/** 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);
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;
}
/* Note: this isn't an assoc_update. The key must not already exist to call this */
int assoc_insert(struct default_engine *engine, uint32_t hash, hash_item *it)
{
struct assoc *assoc = &engine->assoc;
uint32_t bucket = GET_HASH_BUCKET(hash, assoc->hashmask);
uint32_t tabidx;
assert(assoc_find(engine, hash, item_get_key(it), it->nkey) == 0); /* shouldn't have duplicately named things defined */
if (assoc->infotable[bucket].curpower != assoc->rootpower &&
assoc->infotable[bucket].refcount == 0) {
redistribute(engine, bucket);
}
tabidx = GET_HASH_TABIDX(hash, assoc->hashpower,
hashmask(assoc->infotable[bucket].curpower));
// inserting actual hash_item to appropriate assoc_t
it->h_next = assoc->roottable[tabidx].hashtable[bucket];
assoc->roottable[tabidx].hashtable[bucket] = it;
assoc->hash_items++;
if (assoc->hash_items > (hashsize(assoc->hashpower + assoc->rootpower) * 3) / 2) {
assoc_expand(engine);
}
MEMCACHED_ASSOC_INSERT(item_get_key(it), it->nkey, assoc->hash_items);
return 1;
}
/* Note: this isn't an assoc_update. The key must not already exist to call this */
int assoc_insert(struct default_engine *engine, uint32_t hash, hash_item *it)
{
unsigned int oldbucket;
assert(assoc_find(engine, hash, item_get_key(it), it->nkey) == 0); /* shouldn't have duplicately named things defined */
// inserting actual hash_item to appropriate assoc_t
if (engine->assoc.expanding &&
(oldbucket = (hash & hashmask(engine->assoc.hashpower - 1))) >= engine->assoc.expand_bucket)
{
it->h_next = engine->assoc.old_hashtable[oldbucket];
engine->assoc.old_hashtable[oldbucket] = it;
} else {
it->h_next = engine->assoc.primary_hashtable[hash & hashmask(engine->assoc.hashpower)];
engine->assoc.primary_hashtable[hash & hashmask(engine->assoc.hashpower)] = it;
}
engine->assoc.hash_items++;
if (! engine->assoc.expanding && engine->assoc.hash_items > (hashsize(engine->assoc.hashpower) * 3) / 2) {
assoc_expand(engine);
}
MEMCACHED_ASSOC_INSERT(item_get_key(it), it->nkey, engine->assoc.hash_items);
return 1;
}
/** 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;
}
void slabs_alloc_test(void) {
unsigned int total_chunk = 0;
const char *key = "charliezhao";
size_t nkey = strlen(key) + 1;
item *ptr = (item *)slabs_alloc(1024, slabs_clsid(1024), &total_chunk);
strcpy(ITEM_key(ptr), key);
ptr->nkey = nkey;
strcpy(ITEM_data(ptr), "xuechaozhao");
uint32_t hv = jenkins_hash(key, strlen(key));
assoc_insert(ptr, hv);
for(int i = 0; i <= 10922; ++i) {
void *ptr = slabs_alloc(96, slabs_clsid(96), &total_chunk);
if(ptr == NULL) {
fprintf(stderr, "i: %7d slabs_alloc fail\n",
i);
break;
}
else {
slabs_free(ptr, 96, slabs_clsid(96));
}
}
item *ptr2 = assoc_find(key, nkey, hv);
fprintf(stdout, "key:%20s value:%20s\n", ITEM_key(ptr2), ITEM_data(ptr2));
}
/* Note: this isn't an assoc_update. The key must not already exist to call this */
int assoc_insert(item *it) {
uint32_t hv;
unsigned int oldbucket;
assert(assoc_find(ITEM_key(it), it->nkey) == 0); /* shouldn't have duplicately named things defined */
hv = hash(ITEM_key(it), it->nkey, 0);
if (expanding &&
(oldbucket = (hv & hashmask(hashpower - 1))) >= expand_bucket)
{
it->h_next = old_hashtable[oldbucket];
old_hashtable[oldbucket] = it;
} else {
it->h_next = primary_hashtable[hv & hashmask(hashpower)];
primary_hashtable[hv & hashmask(hashpower)] = it;
}
hash_items++;
if (! expanding && hash_items > (hashsize(hashpower) * 3) / 2) {
assoc_expand();
}
MEMCACHED_ASSOC_INSERT(ITEM_key(it), it->nkey, hash_items);
return 1;
}
/** 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;
}
/** returns an item whether or not it's delete-locked or expired. */
item *do_item_get_nocheck(const char *key, const size_t nkey) {
item *it = assoc_find(key, nkey);
if (it) {
it->refcount++;
DEBUG_REFCNT(it, '+');
}
return it;
}
/*
* Returns an item whether or not it's been marked as expired or deleted.
*/
item *mt_item_get_nocheck(char *key, size_t nkey) {
item *it;
pthread_mutex_lock(&cache_lock);
it = assoc_find(key, nkey);
it->refcount++;
pthread_mutex_unlock(&cache_lock);
return it;
}
void complete_nread(conn *c) {
item *it = c->item;
int comm = c->item_comm;
item *old_it;
time_t now = time(0);
stats.set_cmds++;
while(1) {
if (strncmp(ITEM_data(it) + it->nbytes - 2, "\r\n", 2) != 0) {
out_string(c, "CLIENT_ERROR bad data chunk");
break;
}
old_it = assoc_find(ITEM_key(it));
if (old_it && settings.oldest_live &&
old_it->time <= settings.oldest_live) {
item_unlink(old_it);
old_it = 0;
}
if (old_it && old_it->exptime && old_it->exptime < now) {
item_unlink(old_it);
old_it = 0;
}
if (old_it && comm==NREAD_ADD) {
item_update(old_it);
out_string(c, "NOT_STORED");
break;
}
if (!old_it && comm == NREAD_REPLACE) {
out_string(c, "NOT_STORED");
break;
}
if (old_it && (old_it->it_flags & ITEM_DELETED) && (comm == NREAD_REPLACE || comm == NREAD_ADD)) {
out_string(c, "NOT_STORED");
break;
}
if (old_it) {
item_replace(old_it, it);
} else item_link(it);
c->item = 0;
out_string(c, "STORED");
return;
}
item_free(it);
c->item = 0;
return;
}
static rstatus_t
stats_pool_copy_recover(struct context *ctx, struct stats_pool *stp_src, struct hash_table **sit)
{
uint32_t i, j, k;
struct stats *st = ctx->stats;
struct array *sum = &st->sum;
struct stats_pool *stp_dst;
struct stats_metric *stm_src, *stm_dst;
struct stats_server *sts_src, *sts_dst;
char *key;
uint64_t sidx;
for (i = 0;i < array_n(sum); i++) {
/* get pool from array sum */
stp_dst = array_get(sum, i);
/* find the pool */
if (string_compare(&stp_src->name, &stp_dst->name) == 0) {
/* recover the metric array */
for (j = 0;j < array_n(&stp_src->metric);j++) {
stm_src = array_get(&stp_src->metric, j);
stm_dst = array_get(&stp_dst->metric, j);
stats_metric_copy(stm_dst, stm_src);
}
/* recover the server array */
for (j = 0;j < array_n(&stp_dst->server);j++) {
/* recover sts_src data to sts_dst */
sts_dst = array_get(&stp_dst->server, j);
log_debug(LOG_VVVERB, "try recover server %s", sts_dst->name.data);
/* find the server in index hashtable */
key = (char *)sts_dst->name.data;
sidx = (uint64_t)assoc_find(*sit, key, strlen(key));
if (sidx > 0) {
sidx = sidx - 1;
/* get server data in array */
sts_src = array_get(&stp_src->server, (uint32_t)sidx);
ASSERT(string_compare(&sts_dst->name, &sts_src->name) == 0);
log_debug(LOG_VERB, "recovering server stats %s", sts_dst->name.data);
/* recover the old data */
for (k = 0;k < array_n(&sts_src->metric) - 4; k++) {
stm_src = array_get(&sts_src->metric, k);
stm_dst = array_get(&sts_dst->metric, k);
stats_metric_copy(stm_dst, stm_src);
}
}
}
}
}
return NC_OK;
}
static struct item* _item_get(const char *key, uint16_t nkey) {
struct item *it;
it = assoc_find(key, nkey);
if (it == NULL) return NULL;
if (it->exptime != 0 && it->exptime <= time_now()) {
_item_unlink(it);
return NULL;
}
item_acquire_refcount(it);
_item_touch(it);
return it;
}
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;
}
/** 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;
}
/** 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;
}
/* Note: this isn't an assoc_update. The key must not already exist to call this */
int assoc_insert(struct persistent_engine *engine, uint32_t hash, hash_item *it) {
unsigned int oldbucket;
assert(assoc_find(engine, hash, item_get_key(&it->item), it->item.nkey) == 0); /* shouldn't have duplicately named things defined */
if (engine->assoc.expanding &&
(oldbucket = (hash & hashmask(engine->assoc.hashpower - 1))) >= engine->assoc.expand_bucket)
{
it->h_next = engine->assoc.old_hashtable[oldbucket];
engine->assoc.old_hashtable[oldbucket] = it;
} else {
it->h_next = engine->assoc.primary_hashtable[hash & hashmask(engine->assoc.hashpower)];
engine->assoc.primary_hashtable[hash & hashmask(engine->assoc.hashpower)] = it;
}
engine->assoc.hash_items++;
if (! engine->assoc.expanding && engine->assoc.hash_items > (hashsize(engine->assoc.hashpower) * 3) / 2) {
assoc_expand(engine);
}
return 1;
}
int replication_cmd(conn *c, Q_ITEM *q)
{
item *it;
switch (q->type) {
case REPLICATION_REP:
if(it = assoc_find(q->key, strlen(q->key)))
return(replication_rep(c, it));
else
return(replication_del(c, q->key));
case REPLICATION_DEL:
return(replication_del(c, q->key));
case REPLICATION_DEFER_DEL:
return(replication_defer_del(c, q->key, q->time));
case REPLICATION_FLUSH_ALL:
return(replication_flush_all(c, 0));
case REPLICATION_DEFER_FLUSH_ALL:
return(replication_flush_all(c, q->time));
case REPLICATION_MARUGOTO_END:
return(replication_marugoto_end(c));
default:
fprintf(stderr,"replication: got unknown command:%d\n", q->type);
return(0);
}
}
/** 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);
// 查找 key 是否存在
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;
}
}
//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) {
//settings.oldest_live初始化值为0
//检测用户是否使用过flush_all命令，删除所有item
//it->time <= settings.oldest_live就说明用户在使用flush_all命令的时候
//就已经存在该item了。那么该item是要删除的
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");
}
}
//该item已经过期失效了
else if (it->exptime != 0 && it->exptime <= current_time) {
do_item_unlink(it, hv);
do_item_remove(it);
it = NULL;
if (was_found) {
fprintf(stderr, " -nuked by expire");
}
} else {
it->it_flags |= ITEM_FETCHED;
DEBUG_REFCNT(it, '+');
}
}
if (settings.verbose > 2)
fprintf(stderr, "\n");
return it;
}
void process_command(conn *c, char *command) {
int comm = 0;
int incr = 0;
/*
* for commands set/add/replace, we build an item and read the data
* directly into it, then continue in nread_complete().
*/
if (settings.verbose > 1)
fprintf(stderr, "<%d %s\n", c->sfd, command);
/* All incoming commands will require a response, so we cork at the beginning,
and uncork at the very end (usually by means of out_string) */
set_cork(c, 1);
if ((strncmp(command, "add ", 4) == 0 && (comm = NREAD_ADD)) ||
(strncmp(command, "set ", 4) == 0 && (comm = NREAD_SET)) ||
(strncmp(command, "replace ", 8) == 0 && (comm = NREAD_REPLACE))) {
char key[251];
int flags;
time_t expire;
int len, res;
item *it;
res = sscanf(command, "%*s %250s %u %ld %d\n", key, &flags, &expire, &len);
if (res!=4 || strlen(key)==0 ) {
out_string(c, "CLIENT_ERROR bad command line format");
return;
}
expire = realtime(expire);
it = item_alloc(key, flags, expire, len+2);
if (it == 0) {
out_string(c, "SERVER_ERROR out of memory");
/* swallow the data line */
c->write_and_go = conn_swallow;
c->sbytes = len+2;
return;
}
c->item_comm = comm;
c->item = it;
c->rcurr = ITEM_data(it);
c->rlbytes = it->nbytes;
c->state = conn_nread;
return;
}
if ((strncmp(command, "incr ", 5) == 0 && (incr = 1)) ||
(strncmp(command, "decr ", 5) == 0)) {
char temp[32];
unsigned int value;
item *it;
unsigned int delta;
char key[251];
int res;
char *ptr;
time_t now = time(0);
res = sscanf(command, "%*s %250s %u\n", key, &delta);
if (res!=2 || strlen(key)==0 ) {
out_string(c, "CLIENT_ERROR bad command line format");
return;
}
it = assoc_find(key);
if (it && (it->it_flags & ITEM_DELETED)) {
it = 0;
}
if (it && it->exptime && it->exptime < now) {
item_unlink(it);
it = 0;
}
if (!it) {
out_string(c, "NOT_FOUND");
return;
}
ptr = ITEM_data(it);
while (*ptr && (*ptr<'0' && *ptr>'9')) ptr++;
value = atoi(ptr);
if (incr)
value+=delta;
else {
if (delta >= value) value = 0;
else value-=delta;
}
sprintf(temp, "%u", value);
res = strlen(temp);
if (res + 2 > it->nbytes) { /* need to realloc */
item *new_it;
new_it = item_alloc(ITEM_key(it), it->flags, it->exptime, res + 2 );
if (new_it == 0) {
out_string(c, "SERVER_ERROR out of memory");
return;
}
memcpy(ITEM_data(new_it), temp, res);
memcpy(ITEM_data(new_it) + res, "\r\n", 2);
item_replace(it, new_it);
} else { /* replace in-place */
memcpy(ITEM_data(it), temp, res);
memset(ITEM_data(it) + res, ' ', it->nbytes-res-2);
}
out_string(c, temp);
return;
}
if (strncmp(command, "get ", 4) == 0) {
char *start = command + 4;
char key[251];
int next;
int i = 0;
item *it;
time_t now = time(0);
while(sscanf(start, " %250s%n", key, &next) >= 1) {
start+=next;
stats.get_cmds++;
it = assoc_find(key);
if (it && (it->it_flags & ITEM_DELETED)) {
it = 0;
}
if (settings.oldest_live && it &&
it->time <= settings.oldest_live) {
item_unlink(it);
it = 0;
}
if (it && it->exptime && it->exptime < now) {
item_unlink(it);
it = 0;
}
if (it) {
if (i >= c->isize) {
item **new_list = realloc(c->ilist, sizeof(item *)*c->isize*2);
if (new_list) {
c->isize *= 2;
c->ilist = new_list;
} else break;
}
stats.get_hits++;
it->refcount++;
item_update(it);
*(c->ilist + i) = it;
i++;
} else stats.get_misses++;
}
c->icurr = c->ilist;
c->ileft = i;
if (c->ileft) {
c->ipart = 0;
c->state = conn_mwrite;
c->ibytes = 0;
return;
} else {
out_string(c, "END");
return;
}
}
if (strncmp(command, "delete ", 7) == 0) {
char key[251];
item *it;
int res;
time_t exptime = 0;
res = sscanf(command, "%*s %250s %ld", key, &exptime);
it = assoc_find(key);
if (!it) {
out_string(c, "NOT_FOUND");
return;
}
if (exptime == 0) {
item_unlink(it);
out_string(c, "DELETED");
return;
}
if (delcurr >= deltotal) {
item **new_delete = realloc(todelete, sizeof(item *) * deltotal * 2);
if (new_delete) {
todelete = new_delete;
deltotal *= 2;
} else {
/*
* can't delete it immediately, user wants a delay,
* but we ran out of memory for the delete queue
*/
out_string(c, "SERVER_ERROR out of memory");
return;
}
}
exptime = realtime(exptime);
it->refcount++;
/* use its expiration time as its deletion time now */
it->exptime = exptime;
it->it_flags |= ITEM_DELETED;
todelete[delcurr++] = it;
out_string(c, "DELETED");
return;
}
if (strncmp(command, "stats", 5) == 0) {
process_stat(c, command);
return;
}
if (strcmp(command, "flush_all") == 0) {
settings.oldest_live = time(0);
out_string(c, "OK");
return;
}
if (strcmp(command, "version") == 0) {
out_string(c, "VERSION " VERSION);
return;
}
if (strcmp(command, "quit") == 0) {
c->state = conn_closing;
return;
}
if (strncmp(command, "slabs reassign ", 15) == 0) {
int src, dst;
char *start = command+15;
if (sscanf(start, "%u %u\r\n", &src, &dst) == 2) {
int rv = slabs_reassign(src, dst);
if (rv == 1) {
out_string(c, "DONE");
return;
}
if (rv == 0) {
out_string(c, "CANT");
return;
}
if (rv == -1) {
out_string(c, "BUSY");
return;
}
}
out_string(c, "CLIENT_ERROR bogus command");
return;
}
out_string(c, "ERROR");
return;
}
/* allocate all memory with small chunks. allocate one more object. it should
* free up the oldest object. release all objects. this covers case 1 for the
* small item alloc in flat_storage_lru_evict(..). */
static int
all_small_chunks_test(int verbose) {
typedef struct {
item* it;
char key[KEY_MAX_LENGTH];
uint8_t klen;
} all_small_chunks_key_t;
size_t num_objects = fsi.large_free_list_sz * SMALL_CHUNKS_PER_LARGE_CHUNK;
all_small_chunks_key_t* small_items = malloc(sizeof(all_small_chunks_key_t) * num_objects);
item* lru_trigger;
size_t max_key_size = SMALL_TITLE_CHUNK_DATA_SZ;
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_objects; i ++) {
V_PRINTF(2, "\r * allocating object %lu", i);
V_FLUSH(2);
do {
small_items[i].klen = make_random_key(small_items[i].key, max_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) == false);
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_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_objects; i ++) {
do_item_deref(small_items[i].it);
}
V_LPRINTF(2, "alloc after deref\n");
lru_trigger = do_item_alloc(key, klen, FLAGS, 0, 0, addr);
TASSERT(lru_trigger != NULL);
V_LPRINTF(2, "search for evicted object\n");
TASSERT(assoc_find(small_items[0].key, small_items[0].klen) == NULL);
V_LPRINTF(2, "ensuring that objects that shouldn't be evicted are still present\n");
for (i = 1; i < num_objects; i ++) {
TASSERT(assoc_find(small_items[i].key, small_items[i].klen));
}
V_LPRINTF(2, "cleanup objects\n");
for (i = 1; i < num_objects; i ++) {
do_item_unlink(small_items[i].it, UNLINK_NORMAL, small_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 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;
}
/*
* allocate nearly all memory with small items (all memory -
* SMALL_CHUNKS_PER_LARGE_CHUNK - 1). then set it up such that there is only
* one item eligible to be freed (i.e., by removing the remaining items from the
* LRU. allocate one large object. this will require the migration of one
* single chunk item at the LRU head. this covers part of case 1 for the small
* item alloc in flat_storage_lru_evict(..).
*/
static int
all_small_items_migrate_small_single_chunk_item_at_lru_head_test(int verbose) {
typedef struct {
item* it;
char key[KEY_MAX_LENGTH];
uint8_t klen;
} test_keys_t;
size_t num_objects = fsi.large_free_list_sz * SMALL_CHUNKS_PER_LARGE_CHUNK;
test_keys_t* items = malloc(sizeof(test_keys_t) * num_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, count;
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, count = 0;
fsi.large_free_list_sz ||
fsi.small_free_list_sz > SMALL_CHUNKS_PER_LARGE_CHUNK - 1;
i ++, count ++) {
V_PRINTF(2, "\r * allocating small object %lu", i);
V_FLUSH(2);
assert(i < num_objects);
do {
items[i].klen = make_random_key(items[i].key, max_small_key_size, true);
} while (assoc_find(items[i].key, items[i].klen));
items[i].it = do_item_alloc(items[i].key, items[i].klen,
FLAGS, 0, 0, addr);
TASSERT(items[i].it);
TASSERT(is_item_large_chunk(items[i].it) == 0);
do_item_link(items[i].it, items[i].key);
}
V_PRINTF(2, "\n");
TASSERT(fsi.large_free_list_sz == 0);
TASSERT(fsi.small_free_list_sz == SMALL_CHUNKS_PER_LARGE_CHUNK - 1);
// remove all but one item from the LRU. and release our reference to the
// item we don't remove from the LRU.
for (i = 0; i < count - 1; i ++) {
do_item_unlink(items[i].it, UNLINK_NORMAL, items[i].key);
}
do_item_deref(items[count - 1].it);
TASSERT(fsi.lru_head == items[count - 1].it);
TASSERT(fsi.large_free_list_sz == 0);
TASSERT(fsi.small_free_list_sz == SMALL_CHUNKS_PER_LARGE_CHUNK - 1);
V_LPRINTF(2, "alloc\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, min_size_for_large_chunk - klen, addr);
TASSERT(lru_trigger != NULL);
V_LPRINTF(2, "search for evicted object\n");
TASSERT(assoc_find(items[count - 1].key, items[count - 1].klen) == NULL);
V_LPRINTF(2, "cleanup objects\n");
for (i = 0; i < count - 1; i ++) {
do_item_deref(items[i].it);
}
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;
}
/*
* this is a negative test to ensure the proper behavior when we don't have
* sufficient resources. in this case, we have sufficient small items on the
* LRU, and enough of them have refcount == 0, but all the parent broken chunks
* have refcount > 0.
*/
static int
insufficient_available_large_broken_chunks(int verbose) {
typedef struct {
item* it;
char key[KEY_MAX_LENGTH];
uint8_t klen;
} all_small_chunks_key_t;
size_t num_objects = fsi.large_free_list_sz * SMALL_CHUNKS_PER_LARGE_CHUNK;
all_small_chunks_key_t* small_items = malloc(sizeof(all_small_chunks_key_t) * num_objects);
item* lru_trigger;
size_t max_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_objects; i ++) {
V_PRINTF(2, "\r * allocating object %lu", i);
V_FLUSH(2);
do {
small_items[i].klen = make_random_key(small_items[i].key, max_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) == false);
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_key_size, true);
} while (assoc_find(key, klen));
lru_trigger = do_item_alloc(key, klen, FLAGS, 0, min_size_for_large_chunk - klen, addr);
TASSERT(lru_trigger == NULL);
V_LPRINTF(2, "dereferencing objects\n");
for (i = 0; i < num_objects; i += 2) {
do_item_deref(small_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, "ensuring that objects that shouldn't be evicted are still present\n");
for (i = 0; i < num_objects; i ++) {
bool should_be_found;
/* we free everything we encounter that has no refcount until we hit the
* LRU_SEARCH_DEPTH, at which time we cease searching. */
if (i % 2 == 0 && i < (LRU_SEARCH_DEPTH * 2)) {
should_be_found = false;
} else {
should_be_found = true;
}
TASSERT((assoc_find(small_items[i].key, small_items[i].klen) ? (true) : (false)) ==
should_be_found);
}
V_LPRINTF(2, "cleanup objects\n");
for (i = 0; i < num_objects; i ++) {
/* we dereference all the odd numbered items */
if ((i % 2) != 0) {
do_item_deref(small_items[i].it);
}
/* we unlink everything that's still in the LRU. */
if (i % 2 == 0 && i < (LRU_SEARCH_DEPTH * 2)) {
;
} else {
do_item_unlink(small_items[i].it, UNLINK_NORMAL, small_items[i].key);
}
}
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 items. allocate one large object that can be
* covered by the release of small items, but also requires the migration of
* single chunk items. this covers part of case 1 for the large item alloc in
* flat_storage_lru_evict(..).
*/
static int
all_small_items_migrate_small_single_chunk_items_test(int verbose) {
typedef struct {
item* it;
char key[KEY_MAX_LENGTH];
uint8_t klen;
} test_keys_t;
size_t num_objects = fsi.large_free_list_sz * SMALL_CHUNKS_PER_LARGE_CHUNK;
test_keys_t* items = malloc(sizeof(test_keys_t) * num_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_objects; i ++) {
V_PRINTF(2, "\r * allocating small object %lu", i);
V_FLUSH(2);
do {
items[i].klen = make_random_key(items[i].key, max_small_key_size, true);
} while (assoc_find(items[i].key, items[i].klen));
items[i].it = do_item_alloc(items[i].key, items[i].klen,
FLAGS, 0, 0, addr);
TASSERT(items[i].it);
TASSERT(is_item_large_chunk(items[i].it) == 0);
do_item_link(items[i].it, items[i].key);
}
V_PRINTF(2, "\n");
TASSERT(fsi.large_free_list_sz == 0 &&
fsi.small_free_list_sz == 0);
/* access items we don't want to move. */
current_time += ITEM_UPDATE_INTERVAL + 1;
/* touch every other item. the ones that are not touched in (0,
* SMALL_CHUNKS_PER_LARGE_CHUNK * 2) will be evicted. */
for (i = 0; i < SMALL_CHUNKS_PER_LARGE_CHUNK * 2; i += 2) {
do_item_update(items[i].it);
}
/* touch remaining items */
for (i = SMALL_CHUNKS_PER_LARGE_CHUNK * 2; i < num_objects; i ++) {
do_item_update(items[i].it);
}
V_LPRINTF(2, "dereferencing objects\n");
for (i = 0; i < num_objects; i ++) {
do_item_deref(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, min_size_for_large_chunk - klen, addr);
TASSERT(lru_trigger != NULL);
V_LPRINTF(2, "search for evicted object\n");
for (i = 1; i < SMALL_CHUNKS_PER_LARGE_CHUNK * 2; i += 2) {
TASSERT(assoc_find(items[i].key, items[i].klen) == NULL);
}
V_LPRINTF(2, "ensuring that objects that shouldn't be evicted are still present\n");
for (i = 0; i < SMALL_CHUNKS_PER_LARGE_CHUNK * 2; i += 2) {
/* these may have been moved. */
TASSERT((items[i].it = assoc_find(items[i].key, items[i].klen)));
}
for (i = SMALL_CHUNKS_PER_LARGE_CHUNK * 2; i < num_objects; i ++) {
TASSERT(assoc_find(items[i].key, items[i].klen));
}
V_LPRINTF(2, "cleanup objects\n");
for (i = 0; i < SMALL_CHUNKS_PER_LARGE_CHUNK * 2; i += 2) {
do_item_unlink(items[i].it, UNLINK_NORMAL, items[i].key);
}
for (i = SMALL_CHUNKS_PER_LARGE_CHUNK * 2; i < num_objects; i ++) {
do_item_unlink(items[i].it, UNLINK_NORMAL, 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;
}
/** 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;
}
}
//mutex_unlock(&cache_lock);
int was_found = 0;
if (settings.verbose > 2) {
int ii;
if (it == NULL) {
fprintf(stderr, "> NOT FOUND ");
} else {
fprintf(stderr, "> FOUND KEY ");
was_found++;
}
for (ii = 0; ii < nkey; ++ii) {
fprintf(stderr, "%c", key[ii]);
}
}
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) {
do_item_unlink(it, hv);
do_item_remove(it);
it = NULL;
if (was_found) {
fprintf(stderr, " -nuked by expire");
}
} else {
it->it_flags |= ITEM_FETCHED;
DEBUG_REFCNT(it, '+');
if (settings.anti_stampede > 0 && !(it->it_flags & ITEM_FAKE_MISSED)
&& it->exptime != 0
&& it->exptime < current_time + settings.anti_stampede
&& it->exptime <= current_time + rand() % settings.anti_stampede) {
it->it_flags |= ITEM_FAKE_MISSED;
if (settings.verbose > 2)
fprintf(stderr," - FAKE MISS (stampede protection)");
refcount_decr(&it->refcount);
it = NULL;
}
}
}
if (settings.verbose > 2)
fprintf(stderr, "\n");
return it;
}
void process_command(conn *c, char *command) {
int comm = 0;
/*
* for commands set/add/replace, we build an item and read the data
* directly into it, then continue in nread_complete().
*/
if ((strncmp(command, "add ", 4) == 0 && (comm = NREAD_ADD)) ||
(strncmp(command, "set ", 4) == 0 && (comm = NREAD_SET)) ||
(strncmp(command, "replace ", 8) == 0 && (comm = NREAD_REPLACE))) {
char s_comm[10];
char key[256];
int flags;
time_t expire;
int len, res;
item *it;
res = sscanf(command, "%s %s %u %u %d\n", s_comm, key, &flags, &expire, &len);
if (res!=5 || strlen(key)==0 ) {
out_string(c, "CLIENT_ERROR bad command line format");
return;
}
it = item_alloc(key, flags, expire, len+2);
if (it == 0) {
out_string(c, "SERVER_ERROR out of memory");
c->write_and_close = 1;
return;
}
c->item_comm = comm;
c->item = it;
c->rcurr = it->data;
c->rlbytes = it->nbytes;
c->state = conn_nread;
return;
}
if (strncmp(command, "get ", 4) == 0) {
char *start = command + 4;
char key[256];
int next;
int i = 0;
item *it;
time_t now = time(0);
while(sscanf(start, " %s%n", key, &next) >= 1) {
start+=next;
stats.get_cmds++;
it = (item *)assoc_find(key);
if (it && (it->it_flags & ITEM_DELETED)) {
it = 0;
}
if (it && it->exptime && it->exptime < now) {
item_unlink(it);
it = 0;
}
if (it) {
stats.get_hits++;
it->usecount++;
item_update(it);
*(c->ilist + i) = it;
i++;
if (i > c->isize) {
c->isize *= 2;
c->ilist = realloc(c->ilist, sizeof(item *)*c->isize);
}
} else stats.get_misses++;
}
c->icurr = c->ilist;
c->ileft = i;
if (c->ileft) {
c->ipart = 0;
c->state = conn_mwrite;
c->ibytes = 0;
return;
} else {
out_string(c, "END");
return;
}
}
if (strncmp(command, "delete ", 7) == 0) {
char key [256];
char *start = command+7;
item *it;
sscanf(start, " %s", key);
it = assoc_find(key);
if (!it) {
out_string(c, "NOT_FOUND");
return;
} else {
it->usecount++;
/* use its expiration time as its deletion time now */
it->exptime = time(0) + 4;
it->it_flags |= ITEM_DELETED;
todelete[delcurr++] = it;
if (delcurr >= deltotal) {
deltotal *= 2;
todelete = realloc(todelete, sizeof(item *)*deltotal);
}
}
out_string(c, "DELETED");
return;
}
if (strncmp(command, "stats", 5) == 0) {
process_stat(c, command);
return;
}
if (strcmp(command, "version") == 0) {
out_string(c, "VERSION 2.0");
return;
}
out_string(c, "ERROR");
return;
}
static void storage_compact_readback(void *storage, logger *l,
bool drop_unread, char *readback_buf,
uint32_t page_id, uint64_t page_version, uint64_t read_size) {
uint64_t offset = 0;
unsigned int rescues = 0;
unsigned int lost = 0;
unsigned int skipped = 0;
while (offset < read_size) {
item *hdr_it = NULL;
item_hdr *hdr = NULL;
item *it = (item *)(readback_buf+offset);
unsigned int ntotal;
// probably zeroed out junk at the end of the wbuf
if (it->nkey == 0) {
break;
}
ntotal = ITEM_ntotal(it);
uint32_t hv = (uint32_t)it->time;
item_lock(hv);
// We don't have a conn and don't need to do most of do_item_get
hdr_it = assoc_find(ITEM_key(it), it->nkey, hv);
if (hdr_it != NULL) {
bool do_write = false;
refcount_incr(hdr_it);
// Check validity but don't bother removing it.
if ((hdr_it->it_flags & ITEM_HDR) && !item_is_flushed(hdr_it) &&
(hdr_it->exptime == 0 || hdr_it->exptime > current_time)) {
hdr = (item_hdr *)ITEM_data(hdr_it);
if (hdr->page_id == page_id && hdr->page_version == page_version) {
// Item header is still completely valid.
extstore_delete(storage, page_id, page_version, 1, ntotal);
// drop inactive items.
if (drop_unread && GET_LRU(hdr_it->slabs_clsid) == COLD_LRU) {
do_write = false;
skipped++;
} else {
do_write = true;
}
}
}
if (do_write) {
bool do_update = false;
int tries;
obj_io io;
io.len = ntotal;
io.mode = OBJ_IO_WRITE;
for (tries = 10; tries > 0; tries--) {
if (extstore_write_request(storage, PAGE_BUCKET_COMPACT, PAGE_BUCKET_COMPACT, &io) == 0) {
memcpy(io.buf, it, io.len);
extstore_write(storage, &io);
do_update = true;
break;
} else {
usleep(1000);
}
}
if (do_update) {
if (it->refcount == 2) {
hdr->page_version = io.page_version;
hdr->page_id = io.page_id;
hdr->offset = io.offset;
rescues++;
} else {
lost++;
// TODO: re-alloc and replace header.
}
} else {
lost++;
}
}
do_item_remove(hdr_it);
}
item_unlock(hv);
offset += ntotal;
if (read_size - offset < sizeof(struct _stritem))
break;
}
STATS_LOCK();
stats.extstore_compact_lost += lost;
stats.extstore_compact_rescues += rescues;
stats.extstore_compact_skipped += skipped;
STATS_UNLOCK();
LOGGER_LOG(l, LOG_SYSEVENTS, LOGGER_COMPACT_READ_END,
NULL, page_id, offset, rescues, lost, skipped);
}
/** 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, conn *c) {
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, slabs_lock. */
/* This was made unsafe by removal of the cache_lock:
* slab_rebalance_signal and slab_rebal.* are modified in a separate
* thread under slabs_lock. If slab_rebalance_signal = 1, slab_start =
* NULL (0), but slab_end is still equal to some value, this would end
* up unlinking every item fetched.
* This is either an acceptable loss, or if slab_rebalance_signal is
* true, slab_start/slab_end should be put behind the slabs_lock.
* Which would cause a huge potential slowdown.
* Could also use a specific lock for slab_rebal.* and
* slab_rebalance_signal (shorter lock?)
*/
/*if (slab_rebalance_signal &&
((void *)it >= slab_rebal.slab_start && (void *)it < slab_rebal.slab_end)) {
do_item_unlink(it, hv);
do_item_remove(it);
it = NULL;
}*/
}
int was_found = 0;
if (settings.verbose > 2) {
int ii;
if (it == NULL) {
fprintf(stderr, "> NOT FOUND ");
} else {
fprintf(stderr, "> FOUND KEY ");
}
for (ii = 0; ii < nkey; ++ii) {
fprintf(stderr, "%c", key[ii]);
}
}
if (it != NULL) {
was_found = 1;
if (item_is_flushed(it)) {
do_item_unlink(it, hv);
do_item_remove(it);
it = NULL;
pthread_mutex_lock(&c->thread->stats.mutex);
c->thread->stats.get_flushed++;
pthread_mutex_unlock(&c->thread->stats.mutex);
if (settings.verbose > 2) {
fprintf(stderr, " -nuked by flush");
}
was_found = 2;
} else if (it->exptime != 0 && it->exptime <= current_time) {
do_item_unlink(it, hv);
do_item_remove(it);
it = NULL;
pthread_mutex_lock(&c->thread->stats.mutex);
c->thread->stats.get_expired++;
pthread_mutex_unlock(&c->thread->stats.mutex);
if (settings.verbose > 2) {
fprintf(stderr, " -nuked by expire");
}
was_found = 3;
} else {
it->it_flags |= ITEM_FETCHED|ITEM_ACTIVE;
DEBUG_REFCNT(it, '+');
}
}
if (settings.verbose > 2)
fprintf(stderr, "\n");
/* For now this is in addition to the above verbose logging. */
LOGGER_LOG(c->thread->l, LOG_FETCHERS, LOGGER_ITEM_GET, NULL, was_found, key, nkey);
return it;
}
