void mcache_delete(mcache *m, char *key, int key_len) {
(void)key_len;
(void)key;
assert(key);
assert(key_len > 0);
assert(key[key_len] == '\0' ||
key[key_len] == ' ');
if (m == NULL) {
return;
}
if (m->lock) {
cb_mutex_enter(m->lock);
}
if (m->map != NULL) {
void *existing = genhash_find(m->map, key);
if (existing != NULL) {
mcache_item_unlink(m, existing);
genhash_delete(m->map, key);
m->tot_deletes++;
if (settings.verbose > 1) {
moxi_log_write("mcache delete: %s\n", key);
}
}
}
if (m->lock) {
cb_mutex_exit(m->lock);
}
}
static ENGINE_ERROR_CODE mock_get(ENGINE_HANDLE* handle,
const void* cookie,
item** item,
const void* key,
const int nkey) {
*item = genhash_find(get_ht(handle), key, nkey);
return *item ? ENGINE_SUCCESS : ENGINE_KEY_ENOENT;
}
static ENGINE_ERROR_CODE mock_get(ENGINE_HANDLE* handle,
const void* cookie,
item** itm,
const void* key,
const int nkey,
uint16_t vbucket) {
(void)cookie;
(void)vbucket;
*itm = genhash_find(get_ht(handle), key, nkey);
return *itm ? ENGINE_SUCCESS : ENGINE_KEY_ENOENT;
}
static void
assert_hash_val(char *expected, genhash_t* h, const char* key)
{
char *found=(char*)genhash_find(h, key);
if(expected == NULL) {
if(found != NULL) {
fprintf(stderr, "Expected NULL for ``%s'', got ``%s''\n",
key, found);
abort();
}
} else {
if(strcmp(expected, found) != 0) {
fprintf(stderr, "Expected ``%s'' for ``%s'', got ``%s''\n",
expected, key, found);
abort();
}
}
}
/**
* Observe callback. Called internally by libcouchbase's observe handlers
*/
void lcb_durability_dset_update(lcb_t instance,
lcb_durability_set_t *dset,
lcb_error_t err,
const lcb_observe_resp_t *resp)
{
lcb_durability_entry_t *ent;
/**
* So we have two counters to decrement. One is the global 'done' counter
* and the other is the iteration counter.
*
* The iteration counter is only decremented when we receive a NULL signal
* in the callback, whereas the global counter is decremented once, whenever
* the entry's criteria have been satisfied
*/
if (resp->v.v0.key == NULL) {
dset_done_waiting(dset);
return;
}
if (dset->nentries == 1) {
ent = &dset->single.ent;
} else {
ent = genhash_find(dset->ht, resp->v.v0.key, resp->v.v0.nkey);
}
if (ent->done) {
/* ignore subsequent errors */
return;
}
if (err != LCB_SUCCESS) {
RESFLD(ent, err) = err;
return;
}
RESFLD(ent, nresponses)++;
if (resp->v.v0.cas && resp->v.v0.from_master) {
RESFLD(ent, cas) = resp->v.v0.cas;
if (REQFLD(ent, cas) && REQFLD(ent, cas) != resp->v.v0.cas) {
RESFLD(ent, err) = LCB_KEY_EEXISTS;
ent_set_resdone(ent);
return;
}
}
if (DSET_OPTFLD(ent->parent, check_delete)) {
check_negative_durability(ent, resp);
} else {
check_positive_durability(ent, resp);
}
if (ent_is_complete(ent)) {
/* clear any transient errors */
RESFLD(ent, err) = LCB_SUCCESS;
ent_set_resdone(ent);
}
(void)instance;
}
bool multiget_ascii_downstream(downstream *d, conn *uc,
int (*emit_start)(conn *c, char *cmd, int cmd_len),
int (*emit_skey)(conn *c, char *skey, int skey_len),
int (*emit_end)(conn *c),
mcache *front_cache) {
assert(d != NULL);
assert(d->downstream_conns != NULL);
assert(d->multiget == NULL);
assert(uc != NULL);
assert(uc->noreply == false);
proxy_td *ptd = d->ptd;
assert(ptd != NULL);
proxy_stats_cmd *psc_get =
&ptd->stats.stats_cmd[STATS_CMD_TYPE_REGULAR][STATS_CMD_GET];
proxy_stats_cmd *psc_get_key =
&ptd->stats.stats_cmd[STATS_CMD_TYPE_REGULAR][STATS_CMD_GET_KEY];
int nwrite = 0;
int nconns = mcs_server_count(&d->mst);
for (int i = 0; i < nconns; i++) {
if (d->downstream_conns[i] != NULL &&
cproxy_prep_conn_for_write(d->downstream_conns[i]) == false) {
d->ptd->stats.stats.err_downstream_write_prep++;
cproxy_close_conn(d->downstream_conns[i]);
return false;
}
}
if (uc->next != NULL) {
// More than one upstream conn, so we need a hashtable
// to track keys for de-deplication.
//
d->multiget = genhash_init(128, skeyhash_ops);
if (settings.verbose > 1) {
fprintf(stderr, "cproxy multiget hash table new\n");
}
}
// Snapshot the volatile only once.
//
uint32_t msec_current_time_snapshot = msec_current_time;
int uc_num = 0;
conn *uc_cur = uc;
while (uc_cur != NULL) {
assert(uc_cur->cmd == -1);
assert(uc_cur->item == NULL);
assert(uc_cur->state == conn_pause);
assert(IS_ASCII(uc_cur->protocol));
assert(IS_PROXY(uc_cur->protocol));
char *command = uc_cur->cmd_start;
assert(command != NULL);
char *space = strchr(command, ' ');
assert(space > command);
int cmd_len = space - command;
assert(cmd_len == 3 || cmd_len == 4); // Either get or gets.
int cas_emit = (command[3] == 's');
if (settings.verbose > 1) {
fprintf(stderr, "forward multiget %s (%d %d)\n",
command, cmd_len, uc_num);
}
while (space != NULL) {
char *key = space + 1;
char *next_space = strchr(key, ' ');
int key_len;
if (next_space != NULL) {
key_len = next_space - key;
} else {
key_len = strlen(key);
// We've reached the last key.
//
psc_get->read_bytes += (key - command + key_len);
}
// This key_len check helps skip consecutive spaces.
//
if (key_len > 0) {
ptd->stats.stats.tot_multiget_keys++;
psc_get_key->seen++;
psc_get_key->read_bytes += key_len;
// Update key-based statistics.
//
bool do_key_stats =
matcher_check(&ptd->key_stats_matcher,
key, key_len, true) == true &&
matcher_check(&ptd->key_stats_unmatcher,
key, key_len, false) == false;
if (do_key_stats) {
touch_key_stats(ptd, key, key_len,
msec_current_time_snapshot,
STATS_CMD_TYPE_REGULAR,
STATS_CMD_GET_KEY,
1, 0, 0,
key_len, 0);
}
// Handle a front cache hit by queuing response.
//
// Note, front cache stats are part of mcache.
//
if (!cas_emit) {
item *it = mcache_get(front_cache, key, key_len,
msec_current_time_snapshot);
if (it != NULL) {
assert(it->nkey == key_len);
assert(strncmp(ITEM_key(it), key, it->nkey) == 0);
cproxy_upstream_ascii_item_response(it, uc_cur, 0);
psc_get_key->hits++;
psc_get_key->write_bytes += it->nbytes;
if (do_key_stats) {
touch_key_stats(ptd, key, key_len,
msec_current_time_snapshot,
STATS_CMD_TYPE_REGULAR,
STATS_CMD_GET_KEY,
0, 1, 0,
0, it->nbytes);
}
// The refcount was inc'ed by mcache_get() for us.
//
item_remove(it);
goto loop_next;
}
}
bool self = false;
conn *c = cproxy_find_downstream_conn(d, key, key_len,
&self);
if (c != NULL) {
if (self) {
// Optimization for talking with ourselves,
// to avoid extra network hop.
//
ptd->stats.stats.tot_optimize_self++;
item *it = item_get(key, key_len);
if (it != NULL) {
cproxy_upstream_ascii_item_response(it, uc_cur,
cas_emit);
psc_get_key->hits++;
psc_get_key->write_bytes += it->nbytes;
if (do_key_stats) {
touch_key_stats(ptd, key, key_len,
msec_current_time_snapshot,
STATS_CMD_TYPE_REGULAR,
STATS_CMD_GET_KEY,
0, 1, 0,
0, it->nbytes);
}
// The refcount was inc'ed by item_get() for us.
//
item_remove(it);
if (settings.verbose > 1) {
fprintf(stderr,
"optimize self multiget hit: %s\n",
key);
}
} else {
psc_get_key->misses++;
if (do_key_stats) {
touch_key_stats(ptd, key, key_len,
msec_current_time_snapshot,
STATS_CMD_TYPE_REGULAR,
STATS_CMD_GET_KEY,
0, 0, 1,
0, 0);
}
if (settings.verbose > 1) {
fprintf(stderr,
"optimize self multiget miss: %s\n",
key);
}
}
goto loop_next;
}
// See if we've already requested this key via
// the multiget hash table, in order to
// de-deplicate repeated keys.
//
bool first_request = true;
if (d->multiget != NULL) {
// TODO: Use Trond's allocator here.
//
multiget_entry *entry =
calloc(1, sizeof(multiget_entry));
if (entry != NULL) {
entry->upstream_conn = uc_cur;
entry->opaque = 0;
entry->hits = 0;
entry->next = genhash_find(d->multiget, key);
genhash_update(d->multiget, key, entry);
if (entry->next != NULL) {
first_request = false;
}
} else {
// TODO: Handle out of multiget entry memory.
}
}
if (first_request) {
assert(c->item == NULL);
assert(c->state == conn_pause);
assert(IS_PROXY(c->protocol));
assert(c->ilist != NULL);
assert(c->isize > 0);
if (c->msgused <= 1 &&
c->msgbytes <= 0) {
emit_start(c, command, cmd_len);
}
// Provide the preceding space as optimization
// for ascii-to-ascii configuration.
//
emit_skey(c, key - 1, key_len + 1);
} else {
ptd->stats.stats.tot_multiget_keys_dedupe++;
if (settings.verbose > 1) {
char buf[KEY_MAX_LENGTH + 10];
memcpy(buf, key, key_len);
buf[key_len] = '\0';
fprintf(stderr,
"%d cproxy multiget dedpue: %s\n",
uc_cur->sfd, buf);
}
}
} else {
// TODO: Handle when downstream conn is down.
}
}
loop_next:
space = next_space;
}
uc_num++;
uc_cur = uc_cur->next;
}
for (int i = 0; i < nconns; i++) {
conn *c = d->downstream_conns[i];
if (c != NULL &&
(c->msgused > 1 ||
c->msgbytes > 0)) {
emit_end(c);
conn_set_state(c, conn_mwrite);
c->write_and_go = conn_new_cmd;
if (update_event(c, EV_WRITE | EV_PERSIST)) {
nwrite++;
if (uc->noreply) {
c->write_and_go = conn_pause;
}
} else {
if (settings.verbose > 1) {
fprintf(stderr,
"Couldn't update cproxy write event\n");
}
d->ptd->stats.stats.err_oom++;
cproxy_close_conn(c);
}
}
}
if (settings.verbose > 1) {
fprintf(stderr, "forward multiget nwrite %d out of %d\n",
nwrite, nconns);
}
d->downstream_used_start = nwrite;
d->downstream_used = nwrite;
if (cproxy_dettach_if_noreply(d, uc) == false) {
d->upstream_suffix = "END\r\n";
cproxy_start_downstream_timeout(d, NULL);
}
return nwrite > 0;
}
void multiget_ascii_downstream_response(downstream *d, item *it) {
assert(d);
assert(it);
assert(it->nkey > 0);
assert(ITEM_key(it) != NULL);
proxy_td *ptd = d->ptd;
assert(ptd);
proxy_stats_cmd *psc_get_key =
&ptd->stats.stats_cmd[STATS_CMD_TYPE_REGULAR][STATS_CMD_GET_KEY];
proxy *p = ptd->proxy;
assert(p);
uint32_t front_cache_lifespan =
ptd->behavior_pool.base.front_cache_lifespan;
if (front_cache_lifespan > 0) {
if (matcher_check(&p->front_cache_matcher,
ITEM_key(it), it->nkey, true) == true &&
matcher_check(&p->front_cache_unmatcher,
ITEM_key(it), it->nkey, false) == false) {
mcache_set(&p->front_cache, it,
front_cache_lifespan + msec_current_time,
true, false);
}
}
if (d->multiget != NULL) {
// The ITEM_key is not NULL or space terminated.
//
char key_buf[KEY_MAX_LENGTH + 10];
assert(it->nkey <= KEY_MAX_LENGTH);
memcpy(key_buf, ITEM_key(it), it->nkey);
key_buf[it->nkey] = '\0';
multiget_entry *entry_first = genhash_find(d->multiget, key_buf);
if (entry_first != NULL) {
entry_first->hits++;
multiget_entry *entry = entry_first;
while (entry != NULL) {
// The upstream might have been closed mid-request.
//
// TODO: Revisit the -1 cas_emit parameter.
//
conn *uc = entry->upstream_conn;
if (uc != NULL) {
cproxy_upstream_ascii_item_response(it, uc, -1);
psc_get_key->hits++;
psc_get_key->write_bytes += it->nbytes;
if (matcher_check(&ptd->key_stats_matcher,
ITEM_key(it), it->nkey, true) == true &&
matcher_check(&ptd->key_stats_unmatcher,
ITEM_key(it), it->nkey, false) == false) {
touch_key_stats(ptd, ITEM_key(it), it->nkey,
msec_current_time,
STATS_CMD_TYPE_REGULAR,
STATS_CMD_GET_KEY,
0, 1, 0,
0, it->nbytes);
}
if (entry != entry_first) {
ptd->stats.stats.tot_multiget_bytes_dedupe += it->nbytes;
}
}
entry = entry->next;
}
}
} else {
// TODO: We're not tracking miss stats in the simple case.
// Do we always need to use a multiget hashtable?
// Or, perhaps misses equals number of requests - number of hits.
//
conn *uc = d->upstream_conn;
while (uc != NULL) {
// TODO: Revisit the -1 cas_emit parameter.
//
cproxy_upstream_ascii_item_response(it, uc, -1);
psc_get_key->hits++;
psc_get_key->write_bytes += it->nbytes;
if (matcher_check(&ptd->key_stats_matcher,
ITEM_key(it), it->nkey, true) == true &&
matcher_check(&ptd->key_stats_unmatcher,
ITEM_key(it), it->nkey, false) == false) {
touch_key_stats(ptd, ITEM_key(it), it->nkey,
msec_current_time,
STATS_CMD_TYPE_REGULAR,
STATS_CMD_GET_KEY,
0, 1, 0,
0, it->nbytes);
}
uc = uc->next;
}
}
}
void mcache_set(mcache *m, void *it,
uint64_t exptime,
bool add_only,
bool mod_exptime_if_exists) {
assert(it);
assert(m->funcs);
assert(m->funcs->item_get_next(it) == NULL);
assert(m->funcs->item_get_prev(it) == NULL);
if (m == NULL) {
return;
}
/* TODO: Our lock areas are possibly too wide. */
if (m->lock) {
cb_mutex_enter(m->lock);
}
if (m->map != NULL) {
/* Evict some items if necessary. */
int i;
for (i = 0; m->lru_tail != NULL && i < 20; i++) {
void *last_it;
if ((uint32_t)genhash_size(m->map) < m->max) {
break;
}
last_it = m->lru_tail;
mcache_item_unlink(m, last_it);
if (m->key_alloc) {
int len = m->funcs->item_key_len(last_it);
char buf[KEY_MAX_LENGTH + 10];
memcpy(buf, m->funcs->item_key(last_it), len);
buf[len] = '\0';
genhash_delete(m->map, buf);
} else {
genhash_delete(m->map, m->funcs->item_key(last_it));
}
m->tot_evictions++;
}
if ((uint32_t)genhash_size(m->map) < m->max) {
char *key = m->funcs->item_key(it);
int key_len = m->funcs->item_key_len(it);
char *key_buf = NULL;
if (m->key_alloc) {
/* The ITEM_key is not NULL or space terminated, */
/* and we need a copy, too, for hashtable ownership. */
/* TODO: Move this outside the lock area? */
key_buf = malloc(key_len + 1);
if (key_buf != NULL) {
memcpy(key_buf, key, key_len);
key_buf[key_len] = '\0';
key = key_buf;
} else {
key = NULL;
}
}
if (key != NULL) {
void *existing = add_only ? genhash_find(m->map, key) : NULL;
if (existing != NULL) {
mcache_item_unlink(m, existing);
mcache_item_touch(m, existing);
if (mod_exptime_if_exists) {
m->funcs->item_set_exptime(existing, exptime);
}
m->tot_add_skips++;
if (settings.verbose > 1) {
moxi_log_write("mcache add-skip: %s\n", key);
}
if (key_buf != NULL) {
free(key_buf);
}
} else {
m->funcs->item_set_exptime(it, exptime);
m->funcs->item_add_ref(it);
genhash_update(m->map, key, it);
m->tot_adds++;
m->tot_add_bytes += m->funcs->item_len(it);
if (settings.verbose > 1) {
moxi_log_write("mcache add: %s\n", key);
}
}
} else {
m->tot_add_fails++;
}
} else {
m->tot_add_fails++;
}
}
if (m->lock) {
cb_mutex_exit(m->lock);
}
}
void *mcache_get(mcache *m, char *key, int key_len,
uint64_t curr_time) {
(void)key_len;
assert(key);
if (m == NULL) {
return NULL;
}
assert(m->funcs);
if (m->lock) {
cb_mutex_enter(m->lock);
}
if (m->map != NULL) {
void *it = genhash_find(m->map, key);
if (it != NULL) {
mcache_item_unlink(m, it);
uint64_t exptime = m->funcs->item_get_exptime(it);
if ((exptime <= 0) ||
(exptime >= curr_time &&
exptime >= m->oldest_live)) {
mcache_item_touch(m, it);
m->funcs->item_add_ref(it); /* TODO: Need lock here? */
m->tot_get_hits++;
m->tot_get_bytes += m->funcs->item_len(it);
if (m->lock) {
cb_mutex_exit(m->lock);
}
if (settings.verbose > 1) {
moxi_log_write("mcache hit: %s\n", key);
}
return it;
}
/* Handle item expiration. */
m->tot_get_expires++;
if (settings.verbose > 1) {
moxi_log_write("mcache expire: %s\n", key);
}
genhash_delete(m->map, key);
} else {
m->tot_get_misses++;
if (settings.verbose > 1) {
moxi_log_write("mcache miss: %s\n", key);
}
}
}
if (m->lock) {
cb_mutex_exit(m->lock);
}
return NULL;
}
bool protocol_stats_merge_name_val(genhash_t *merger,
char *prefix,
int prefix_len,
char *name,
int name_len,
char *val,
int val_len) {
char *key;
int key_len;
cb_assert(merger);
cb_assert(name);
cb_assert(val);
key = name + name_len - 1; /* Key part for merge rule lookup. */
while (key >= name && *key != ':') { /* Scan for last colon. */
key--;
}
if (key < name) {
key = name;
}
key_len = name_len - (key - name);
if (key_len > 0 && key_len < MERGE_BUF_SIZE) {
char buf_name[MERGE_BUF_SIZE];
char buf_key[MERGE_BUF_SIZE];
char buf_val[MERGE_BUF_SIZE];
char *prev;
token_t prev_tokens[MAX_TOKENS];
size_t prev_ntokens;
bool ok;
strncpy(buf_name, name, name_len);
buf_name[name_len] = '\0';
prev = (char *) genhash_find(merger, buf_name);
if (prev == NULL) {
char *hval = malloc(prefix_len + 1 +
name_len + 1 +
val_len + 1);
if (hval != NULL) {
memcpy(hval, prefix, prefix_len);
hval[prefix_len] = ' ';
memcpy(hval + prefix_len + 1, name, name_len);
hval[prefix_len + 1 + name_len] = ' ';
memcpy(hval + prefix_len + 1 + name_len + 1, val, val_len);
hval[prefix_len + 1 + name_len + 1 + val_len] = '\0';
genhash_update(merger, hval + prefix_len + 1, hval);
}
return true;
}
strncpy(buf_key, key, key_len);
buf_key[key_len] = '\0';
if (strstr(protocol_stats_keys_first, buf_key) != NULL) {
return true;
}
prev_ntokens = scan_tokens(prev, prev_tokens, MAX_TOKENS, NULL);
if (prev_ntokens != 4) {
return true;
}
strncpy(buf_val, val, val_len);
buf_val[val_len] = '\0';
if (strstr(protocol_stats_keys_smallest, buf_key) != NULL) {
ok = protocol_stats_merge_smallest(prev_tokens[VALUE_TOKEN].value,
prev_tokens[VALUE_TOKEN].length,
buf_val, val_len,
buf_val, MERGE_BUF_SIZE);
} else {
ok = protocol_stats_merge_sum(prev_tokens[VALUE_TOKEN].value,
prev_tokens[VALUE_TOKEN].length,
buf_val, val_len,
buf_val, MERGE_BUF_SIZE);
}
if (ok) {
int vlen = strlen(buf_val);
char *hval = malloc(prefix_len + 1 +
name_len + 1 +
vlen + 1);
if (hval != NULL) {
memcpy(hval, prefix, prefix_len);
hval[prefix_len] = ' ';
memcpy(hval + prefix_len + 1, name, name_len);
hval[prefix_len + 1 + name_len] = ' ';
strcpy(hval + prefix_len + 1 + name_len + 1, buf_val);
hval[prefix_len + 1 + name_len + 1 + vlen] = '\0';
genhash_update(merger, hval + prefix_len + 1, hval);
free(prev);
}
}
/* Note, if we couldn't merge, then just keep */
/* the previous value. */
return true;
}
return false;
}
/**
* Observe callback. Called internally by libcouchbase's observe handlers
*/
void
lcb_durability_dset_update(lcb_t instance,
lcb_DURSET *dset, lcb_error_t err, const lcb_RESPOBSERVE *resp)
{
lcb_DURITEM *ent;
/**
* So we have two counters to decrement. One is the global 'done' counter
* and the other is the iteration counter.
*
* The iteration counter is only decremented when we receive a NULL signal
* in the callback, whereas the global counter is decremented once, whenever
* the entry's criteria have been satisfied
*/
if (resp->key == NULL) {
dset_done_waiting(dset);
return;
}
if (dset->nentries == 1) {
ent = &dset->single.ent;
} else {
ent = genhash_find(dset->ht, resp->key, resp->nkey);
}
if (ent->done) {
/* ignore subsequent errors */
return;
}
if (err != LCB_SUCCESS) {
RESFLD(ent, rc) = err;
/* If it's a non-scheduling error then the item will be retried in the
* next iteration */
if (err == LCB_SCHEDFAIL_INTERNAL) {
ent_set_resdone(ent);
}
return;
}
RESFLD(ent, nresponses)++;
if (resp->cas && resp->ismaster) {
RESFLD(ent, cas) = resp->cas;
if (ent->reqcas && ent->reqcas != resp->cas) {
RESFLD(ent, rc) = LCB_KEY_EEXISTS;
ent_set_resdone(ent);
return;
}
}
if (DSET_OPTFLD(ent->parent, check_delete)) {
check_negative_durability(ent, resp);
} else {
check_positive_durability(ent, resp);
}
if (ent_is_complete(ent)) {
/* clear any transient errors */
RESFLD(ent, rc) = LCB_SUCCESS;
ent_set_resdone(ent);
}
(void)instance;
}