//将item插入到哈希表和LRU队列中，hv为哈希值
int do_item_link(item *it, const uint32_t hv) {
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nkey, it->nbytes);
//确保这个item已经从slab分配出去并且还没插入到LRU队列中
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
mutex_lock(&cache_lock);
//加入link标记
it->it_flags |= ITEM_LINKED;
it->time = current_time;
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
/* Allocate a new CAS ID on link. */
ITEM_set_cas(it, (settings.use_cas) ? get_cas_id() : 0);
//插入到hash表中
assoc_insert(it, hv);
//item插入到链表中
item_link_q(it);
//引用计数加1
refcount_incr(&it->refcount);
mutex_unlock(&cache_lock);
return 1;
}
/* 形成了一个完成的 item 后, 就要把它放入两个数据结构中, 一是 memcached 的哈希表,
memcached 运行过程中只有一个哈希表, 二是 item 所在的 slabclass 的 LRU 队列. */
int do_item_link(item *it, const uint32_t hv) {
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nkey, it->nbytes);
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
mutex_lock(&cache_lock);
it->it_flags |= ITEM_LINKED;
it->time = current_time;
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
/* Allocate a new CAS ID on link. */
ITEM_set_cas(it, (settings.use_cas) ? get_cas_id() : 0);
/* 把 item 放入哈希表 */
assoc_insert(it, hv);
/* 把 item 放入 LRU 队列*/
item_link_q(it);
refcount_incr(&it->refcount);
mutex_unlock(&cache_lock);
return 1;
}
int do_item_link(item *it) {
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nkey, it->nbytes);
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
assert(it->nbytes < (1024 * 1024)); /* 1MB max size */
it->it_flags |= ITEM_LINKED;
it->time = current_time;
assoc_insert(it);
#ifdef MOXI_ITEM_MALLOC
it->refcount++;
#endif
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
/* Allocate a new CAS ID on link. */
ITEM_set_cas(it, (settings.use_cas) ? get_cas_id() : 0);
item_link_q(it);
return 1;
}
//将item加入到hashtable和LRU链中
int do_item_link(item *it, const uint32_t hv) {
syslog(LOG_INFO, "[%s:%s:%d]", __FILE__, __func__, __LINE__);
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nkey, it->nbytes);
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0); //判断状态，即没有在hash表LRU链中或被释放
mutex_lock(&cache_lock);
it->it_flags |= ITEM_LINKED; //设置linked状态
it->time = current_time;//设置最近访问的时间
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it); //增加每个item所需要的字节大小，包括item结构体和item内容大小
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
/* Allocate a new CAS ID on link. */
ITEM_set_cas(it, (settings.use_cas) ? get_cas_id() : 0); //设置新CAS，CAS是memcache用来处理并发请求的一种机制
assoc_insert(it, hv);//插入hashtable
item_link_q(it); //加入LRU链
refcount_incr(&it->refcount);
mutex_unlock(&cache_lock);
return 1;
}
void cproxy_process_a2a_downstream(conn *c, char *line) {
assert(c != NULL);
assert(c->next == NULL);
assert(c->extra != NULL);
assert(c->cmd == -1);
assert(c->item == NULL);
assert(line != NULL);
assert(line == c->rcurr);
assert(IS_ASCII(c->protocol));
assert(IS_PROXY(c->protocol));
if (settings.verbose > 1)
fprintf(stderr, "<%d cproxy_process_a2a_downstream %s\n",
c->sfd, line);
downstream *d = c->extra;
assert(d != NULL);
assert(d->ptd != NULL);
assert(d->ptd->proxy != NULL);
if (strncmp(line, "VALUE ", 6) == 0) {
token_t tokens[MAX_TOKENS];
size_t ntokens;
unsigned int flags;
int clen = 0;
int vlen;
uint64_t cas = CPROXY_NOT_CAS;
ntokens = scan_tokens(line, tokens, MAX_TOKENS, &clen);
if (ntokens >= 5 && // Accounts for extra termimation token.
ntokens <= 6 &&
tokens[KEY_TOKEN].length <= KEY_MAX_LENGTH &&
safe_strtoul(tokens[2].value, (uint32_t *) &flags) &&
safe_strtoul(tokens[3].value, (uint32_t *) &vlen)) {
char *key = tokens[KEY_TOKEN].value;
size_t nkey = tokens[KEY_TOKEN].length;
item *it = item_alloc(key, nkey, flags, 0, vlen + 2);
if (it != NULL) {
if (ntokens == 5 ||
safe_strtoull(tokens[4].value, &cas)) {
ITEM_set_cas(it, cas);
c->item = it;
c->ritem = ITEM_data(it);
c->rlbytes = it->nbytes;
c->cmd = -1;
conn_set_state(c, conn_nread);
return; // Success.
} else {
if (settings.verbose > 1)
fprintf(stderr, "cproxy could not parse cas\n");
}
} else {
if (settings.verbose > 1)
fprintf(stderr, "cproxy could not item_alloc size %u\n",
vlen + 2);
}
if (it != NULL)
item_remove(it);
it = NULL;
c->sbytes = vlen + 2; // Number of bytes to swallow.
conn_set_state(c, conn_swallow);
// Note, eventually, we'll see an END later.
} else {
// We don't know how much to swallow, so close the downstream.
// The conn_closing should release the downstream,
// which should write a suffix/error to the upstream.
//
conn_set_state(c, conn_closing);
}
} else if (strncmp(line, "END", 3) == 0) {
conn_set_state(c, conn_pause);
} else if (strncmp(line, "OK", 2) == 0) {
conn_set_state(c, conn_pause);
// TODO: Handle flush_all's expiration parameter against
// the front_cache.
//
// TODO: We flush the front_cache too often, inefficiently
// on every downstream flush_all OK response, rather than
// on just the last flush_all OK response.
//
conn *uc = d->upstream_conn;
if (uc != NULL &&
uc->cmd_curr == PROTOCOL_BINARY_CMD_FLUSH) {
mcache_flush_all(&d->ptd->proxy->front_cache, 0);
}
} else if (strncmp(line, "STAT ", 5) == 0 ||
strncmp(line, "ITEM ", 5) == 0 ||
strncmp(line, "PREFIX ", 7) == 0) {
assert(d->merger != NULL);
conn *uc = d->upstream_conn;
if (uc != NULL) {
assert(uc->next == NULL);
if (protocol_stats_merge_line(d->merger, line) == false) {
// Forward the line as-is if we couldn't merge it.
//
int nline = strlen(line);
item *it = item_alloc("s", 1, 0, 0, nline + 2);
if (it != NULL) {
strncpy(ITEM_data(it), line, nline);
strncpy(ITEM_data(it) + nline, "\r\n", 2);
if (add_conn_item(uc, it)) {
add_iov(uc, ITEM_data(it), nline + 2);
it = NULL;
}
if (it != NULL)
item_remove(it);
}
}
}
conn_set_state(c, conn_new_cmd);
} else {
conn_set_state(c, conn_pause);
// The upstream conn might be NULL when closed already
// or while handling a noreply.
//
conn *uc = d->upstream_conn;
if (uc != NULL) {
assert(uc->next == NULL);
out_string(uc, line);
if (!update_event(uc, EV_WRITE | EV_PERSIST)) {
if (settings.verbose > 1)
fprintf(stderr,
"Can't update upstream write event\n");
d->ptd->stats.stats.err_oom++;
cproxy_close_conn(uc);
}
cproxy_del_front_cache_key_ascii_response(d, line,
uc->cmd_start);
}
}
}
static int storage_write(void *storage, const int clsid, const int item_age) {
int did_moves = 0;
struct lru_pull_tail_return it_info;
it_info.it = NULL;
lru_pull_tail(clsid, COLD_LRU, 0, LRU_PULL_RETURN_ITEM, 0, &it_info);
/* Item is locked, and we have a reference to it. */
if (it_info.it == NULL) {
return did_moves;
}
obj_io io;
item *it = it_info.it;
/* First, storage for the header object */
size_t orig_ntotal = ITEM_ntotal(it);
uint32_t flags;
if ((it->it_flags & ITEM_HDR) == 0 &&
(item_age == 0 || current_time - it->time > item_age)) {
FLAGS_CONV(it, flags);
item *hdr_it = do_item_alloc(ITEM_key(it), it->nkey, flags, it->exptime, sizeof(item_hdr));
/* Run the storage write understanding the start of the item is dirty.
* We will fill it (time/exptime/etc) from the header item on read.
*/
if (hdr_it != NULL) {
int bucket = (it->it_flags & ITEM_CHUNKED) ?
PAGE_BUCKET_CHUNKED : PAGE_BUCKET_DEFAULT;
// Compress soon to expire items into similar pages.
if (it->exptime - current_time < settings.ext_low_ttl) {
bucket = PAGE_BUCKET_LOWTTL;
}
hdr_it->it_flags |= ITEM_HDR;
io.len = orig_ntotal;
io.mode = OBJ_IO_WRITE;
// NOTE: when the item is read back in, the slab mover
// may see it. Important to have refcount>=2 or ~ITEM_LINKED
assert(it->refcount >= 2);
// NOTE: write bucket vs free page bucket will disambiguate once
// lowttl feature is better understood.
if (extstore_write_request(storage, bucket, bucket, &io) == 0) {
// cuddle the hash value into the time field so we don't have
// to recalculate it.
item *buf_it = (item *) io.buf;
buf_it->time = it_info.hv;
// copy from past the headers + time headers.
// TODO: should be in items.c
if (it->it_flags & ITEM_CHUNKED) {
// Need to loop through the item and copy
item_chunk *sch = (item_chunk *) ITEM_schunk(it);
int remain = orig_ntotal;
int copied = 0;
// copy original header
int hdrtotal = ITEM_ntotal(it) - it->nbytes;
memcpy((char *)io.buf+STORE_OFFSET, (char *)it+STORE_OFFSET, hdrtotal - STORE_OFFSET);
copied = hdrtotal;
// copy data in like it were one large object.
while (sch && remain) {
assert(remain >= sch->used);
memcpy((char *)io.buf+copied, sch->data, sch->used);
// FIXME: use one variable?
remain -= sch->used;
copied += sch->used;
sch = sch->next;
}
} else {
memcpy((char *)io.buf+STORE_OFFSET, (char *)it+STORE_OFFSET, io.len-STORE_OFFSET);
}
// crc what we copied so we can do it sequentially.
buf_it->it_flags &= ~ITEM_LINKED;
buf_it->exptime = crc32c(0, (char*)io.buf+STORE_OFFSET, orig_ntotal-STORE_OFFSET);
extstore_write(storage, &io);
item_hdr *hdr = (item_hdr *) ITEM_data(hdr_it);
hdr->page_version = io.page_version;
hdr->page_id = io.page_id;
hdr->offset = io.offset;
// overload nbytes for the header it
hdr_it->nbytes = it->nbytes;
/* success! Now we need to fill relevant data into the new
* header and replace. Most of this requires the item lock
*/
/* CAS gets set while linking. Copy post-replace */
item_replace(it, hdr_it, it_info.hv);
ITEM_set_cas(hdr_it, ITEM_get_cas(it));
do_item_remove(hdr_it);
did_moves = 1;
LOGGER_LOG(NULL, LOG_EVICTIONS, LOGGER_EXTSTORE_WRITE, it, bucket);
} else {
/* Failed to write for some reason, can't continue. */
slabs_free(hdr_it, ITEM_ntotal(hdr_it), ITEM_clsid(hdr_it));
}
}
}
do_item_remove(it);
item_unlock(it_info.hv);
return did_moves;
}
