static void
req_forward_remote_dc(struct context *ctx, struct conn *c_conn, struct msg *msg,
struct mbuf *orig_mbuf, uint8_t *key, uint32_t keylen,
struct datacenter *dc)
{
uint32_t rack_cnt = array_n(&dc->racks);
if (rack_cnt == 0)
return;
struct rack *rack = dc->preselected_rack_for_replication;
if (rack == NULL)
rack = array_get(&dc->racks, 0);
struct msg *rack_msg = msg_get(c_conn, msg->request, __FUNCTION__);
if (rack_msg == NULL) {
log_debug(LOG_VERB, "whelp, looks like yer screwed now, buddy. no inter-rack messages for you!");
msg_put(rack_msg);
return;
}
msg_clone(msg, orig_mbuf, rack_msg);
log_info("msg (%d:%d) clone to remote rack msg (%d:%d)",
msg->id, msg->parent_id, rack_msg->id, rack_msg->parent_id);
rack_msg->swallow = true;
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "forwarding request to conn '%s' on rack '%.*s'",
dn_unresolve_peer_desc(c_conn->sd), rack->name->len, rack->name->data);
}
remote_req_forward(ctx, c_conn, rack_msg, rack, key, keylen);
}
static void
core_close_log(struct conn *conn)
{
char *addrstr;
if ((conn->type == CONN_CLIENT) || (conn->type == CONN_DNODE_PEER_CLIENT)) {
addrstr = dn_unresolve_peer_desc(conn->sd);
} else {
addrstr = dn_unresolve_addr(conn->addr, conn->addrlen);
}
log_debug(LOG_NOTICE, "close %s %d '%s' on event %04"PRIX32" eof %d done "
"%d rb %zu sb %zu%c %s", conn_get_type_string(conn), conn->sd,
addrstr, conn->events, conn->eof, conn->done, conn->recv_bytes,
conn->send_bytes,
conn->err ? ':' : ' ', conn->err ? strerror(conn->err) : "");
}
static void
req_forward_all_local_racks(struct context *ctx, struct conn *c_conn,
struct msg *msg, struct mbuf *orig_mbuf,
uint8_t *key, uint32_t keylen, struct datacenter *dc)
{
//log_debug(LOG_DEBUG, "dc name '%.*s'",
// dc->name->len, dc->name->data);
uint8_t rack_cnt = (uint8_t)array_n(&dc->racks);
uint8_t rack_index;
msg->rsp_handler = msg_get_rsp_handler(msg);
init_response_mgr(&msg->rspmgr, msg, msg->is_read, rack_cnt, c_conn);
log_info("msg %d:%d same DC racks:%d expect replies %d",
msg->id, msg->parent_id, rack_cnt, msg->rspmgr.max_responses);
for(rack_index = 0; rack_index < rack_cnt; rack_index++) {
struct rack *rack = array_get(&dc->racks, rack_index);
//log_debug(LOG_DEBUG, "rack name '%.*s'",
// rack->name->len, rack->name->data);
struct msg *rack_msg;
// clone message even for local node
struct server_pool *pool = c_conn->owner;
if (string_compare(rack->name, &pool->rack) == 0 ) {
rack_msg = msg;
} else {
rack_msg = msg_get(c_conn, msg->request, __FUNCTION__);
if (rack_msg == NULL) {
log_debug(LOG_VERB, "whelp, looks like yer screwed "
"now, buddy. no inter-rack messages for "
"you!");
continue;
}
msg_clone(msg, orig_mbuf, rack_msg);
log_info("msg (%d:%d) clone to rack msg (%d:%d)",
msg->id, msg->parent_id, rack_msg->id, rack_msg->parent_id);
rack_msg->swallow = true;
}
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "forwarding request to conn '%s' on rack '%.*s'",
dn_unresolve_peer_desc(c_conn->sd), rack->name->len, rack->name->data);
}
log_debug(LOG_VERB, "c_conn: %p forwarding (%d:%d)",
c_conn, rack_msg->id, rack_msg->parent_id);
remote_req_forward(ctx, c_conn, rack_msg, rack, key, keylen);
}
}
static void
core_close_log(struct conn *conn)
{
char type, *addrstr;
if (conn->client) {
type = 'c';
addrstr = dn_unresolve_peer_desc(conn->sd);
} else {
type = conn->proxy ? 'p' : 's';
addrstr = dn_unresolve_addr(conn->addr, conn->addrlen);
}
log_debug(LOG_NOTICE, "close %c %d '%s' on event %04"PRIX32" eof %d done "
"%d rb %zu sb %zu%c %s", type, conn->sd, addrstr, conn->events,
conn->eof, conn->done, conn->recv_bytes, conn->send_bytes,
conn->err ? ':' : ' ', conn->err ? strerror(conn->err) : "");
}
static rstatus_t
proxy_accept(struct context *ctx, struct conn *p)
{
rstatus_t status;
struct conn *c;
int sd;
ASSERT(p->proxy && !p->client);
ASSERT(p->sd > 0);
ASSERT(p->recv_active && p->recv_ready);
for (;;) {
sd = accept(p->sd, NULL, NULL);
if (sd < 0) {
if (errno == EINTR) {
log_debug(LOG_VERB, "accept on p %d not ready - eintr", p->sd);
continue;
}
if (errno == EAGAIN || errno == EWOULDBLOCK) {
log_debug(LOG_VERB, "accept on p %d not ready - eagain", p->sd);
p->recv_ready = 0;
return DN_OK;
}
/*
* FIXME: On EMFILE or ENFILE mask out IN event on the proxy; mask
* it back in when some existing connection gets closed
*/
log_error("accept on p %d failed: %s", p->sd, strerror(errno));
return DN_ERROR;
}
break;
}
c = conn_get(p->owner, true, p->data_store);
if (c == NULL) {
log_error("get conn for c %d from p %d failed: %s", sd, p->sd,
strerror(errno));
status = close(sd);
if (status < 0) {
log_error("close c %d failed, ignored: %s", sd, strerror(errno));
}
return DN_ENOMEM;
}
c->sd = sd;
stats_pool_incr(ctx, c->owner, client_connections);
status = dn_set_nonblocking(c->sd);
if (status < 0) {
log_error("set nonblock on c %d from p %d failed: %s", c->sd, p->sd,
strerror(errno));
c->close(ctx, c);
return status;
}
if (p->family == AF_INET || p->family == AF_INET6) {
status = dn_set_tcpnodelay(c->sd);
if (status < 0) {
log_warn("set tcpnodelay on c %d from p %d failed, ignored: %s",
c->sd, p->sd, strerror(errno));
}
}
status = event_add_conn(ctx->evb, c);
if (status < 0) {
log_error("event add conn from p %d failed: %s", p->sd,
strerror(errno));
c->close(ctx, c);
return status;
}
log_debug(LOG_NOTICE, "accepted c %d on p %d from '%s'", c->sd, p->sd,
dn_unresolve_peer_desc(c->sd));
return DN_OK;
}
static void
dnode_req_forward(struct context *ctx, struct conn *conn, struct msg *msg)
{
struct server_pool *pool;
uint8_t *key;
uint32_t keylen;
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "dnode_req_forward entering ");
}
log_debug(LOG_DEBUG, "DNODE REQ RECEIVED %s %d dmsg->id %u",
conn_get_type_string(conn), conn->sd, msg->dmsg->id);
ASSERT(conn->type == CONN_DNODE_PEER_CLIENT);
pool = conn->owner;
key = NULL;
keylen = 0;
log_debug(LOG_DEBUG, "conn %p adding message %d:%d", conn, msg->id, msg->parent_id);
dictAdd(conn->outstanding_msgs_dict, &msg->id, msg);
if (!string_empty(&pool->hash_tag)) {
struct string *tag = &pool->hash_tag;
uint8_t *tag_start, *tag_end;
tag_start = dn_strchr(msg->key_start, msg->key_end, tag->data[0]);
if (tag_start != NULL) {
tag_end = dn_strchr(tag_start + 1, msg->key_end, tag->data[1]);
if (tag_end != NULL) {
key = tag_start + 1;
keylen = (uint32_t)(tag_end - key);
}
}
}
if (keylen == 0) {
key = msg->key_start;
keylen = (uint32_t)(msg->key_end - msg->key_start);
}
ASSERT(msg->dmsg != NULL);
if (msg->dmsg->type == DMSG_REQ) {
local_req_forward(ctx, conn, msg, key, keylen);
} else if (msg->dmsg->type == DMSG_REQ_FORWARD) {
struct mbuf *orig_mbuf = STAILQ_FIRST(&msg->mhdr);
struct datacenter *dc = server_get_dc(pool, &pool->dc);
uint32_t rack_cnt = array_n(&dc->racks);
uint32_t rack_index;
for(rack_index = 0; rack_index < rack_cnt; rack_index++) {
struct rack *rack = array_get(&dc->racks, rack_index);
//log_debug(LOG_DEBUG, "forwarding to rack '%.*s'",
// rack->name->len, rack->name->data);
struct msg *rack_msg;
if (string_compare(rack->name, &pool->rack) == 0 ) {
rack_msg = msg;
} else {
rack_msg = msg_get(conn, msg->request, __FUNCTION__);
if (rack_msg == NULL) {
log_debug(LOG_VERB, "whelp, looks like yer screwed now, buddy. no inter-rack messages for you!");
continue;
}
if (msg_clone(msg, orig_mbuf, rack_msg) != DN_OK) {
msg_put(rack_msg);
continue;
}
rack_msg->swallow = true;
}
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "forwarding request from conn '%s' to rack '%.*s' dc '%.*s' ",
dn_unresolve_peer_desc(conn->sd), rack->name->len, rack->name->data, rack->dc->len, rack->dc->data);
}
remote_req_forward(ctx, conn, rack_msg, rack, key, keylen);
}
}
}
/* Forward a client request over to a peer */
void dnode_peer_req_forward(struct context *ctx, struct conn *c_conn,
struct conn *p_conn, struct msg *msg,
struct rack *rack, uint8_t *key, uint32_t keylen)
{
struct server *server = p_conn->owner;
log_debug(LOG_DEBUG, "forwarding request from client conn '%s' to peer conn '%s' on rack '%.*s' dc '%.*s' ",
dn_unresolve_peer_desc(c_conn->sd), dn_unresolve_peer_desc(p_conn->sd),
rack->name->len, rack->name->data,
server->dc.len, server->dc.data);
struct string *dc = rack->dc;
rstatus_t status;
/* enqueue message (request) into client outq, if response is expected */
if (!msg->noreply && !msg->swallow) {
conn_enqueue_outq(ctx, c_conn, msg);
}
ASSERT(p_conn->type == CONN_DNODE_PEER_SERVER);
ASSERT((c_conn->type == CONN_CLIENT) ||
(c_conn->type == CONN_DNODE_PEER_CLIENT));
/* enqueue the message (request) into peer inq */
status = event_add_out(ctx->evb, p_conn);
if (status != DN_OK) {
dnode_req_forward_error(ctx, p_conn, msg);
p_conn->err = errno;
return;
}
struct mbuf *header_buf = mbuf_get();
if (header_buf == NULL) {
loga("Unable to obtain an mbuf for dnode msg's header!");
req_put(msg);
return;
}
struct server_pool *pool = c_conn->owner;
dmsg_type_t msg_type = (string_compare(&pool->dc, dc) != 0)? DMSG_REQ_FORWARD : DMSG_REQ;
if (p_conn->dnode_secured) {
//Encrypting and adding header for a request
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "AES encryption key: %s\n", base64_encode(p_conn->aes_key, AES_KEYLEN));
}
//write dnode header
if (ENCRYPTION) {
status = dyn_aes_encrypt_msg(msg, p_conn->aes_key);
if (status == DN_ERROR) {
loga("OOM to obtain an mbuf for encryption!");
mbuf_put(header_buf);
req_put(msg);
return;
}
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "#encrypted bytes : %d", status);
}
dmsg_write(header_buf, msg->id, msg_type, p_conn, msg_length(msg));
} else {
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "no encryption on the msg payload");
}
dmsg_write(header_buf, msg->id, msg_type, p_conn, msg_length(msg));
}
} else {
//write dnode header
dmsg_write(header_buf, msg->id, msg_type, p_conn, msg_length(msg));
}
mbuf_insert_head(&msg->mhdr, header_buf);
if (log_loggable(LOG_VVERB)) {
log_hexdump(LOG_VVERB, header_buf->pos, mbuf_length(header_buf), "dyn message header: ");
msg_dump(msg);
}
conn_enqueue_inq(ctx, p_conn, msg);
dnode_peer_req_forward_stats(ctx, p_conn->owner, msg);
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, "remote forward from c %d to s %d req %"PRIu64" len %"PRIu32
" type %d with key '%.*s'", c_conn->sd, p_conn->sd, msg->id,
msg->mlen, msg->type, keylen, key);
}
}
void
local_req_forward(struct context *ctx, struct conn *c_conn, struct msg *msg,
uint8_t *key, uint32_t keylen)
{
rstatus_t status;
struct conn *s_conn;
if (log_loggable(LOG_VVERB)) {
loga("local_req_forward entering ............");
}
ASSERT((c_conn->type == CONN_CLIENT) ||
(c_conn->type == CONN_DNODE_PEER_CLIENT));
/* enqueue message (request) into client outq, if response is expected */
if (msg->expect_datastore_reply) {
conn_enqueue_outq(ctx, c_conn, msg);
}
s_conn = get_datastore_conn(ctx, c_conn->owner);
log_debug(LOG_VERB, "c_conn %p got server conn %p", c_conn, s_conn);
if (s_conn == NULL) {
req_forward_error(ctx, c_conn, msg, errno);
return;
}
ASSERT(s_conn->type == CONN_SERVER);
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "forwarding request from client conn '%s' to storage conn '%s'",
dn_unresolve_peer_desc(c_conn->sd), dn_unresolve_peer_desc(s_conn->sd));
}
if (ctx->dyn_state == NORMAL) {
/* enqueue the message (request) into server inq */
if (TAILQ_EMPTY(&s_conn->imsg_q)) {
status = event_add_out(ctx->evb, s_conn);
if (status != DN_OK) {
req_forward_error(ctx, c_conn, msg, errno);
s_conn->err = errno;
return;
}
}
} else if (ctx->dyn_state == STANDBY) { //no reads/writes from peers/clients
log_debug(LOG_INFO, "Node is in STANDBY state. Drop write/read requests");
req_forward_error(ctx, c_conn, msg, errno);
return;
} else if (ctx->dyn_state == WRITES_ONLY && msg->is_read) {
//no reads from peers/clients but allow writes from peers/clients
log_debug(LOG_INFO, "Node is in WRITES_ONLY state. Drop read requests");
req_forward_error(ctx, c_conn, msg, errno);
return;
} else if (ctx->dyn_state == RESUMING) {
log_debug(LOG_INFO, "Node is in RESUMING state. Still drop read requests and flush out all the queued writes");
if (msg->is_read) {
req_forward_error(ctx, c_conn, msg, errno);
return;
}
status = event_add_out(ctx->evb, s_conn);
if (status != DN_OK) {
req_forward_error(ctx, c_conn, msg, errno);
s_conn->err = errno;
return;
}
}
conn_enqueue_inq(ctx, s_conn, msg);
req_forward_stats(ctx, msg);
if(g_data_store == DATA_REDIS){
req_redis_stats(ctx, msg);
}
if (log_loggable(LOG_VERB)) {
log_debug(LOG_VERB, "local forward from c %d to s %d req %"PRIu64" len %"PRIu32
" type %d with key '%.*s'", c_conn->sd, s_conn->sd, msg->id,
msg->mlen, msg->type, keylen, key);
}
}
static void
req_forward(struct context *ctx, struct conn *c_conn, struct msg *msg)
{
struct server_pool *pool = c_conn->owner;
uint8_t *key;
uint32_t keylen;
ASSERT(c_conn->client && !c_conn->proxy);
if (msg->is_read)
stats_pool_incr(ctx, pool, client_read_requests);
else
stats_pool_incr(ctx, pool, client_write_requests);
key = NULL;
keylen = 0;
if (!string_empty(&pool->hash_tag)) {
struct string *tag = &pool->hash_tag;
uint8_t *tag_start, *tag_end;
tag_start = dn_strchr(msg->key_start, msg->key_end, tag->data[0]);
if (tag_start != NULL) {
tag_end = dn_strchr(tag_start + 1, msg->key_end, tag->data[1]);
if (tag_end != NULL) {
key = tag_start + 1;
keylen = (uint32_t)(tag_end - key);
}
}
}
if (keylen == 0) {
key = msg->key_start;
keylen = (uint32_t)(msg->key_end - msg->key_start);
}
// need to capture the initial mbuf location as once we add in the dynomite headers (as mbufs to the src msg),
// that will bork the request sent to secondary racks
struct mbuf *orig_mbuf = STAILQ_FIRST(&msg->mhdr);
if (request_send_to_all_racks(msg)) {
uint32_t dc_cnt = array_n(&pool->datacenters);
uint32_t dc_index;
for(dc_index = 0; dc_index < dc_cnt; dc_index++) {
struct datacenter *dc = array_get(&pool->datacenters, dc_index);
if (dc == NULL) {
log_error("Wow, this is very bad, dc is NULL");
return;
}
if (string_compare(dc->name, &pool->dc) == 0) { //send to all local racks
//log_debug(LOG_DEBUG, "dc name '%.*s'", dc->name->len, dc->name->data);
uint32_t rack_cnt = array_n(&dc->racks);
uint32_t rack_index;
for(rack_index = 0; rack_index < rack_cnt; rack_index++) {
struct rack *rack = array_get(&dc->racks, rack_index);
//log_debug(LOG_DEBUG, "rack name '%.*s'", rack->name->len, rack->name->data);
struct msg *rack_msg;
if (string_compare(rack->name, &pool->rack) == 0 ) {
rack_msg = msg;
} else {
rack_msg = msg_get(c_conn, msg->request, msg->redis);
if (rack_msg == NULL) {
log_debug(LOG_VERB, "whelp, looks like yer screwed now, buddy. no inter-rack messages for you!");
continue;
}
msg_clone(msg, orig_mbuf, rack_msg);
rack_msg->noreply = true;
}
log_debug(LOG_DEBUG, "forwarding request to conn '%s' on rack '%.*s'",
dn_unresolve_peer_desc(c_conn->sd), rack->name->len, rack->name->data);
remote_req_forward(ctx, c_conn, rack_msg, rack, key, keylen);
}
} else {
uint32_t rack_cnt = array_n(&dc->racks);
if (rack_cnt == 0)
continue;
uint32_t ran_index = rand() % rack_cnt;
struct rack *rack = array_get(&dc->racks, ran_index);
struct msg *rack_msg = msg_get(c_conn, msg->request, msg->redis);
if (rack_msg == NULL) {
log_debug(LOG_VERB, "whelp, looks like yer screwed now, buddy. no inter-rack messages for you!");
continue;
}
msg_clone(msg, orig_mbuf, rack_msg);
rack_msg->noreply = true;
log_debug(LOG_DEBUG, "forwarding request to conn '%s' on rack '%.*s'",
dn_unresolve_peer_desc(c_conn->sd), rack->name->len, rack->name->data);
remote_req_forward(ctx, c_conn, rack_msg, rack, key, keylen);
}
}
} else { //for read only requests
struct rack * rack = server_get_rack_by_dc_rack(pool, &pool->rack, &pool->dc);
remote_req_forward(ctx, c_conn, msg, rack, key, keylen);
}
}