix);
it->nsuffix = nsuffix;
return it;
}
/*
* allocate all memory with small and large chunks. link them such that the
* small items are the oldest. allocate one large object that can be covered by
* the release of one large item. this covers part of case 4 for the large item
* alloc in flat_storage_lru_evict(..).
*/
static int
mixed_items_release_one_large_item_test(int verbose) {
typedef struct {
item* it;
char key[KEY_MAX_LENGTH];
uint8_t klen;
} mixed_items_release_one_small_item_t;
size_t num_small_objects = (fsi.large_free_list_sz / 2) * SMALL_CHUNKS_PER_LARGE_CHUNK;
/* this is not the same as fsi.large_free_list_sz / 2 due to rounding. */
size_t num_large_objects = fsi.large_free_list_sz - (fsi.large_free_list_sz / 2);
mixed_items_release_one_small_item_t* large_items = malloc(sizeof(mixed_items_release_one_small_item_t) *
num_large_objects);
mixed_items_release_one_small_item_t* small_items = malloc(sizeof(mixed_items_release_one_small_item_t) *
num_small_objects);
item* lru_trigger;
size_t max_small_key_size = SMALL_TITLE_CHUNK_DATA_SZ;
size_t min_size_for_large_chunk = ( sizeof( ((small_title_chunk_t*) 0)->data ) ) +
( (SMALL_CHUNKS_PER_LARGE_CHUNK - 1) * sizeof( ((small_body_chunk_t*) 0)->data ) ) +
1;
size_t i;
char key[KEY_MAX_LENGTH];
size_t klen;
size_t large_free_list_sz = fsi.large_free_list_sz, small_free_list_sz = fsi.small_free_list_sz;
V_PRINTF(1, " * %s\n", __FUNCTION__);
TASSERT(fsi.large_free_list_sz != 0);
TASSERT(fsi.small_free_list_sz == 0);
for (i = 0; i < num_large_objects; i ++) {
V_PRINTF(2, "\r * allocating large object %lu", i);
V_FLUSH(2);
do {
large_items[i].klen = make_random_key(large_items[i].key, KEY_MAX_LENGTH, true);
} while (assoc_find(large_items[i].key, large_items[i].klen));
large_items[i].it = do_item_alloc(large_items[i].key, large_items[i].klen,
FLAGS, 0,
min_size_for_large_chunk - large_items[i].klen,
addr);
TASSERT(large_items[i].it);
TASSERT(is_item_large_chunk(large_items[i].it));
do_item_link(large_items[i].it, large_items[i].key);
}
V_PRINTF(2, "\n");
for (i = 0; i < num_small_objects; i ++) {
V_PRINTF(2, "\r * allocating small object %lu", i);
V_FLUSH(2);
do {
small_items[i].klen = make_random_key(small_items[i].key, max_small_key_size, true);
} while (assoc_find(small_items[i].key, small_items[i].klen));
small_items[i].it = do_item_alloc(small_items[i].key, small_items[i].klen,
FLAGS, 0,
0, addr);
TASSERT(small_items[i].it);
TASSERT(is_item_large_chunk(small_items[i].it) == 0);
do_item_link(small_items[i].it, small_items[i].key);
}
V_PRINTF(2, "\n");
TASSERT(fsi.large_free_list_sz == 0 &&
fsi.small_free_list_sz == 0);
V_LPRINTF(2, "alloc before deref\n");
do {
klen = make_random_key(key, max_small_key_size, true);
} while (assoc_find(key, klen));
lru_trigger = do_item_alloc(key, klen, FLAGS, 0, 0, addr);
TASSERT(lru_trigger == NULL);
V_LPRINTF(2, "dereferencing objects\n");
for (i = 0; i < num_small_objects; i ++) {
do_item_deref(small_items[i].it);
}
for (i = 0; i < num_large_objects; i ++) {
do_item_deref(large_items[i].it);
}
V_LPRINTF(2, "alloc after deref\n");
lru_trigger = do_item_alloc(key, klen, FLAGS, 0, min_size_for_large_chunk - klen, addr);
TASSERT(lru_trigger != NULL);
V_LPRINTF(2, "search for evicted object\n");
TASSERT(assoc_find(large_items[0].key, large_items[0].klen) == NULL);
V_LPRINTF(2, "ensuring that objects that shouldn't be evicted are still present\n");
for (i = 0; i < num_small_objects; i ++) {
TASSERT(assoc_find(small_items[i].key, small_items[i].klen));
}
for (i = 1; i < num_large_objects; i ++) {
TASSERT(assoc_find(large_items[i].key, large_items[i].klen));
}
V_LPRINTF(2, "cleanup objects\n");
for (i = 0; i < num_small_objects; i ++) {
do_item_unlink(small_items[i].it, UNLINK_NORMAL, small_items[i].key);
}
for (i = 1; i < num_large_objects; i ++) {
do_item_unlink(large_items[i].it, UNLINK_NORMAL, large_items[i].key);
}
do_item_deref(lru_trigger);
TASSERT(fsi.large_free_list_sz == large_free_list_sz &&
fsi.small_free_list_sz == small_free_list_sz);
return 0;
}
/*
* allocate all memory with small and large chunks. link them such the
* allocation of a large object will start evicting small chunks but then stop
* because the large chunk LRU has an older item. this covers part of case 3
* and part of case 4 for the small item alloc in flat_storage_lru_evict(..).
*/
static int
mixed_items_release_small_and_large_items_scan_stop_test(int verbose) {
typedef struct {
item* it;
char key[KEY_MAX_LENGTH];
uint8_t klen;
} mixed_items_release_one_small_item_t;
size_t num_small_objects = (fsi.large_free_list_sz / 2) * SMALL_CHUNKS_PER_LARGE_CHUNK;
/* this is not the same as fsi.large_free_list_sz / 2 due to rounding. */
size_t num_large_objects = fsi.large_free_list_sz - (fsi.large_free_list_sz / 2);
mixed_items_release_one_small_item_t* large_items = malloc(sizeof(mixed_items_release_one_small_item_t) *
num_large_objects);
mixed_items_release_one_small_item_t* small_items = malloc(sizeof(mixed_items_release_one_small_item_t) *
num_small_objects);
item* lru_trigger;
size_t max_small_key_size = SMALL_TITLE_CHUNK_DATA_SZ;
size_t min_size_for_large_chunk = ( sizeof( ((small_title_chunk_t*) 0)->data ) ) +
( (SMALL_CHUNKS_PER_LARGE_CHUNK - 1) * sizeof( ((small_body_chunk_t*) 0)->data ) ) +
1;
size_t i;
char key[KEY_MAX_LENGTH];
size_t klen;
size_t large_free_list_sz = fsi.large_free_list_sz, small_free_list_sz = fsi.small_free_list_sz;
V_PRINTF(1, " * %s\n", __FUNCTION__);
TASSERT(fsi.large_free_list_sz != 0);
TASSERT(fsi.small_free_list_sz == 0);
for (i = 0; i < num_small_objects; i ++) {
V_PRINTF(2, "\r * allocating small object %lu", i);
V_FLUSH(2);
do {
small_items[i].klen = make_random_key(small_items[i].key, max_small_key_size, true);
} while (assoc_find(small_items[i].key, small_items[i].klen));
small_items[i].it = do_item_alloc(small_items[i].key, small_items[i].klen,
FLAGS, 0,
0, addr);
TASSERT(small_items[i].it);
TASSERT(is_item_large_chunk(small_items[i].it) == 0);
do_item_link(small_items[i].it, small_items[i].key);
}
V_PRINTF(2, "\n");
for (i = 0; i < num_large_objects; i ++) {
V_PRINTF(2, "\r * allocating large object %lu", i);
V_FLUSH(2);
do {
large_items[i].klen = make_random_key(large_items[i].key, KEY_MAX_LENGTH, true);
} while (assoc_find(large_items[i].key, large_items[i].klen));
large_items[i].it = do_item_alloc(large_items[i].key, large_items[i].klen,
FLAGS, 0,
min_size_for_large_chunk - large_items[i].klen, addr);
TASSERT(large_items[i].it);
TASSERT(is_item_large_chunk(large_items[i].it));
do_item_link(large_items[i].it, large_items[i].key);
}
V_PRINTF(2, "\n");
TASSERT(fsi.large_free_list_sz == 0 &&
fsi.small_free_list_sz == 0);
V_LPRINTF(2, "update items\n");
/* update the objects we want to clobber *first*. but since ties go to the
* large item, we need to bump the time stamp to ensure the small item is
* released first. */
current_time += ITEM_UPDATE_INTERVAL + 1; /* initial bump to ensure that
* LRU reordering takes place. */
do_item_update(small_items[0].it);
current_time += 1;
do_item_update(large_items[0].it);
/* bump the timestamp and add the remaining items. */
current_time += 1;
for (i = 1; i < num_small_objects; i ++) {
do_item_update(small_items[i].it);
}
for (i = 1; i < num_large_objects; i ++) {
do_item_update(large_items[i].it);
}
V_LPRINTF(2, "dereferencing objects\n");
for (i = 0; i < num_small_objects; i ++) {
do_item_deref(small_items[i].it);
}
for (i = 0; i < num_large_objects; i ++) {
do_item_deref(large_items[i].it);
}
V_LPRINTF(2, "alloc after deref\n");
do {
klen = make_random_key(key, max_small_key_size, true);
} while (assoc_find(key, klen));
lru_trigger = do_item_alloc(key, klen, FLAGS, 0, LARGE_TITLE_CHUNK_DATA_SZ - klen, addr);
TASSERT(lru_trigger != NULL);
TASSERT(is_item_large_chunk(lru_trigger));
V_LPRINTF(2, "search for evicted objects\n");
TASSERT(assoc_find(small_items[0].key, small_items[0].klen) == NULL);
TASSERT(assoc_find(large_items[0].key, large_items[0].klen) == NULL);
V_LPRINTF(2, "ensuring that objects that shouldn't be evicted are still present\n");
for (i = 1; i < num_small_objects; i ++) {
TASSERT(assoc_find(small_items[i].key, small_items[i].klen));
}
for (i = 1; i < num_large_objects; i ++) {
TASSERT(assoc_find(large_items[i].key, large_items[i].klen));
}
V_LPRINTF(2, "cleanup objects\n");
for (i = 1; i < num_small_objects; i ++) {
do_item_unlink(small_items[i].it, UNLINK_NORMAL, small_items[i].key);
}
for (i = 1; i < num_large_objects; i ++) {
do_item_unlink(large_items[i].it, UNLINK_NORMAL, large_items[i].key);
}
do_item_deref(lru_trigger);
TASSERT(fsi.large_free_list_sz == large_free_list_sz &&
fsi.small_free_list_sz == small_free_list_sz);
return 0;
}