3 * Author: Mingqiang Zhuang
5 * Created on February 10, 2009
7 * (c) Copyright 2009, Schooner Information Technology, Inc.
8 * http://www.schoonerinfotech.com/
20 #include <netinet/tcp.h>
21 #include <arpa/inet.h>
22 #if TIME_WITH_SYS_TIME
23 # include <sys/time.h>
27 # include <sys/time.h>
32 #include "ms_setting.h"
33 #include "ms_thread.h"
34 #include "ms_atomic.h"
37 /* /usr/include/netinet/in.h defines macros from ntohs() to _bswap_nn to
38 * optimize the conversion functions, but the prototypes generate warnings
39 * from gcc. The conversion methods isn't the bottleneck for my app, so
40 * just remove the warnings by undef'ing the optimization ..
48 /* for network write */
49 #define TRANSMIT_COMPLETE 0
50 #define TRANSMIT_INCOMPLETE 1
51 #define TRANSMIT_SOFT_ERROR 2
52 #define TRANSMIT_HARD_ERROR 3
54 /* for generating key */
55 #define KEY_PREFIX_BASE 0x1010101010101010 /* not include ' ' '\r' '\n' '\0' */
56 #define KEY_PREFIX_MASK 0x1010101010101010
58 /* For parse the value length return by server */
60 #define VALUELEN_TOKEN 3
62 /* global increasing counter, to ensure the key prefix unique */
63 static uint64_t key_prefix_seq
= KEY_PREFIX_BASE
;
65 /* global increasing counter, generating request id for UDP */
66 static volatile uint32_t udp_request_id
= 0;
68 extern pthread_key_t ms_thread_key
;
70 /* generate upd request id */
71 static uint32_t ms_get_udp_request_id(void);
74 /* connect initialize */
75 static void ms_task_init(ms_conn_t
*c
);
76 static int ms_conn_udp_init(ms_conn_t
*c
, const bool is_udp
);
77 static int ms_conn_sock_init(ms_conn_t
*c
);
78 static int ms_conn_event_init(ms_conn_t
*c
);
79 static int ms_conn_init(ms_conn_t
*c
,
81 const int read_buffer_size
,
83 static void ms_warmup_num_init(ms_conn_t
*c
);
84 static int ms_item_win_init(ms_conn_t
*c
);
87 /* connection close */
88 void ms_conn_free(ms_conn_t
*c
);
89 static void ms_conn_close(ms_conn_t
*c
);
92 /* create network connection */
93 static int ms_new_socket(struct addrinfo
*ai
);
94 static void ms_maximize_sndbuf(const int sfd
);
95 static int ms_network_connect(ms_conn_t
*c
,
100 static int ms_reconn(ms_conn_t
*c
);
104 static int ms_tokenize_command(char *command
,
106 const int max_tokens
);
107 static int ms_ascii_process_line(ms_conn_t
*c
, char *command
);
108 static int ms_try_read_line(ms_conn_t
*c
);
109 static int ms_sort_udp_packet(ms_conn_t
*c
, char *buf
, int rbytes
);
110 static int ms_udp_read(ms_conn_t
*c
, char *buf
, int len
);
111 static int ms_try_read_network(ms_conn_t
*c
);
112 static void ms_verify_value(ms_conn_t
*c
,
113 ms_mlget_task_item_t
*mlget_item
,
116 static void ms_ascii_complete_nread(ms_conn_t
*c
);
117 static void ms_bin_complete_nread(ms_conn_t
*c
);
118 static void ms_complete_nread(ms_conn_t
*c
);
122 static int ms_add_msghdr(ms_conn_t
*c
);
123 static int ms_ensure_iov_space(ms_conn_t
*c
);
124 static int ms_add_iov(ms_conn_t
*c
, const void *buf
, int len
);
125 static int ms_build_udp_headers(ms_conn_t
*c
);
126 static int ms_transmit(ms_conn_t
*c
);
129 /* status adjustment */
130 static void ms_conn_shrink(ms_conn_t
*c
);
131 static void ms_conn_set_state(ms_conn_t
*c
, int state
);
132 static bool ms_update_event(ms_conn_t
*c
, const int new_flags
);
133 static int ms_get_rep_sock_index(ms_conn_t
*c
, int cmd
);
134 static int ms_get_next_sock_index(ms_conn_t
*c
);
135 static int ms_update_conn_sock_event(ms_conn_t
*c
);
136 static bool ms_need_yield(ms_conn_t
*c
);
137 static void ms_update_start_time(ms_conn_t
*c
);
141 static void ms_drive_machine(ms_conn_t
*c
);
142 void ms_event_handler(const int fd
, const short which
, void *arg
);
146 static int ms_build_ascii_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
);
147 static int ms_build_ascii_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
);
148 static int ms_build_ascii_write_buf_mlget(ms_conn_t
*c
);
151 /* binary protocol */
152 static int ms_bin_process_response(ms_conn_t
*c
);
153 static void ms_add_bin_header(ms_conn_t
*c
,
158 static void ms_add_key_to_iov(ms_conn_t
*c
, ms_task_item_t
*item
);
159 static int ms_build_bin_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
);
160 static int ms_build_bin_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
);
161 static int ms_build_bin_write_buf_mlget(ms_conn_t
*c
);
165 * each key has two parts, prefix and suffix. The suffix is a
166 * string random get form the character table. The prefix is a
167 * uint64_t variable. And the prefix must be unique. we use the
168 * prefix to identify a key. And the prefix can't include
169 * character ' ' '\r' '\n' '\0'.
173 uint64_t ms_get_key_prefix(void)
177 pthread_mutex_lock(&ms_global
.seq_mutex
);
178 key_prefix_seq
|= KEY_PREFIX_MASK
;
179 key_prefix
= key_prefix_seq
;
181 pthread_mutex_unlock(&ms_global
.seq_mutex
);
184 } /* ms_get_key_prefix */
188 * get an unique udp request id
190 * @return an unique UDP request id
192 static uint32_t ms_get_udp_request_id(void)
194 return atomic_add_32_nv(&udp_request_id
, 1);
199 * initialize current task structure
201 * @param c, pointer of the concurrency
203 static void ms_task_init(ms_conn_t
*c
)
205 c
->curr_task
.cmd
= CMD_NULL
;
206 c
->curr_task
.item
= 0;
207 c
->curr_task
.verify
= false;
208 c
->curr_task
.finish_verify
= true;
209 c
->curr_task
.get_miss
= true;
211 c
->curr_task
.get_opt
= 0;
212 c
->curr_task
.set_opt
= 0;
213 c
->curr_task
.cycle_undo_get
= 0;
214 c
->curr_task
.cycle_undo_set
= 0;
215 c
->curr_task
.verified_get
= 0;
216 c
->curr_task
.overwrite_set
= 0;
221 * initialize udp for the connection structure
223 * @param c, pointer of the concurrency
224 * @param is_udp, whether it's udp
226 * @return int, if success, return 0, else return -1
228 static int ms_conn_udp_init(ms_conn_t
*c
, const bool is_udp
)
234 c
->rudpsize
= UDP_DATA_BUFFER_SIZE
;
245 if (c
->udp
|| (! c
->udp
&& ms_setting
.facebook_test
))
247 c
->rudpbuf
= (char *)malloc((size_t)c
->rudpsize
);
248 c
->udppkt
= (ms_udppkt_t
*)malloc(MAX_UDP_PACKET
* sizeof(ms_udppkt_t
));
250 if ((c
->rudpbuf
== NULL
) || (c
->udppkt
== NULL
))
252 if (c
->rudpbuf
!= NULL
)
254 if (c
->udppkt
!= NULL
)
256 fprintf(stderr
, "malloc()\n");
259 memset(c
->udppkt
, 0, MAX_UDP_PACKET
* sizeof(ms_udppkt_t
));
263 } /* ms_conn_udp_init */
267 * initialize the connection structure
269 * @param c, pointer of the concurrency
270 * @param init_state, (conn_read, conn_write, conn_closing)
271 * @param read_buffer_size
272 * @param is_udp, whether it's udp
274 * @return int, if success, return 0, else return -1
276 static int ms_conn_init(ms_conn_t
*c
,
277 const int init_state
,
278 const int read_buffer_size
,
287 c
->rsize
= read_buffer_size
;
288 c
->wsize
= WRITE_BUFFER_SIZE
;
289 c
->iovsize
= IOV_LIST_INITIAL
;
290 c
->msgsize
= MSG_LIST_INITIAL
;
292 /* for replication, each connection need connect all the server */
293 if (ms_setting
.rep_write_srv
> 0)
295 c
->total_sfds
= ms_setting
.srv_cnt
;
299 c
->total_sfds
= ms_setting
.sock_per_conn
;
303 c
->rbuf
= (char *)malloc((size_t)c
->rsize
);
304 c
->wbuf
= (char *)malloc((size_t)c
->wsize
);
305 c
->iov
= (struct iovec
*)malloc(sizeof(struct iovec
) * (size_t)c
->iovsize
);
306 c
->msglist
= (struct msghdr
*)malloc(
307 sizeof(struct msghdr
) * (size_t)c
->msgsize
);
308 if (ms_setting
.mult_key_num
> 1)
310 c
->mlget_task
.mlget_item
= (ms_mlget_task_item_t
*)
312 sizeof(ms_mlget_task_item_t
) * (size_t)ms_setting
.mult_key_num
);
314 c
->tcpsfd
= (int *)malloc((size_t)c
->total_sfds
* sizeof(int));
316 if ((c
->rbuf
== NULL
) || (c
->wbuf
== NULL
) || (c
->iov
== NULL
)
317 || (c
->msglist
== NULL
) || (c
->tcpsfd
== NULL
)
318 || ((ms_setting
.mult_key_num
> 1)
319 && (c
->mlget_task
.mlget_item
== NULL
)))
327 if (c
->msglist
!= NULL
)
329 if (c
->mlget_task
.mlget_item
!= NULL
)
330 free(c
->mlget_task
.mlget_item
);
331 if (c
->tcpsfd
!= NULL
)
333 fprintf(stderr
, "malloc()\n");
337 c
->state
= init_state
;
345 c
->cur_idx
= c
->total_sfds
; /* default index is a invalid value */
349 c
->change_sfd
= false;
351 c
->precmd
.cmd
= c
->currcmd
.cmd
= CMD_NULL
;
352 c
->precmd
.isfinish
= true; /* default the previous command finished */
353 c
->currcmd
.isfinish
= false;
354 c
->precmd
.retstat
= c
->currcmd
.retstat
= MCD_FAILURE
;
355 c
->precmd
.key_prefix
= c
->currcmd
.key_prefix
= 0;
357 c
->mlget_task
.mlget_num
= 0;
358 c
->mlget_task
.value_index
= -1; /* default invalid value */
360 if (ms_setting
.binary_prot
)
362 c
->protocol
= binary_prot
;
366 c
->protocol
= ascii_udp_prot
;
370 c
->protocol
= ascii_prot
;
374 if (ms_conn_udp_init(c
, is_udp
) != 0)
379 /* initialize task */
382 if (! (ms_setting
.facebook_test
&& is_udp
))
384 atomic_add_32(&ms_stats
.active_conns
, 1);
392 * when doing 100% get operation, it could preset some objects
393 * to warmup the server. this function is used to initialize the
394 * number of the objects to preset.
396 * @param c, pointer of the concurrency
398 static void ms_warmup_num_init(ms_conn_t
*c
)
400 /* no set operation, preset all the items in the window */
401 if (ms_setting
.cmd_distr
[CMD_SET
].cmd_prop
< PROP_ERROR
)
403 c
->warmup_num
= c
->win_size
;
404 c
->remain_warmup_num
= c
->warmup_num
;
409 c
->remain_warmup_num
= c
->warmup_num
;
411 } /* ms_warmup_num_init */
415 * each connection has an item window, this function initialize
416 * the window. The window is used to generate task.
418 * @param c, pointer of the concurrency
420 * @return int, if success, return 0, else return -1
422 static int ms_item_win_init(ms_conn_t
*c
)
424 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
427 c
->win_size
= (int)ms_setting
.win_size
;
429 c
->exec_num
= ms_thread
->thread_ctx
->exec_num_perconn
;
430 c
->remain_exec_num
= c
->exec_num
;
432 c
->item_win
= (ms_task_item_t
*)malloc(
433 sizeof(ms_task_item_t
) * (size_t)c
->win_size
);
434 if (c
->item_win
== NULL
)
436 fprintf(stderr
, "Can't allocate task item array for conn.\n");
439 memset(c
->item_win
, 0, sizeof(ms_task_item_t
) * (size_t)c
->win_size
);
441 for (int i
= 0; i
< c
->win_size
; i
++)
443 c
->item_win
[i
].key_size
= (int)ms_setting
.distr
[i
].key_size
;
444 c
->item_win
[i
].key_prefix
= ms_get_key_prefix();
445 c
->item_win
[i
].key_suffix_offset
= ms_setting
.distr
[i
].key_offset
;
446 c
->item_win
[i
].value_size
= (int)ms_setting
.distr
[i
].value_size
;
447 c
->item_win
[i
].value_offset
= INVALID_OFFSET
; /* default in invalid offset */
448 c
->item_win
[i
].client_time
= 0;
450 /* set expire time base on the proportion */
451 if (exp_cnt
< ms_setting
.exp_ver_per
* i
)
453 c
->item_win
[i
].exp_time
= FIXED_EXPIRE_TIME
;
458 c
->item_win
[i
].exp_time
= 0;
462 ms_warmup_num_init(c
);
465 } /* ms_item_win_init */
469 * each connection structure can include one or more sock
470 * handlers. this function create these socks and connect the
473 * @param c, pointer of the concurrency
475 * @return int, if success, return 0, else return -1
477 static int ms_conn_sock_init(ms_conn_t
*c
)
479 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
485 assert(c
->tcpsfd
!= NULL
);
487 for (i
= 0; i
< c
->total_sfds
; i
++)
490 if (ms_setting
.rep_write_srv
> 0)
492 /* for replication, each connection need connect all the server */
497 /* all the connections in a thread connects the same server */
498 srv_idx
= ms_thread
->thread_ctx
->srv_idx
;
501 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
502 ms_setting
.servers
[srv_idx
].srv_port
,
503 ms_setting
.udp
, &ret_sfd
) != 0)
513 if (! ms_setting
.udp
)
515 c
->tcpsfd
[i
]= ret_sfd
;
521 /* initialize udp sock handler if necessary */
522 if (ms_setting
.facebook_test
)
525 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
526 ms_setting
.servers
[srv_idx
].srv_port
,
527 true, &ret_sfd
) != 0)
537 if ((i
!= c
->total_sfds
) || (ms_setting
.facebook_test
&& (c
->udpsfd
== 0)))
545 for (int j
= 0; j
< i
; j
++)
560 } /* ms_conn_sock_init */
564 * each connection is managed by libevent, this function
565 * initialize the event of the connection structure.
567 * @param c, pointer of the concurrency
569 * @return int, if success, return 0, else return -1
571 static int ms_conn_event_init(ms_conn_t
*c
)
573 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
574 /* default event timeout 10 seconds */
577 .tv_sec
= EVENT_TIMEOUT
, .tv_usec
= 0
579 short event_flags
= EV_WRITE
| EV_PERSIST
;
581 event_set(&c
->event
, c
->sfd
, event_flags
, ms_event_handler
, (void *)c
);
582 event_base_set(ms_thread
->base
, &c
->event
);
583 c
->ev_flags
= event_flags
;
585 if (c
->total_sfds
== 1)
587 if (event_add(&c
->event
, NULL
) == -1)
594 if (event_add(&c
->event
, &t
) == -1)
601 } /* ms_conn_event_init */
605 * setup a connection, each connection structure of each
606 * thread must call this function to initialize.
608 * @param c, pointer of the concurrency
610 * @return int, if success, return 0, else return -1
612 int ms_setup_conn(ms_conn_t
*c
)
614 if (ms_item_win_init(c
) != 0)
619 if (ms_conn_init(c
, conn_write
, DATA_BUFFER_SIZE
, ms_setting
.udp
) != 0)
624 if (ms_conn_sock_init(c
) != 0)
629 if (ms_conn_event_init(c
) != 0)
635 } /* ms_setup_conn */
639 * Frees a connection.
641 * @param c, pointer of the concurrency
643 void ms_conn_free(ms_conn_t
*c
)
645 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
648 if (c
->hdrbuf
!= NULL
)
650 if (c
->msglist
!= NULL
)
658 if (c
->mlget_task
.mlget_item
!= NULL
)
659 free(c
->mlget_task
.mlget_item
);
660 if (c
->rudpbuf
!= NULL
)
662 if (c
->udppkt
!= NULL
)
664 if (c
->item_win
!= NULL
)
666 if (c
->tcpsfd
!= NULL
)
669 if (--ms_thread
->nactive_conn
== 0)
671 free(ms_thread
->conn
);
680 * @param c, pointer of the concurrency
682 static void ms_conn_close(ms_conn_t
*c
)
684 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
687 /* delete the event, the socket and the connection */
688 event_del(&c
->event
);
690 for (int i
= 0; i
< c
->total_sfds
; i
++)
692 if (c
->tcpsfd
[i
] > 0)
699 if (ms_setting
.facebook_test
)
704 atomic_dec_32(&ms_stats
.active_conns
);
708 if (ms_setting
.run_time
== 0)
710 pthread_mutex_lock(&ms_global
.run_lock
.lock
);
711 ms_global
.run_lock
.count
++;
712 pthread_cond_signal(&ms_global
.run_lock
.cond
);
713 pthread_mutex_unlock(&ms_global
.run_lock
.lock
);
716 if (ms_thread
->nactive_conn
== 0)
720 } /* ms_conn_close */
726 * @param ai, server address information
728 * @return int, if success, return 0, else return -1
730 static int ms_new_socket(struct addrinfo
*ai
)
734 if ((sfd
= socket(ai
->ai_family
, ai
->ai_socktype
, ai
->ai_protocol
)) == -1)
736 fprintf(stderr
, "socket() error: %s.\n", strerror(errno
));
741 } /* ms_new_socket */
745 * Sets a socket's send buffer size to the maximum allowed by the system.
747 * @param sfd, file descriptor of socket
749 static void ms_maximize_sndbuf(const int sfd
)
751 socklen_t intsize
= sizeof(int);
752 unsigned int last_good
= 0;
753 unsigned int min
, max
, avg
;
754 unsigned int old_size
;
756 /* Start with the default size. */
757 if (getsockopt(sfd
, SOL_SOCKET
, SO_SNDBUF
, &old_size
, &intsize
) != 0)
759 fprintf(stderr
, "getsockopt(SO_SNDBUF)\n");
763 /* Binary-search for the real maximum. */
765 max
= MAX_SENDBUF_SIZE
;
769 avg
= ((unsigned int)(min
+ max
)) / 2;
770 if (setsockopt(sfd
, SOL_SOCKET
, SO_SNDBUF
, (void *)&avg
, intsize
) == 0)
780 } /* ms_maximize_sndbuf */
784 * socket connects the server
786 * @param c, pointer of the concurrency
787 * @param srv_host_name, the host name of the server
788 * @param srv_port, port of server
789 * @param is_udp, whether it's udp
790 * @param ret_sfd, the connected socket file descriptor
792 * @return int, if success, return 0, else return -1
794 static int ms_network_connect(ms_conn_t
*c
,
806 struct addrinfo
*next
;
807 struct addrinfo hints
;
808 char port_buf
[NI_MAXSERV
];
815 * the memset call clears nonstandard fields in some impementations
816 * that otherwise mess things up.
818 memset(&hints
, 0, sizeof(hints
));
819 hints
.ai_flags
= AI_PASSIVE
| AI_ADDRCONFIG
;
822 hints
.ai_protocol
= IPPROTO_UDP
;
823 hints
.ai_socktype
= SOCK_DGRAM
;
824 hints
.ai_family
= AF_INET
; /* This left here because of issues with OSX 10.5 */
828 hints
.ai_family
= AF_UNSPEC
;
829 hints
.ai_protocol
= IPPROTO_TCP
;
830 hints
.ai_socktype
= SOCK_STREAM
;
833 snprintf(port_buf
, NI_MAXSERV
, "%d", srv_port
);
834 error
= getaddrinfo(srv_host_name
, port_buf
, &hints
, &ai
);
837 if (error
!= EAI_SYSTEM
)
838 fprintf(stderr
, "getaddrinfo(): %s.\n", gai_strerror(error
));
840 perror("getaddrinfo()\n");
845 for (next
= ai
; next
; next
= next
->ai_next
)
847 if ((sfd
= ms_new_socket(next
)) == -1)
853 setsockopt(sfd
, SOL_SOCKET
, SO_REUSEADDR
, (void *)&flags
, sizeof(flags
));
856 ms_maximize_sndbuf(sfd
);
860 setsockopt(sfd
, SOL_SOCKET
, SO_KEEPALIVE
, (void *)&flags
,
862 setsockopt(sfd
, SOL_SOCKET
, SO_LINGER
, (void *)&ling
, sizeof(ling
));
863 setsockopt(sfd
, IPPROTO_TCP
, TCP_NODELAY
, (void *)&flags
,
869 c
->srv_recv_addr_size
= sizeof(struct sockaddr
);
870 memcpy(&c
->srv_recv_addr
, next
->ai_addr
, c
->srv_recv_addr_size
);
874 if (connect(sfd
, next
->ai_addr
, next
->ai_addrlen
) == -1)
882 if (((flags
= fcntl(sfd
, F_GETFL
, 0)) < 0)
883 || (fcntl(sfd
, F_SETFL
, flags
| O_NONBLOCK
) < 0))
885 fprintf(stderr
, "setting O_NONBLOCK\n");
901 /* Return zero if we detected no errors in starting up connections */
903 } /* ms_network_connect */
907 * reconnect a disconnected sock
909 * @param c, pointer of the concurrency
911 * @return int, if success, return 0, else return -1
913 static int ms_reconn(ms_conn_t
*c
)
915 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
917 int32_t srv_conn_cnt
= 0;
919 if (ms_setting
.rep_write_srv
> 0)
922 srv_conn_cnt
= ms_setting
.nconns
;
926 srv_idx
= ms_thread
->thread_ctx
->srv_idx
;
927 srv_conn_cnt
= ms_setting
.nconns
/ ms_setting
.srv_cnt
;
930 /* close the old socket handler */
932 c
->tcpsfd
[c
->cur_idx
]= 0;
934 if (atomic_add_32_nv(&ms_setting
.servers
[srv_idx
].disconn_cnt
, 1)
935 % (uint32_t)srv_conn_cnt
== 0)
937 gettimeofday(&ms_setting
.servers
[srv_idx
].disconn_time
, NULL
);
938 fprintf(stderr
, "Server %s:%d disconnect\n",
939 ms_setting
.servers
[srv_idx
].srv_host_name
,
940 ms_setting
.servers
[srv_idx
].srv_port
);
943 if (ms_setting
.rep_write_srv
> 0)
946 for (i
= 0; i
< c
->total_sfds
; i
++)
948 if (c
->tcpsfd
[i
] != 0)
954 /* all socks disconnect */
955 if (i
== c
->total_sfds
)
964 /* reconnect success, break the loop */
965 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
966 ms_setting
.servers
[srv_idx
].srv_port
,
967 ms_setting
.udp
, &c
->sfd
) == 0)
969 c
->tcpsfd
[c
->cur_idx
]= c
->sfd
;
970 if (atomic_add_32_nv(&ms_setting
.servers
[srv_idx
].reconn_cnt
, 1)
971 % (uint32_t)srv_conn_cnt
== 0)
973 gettimeofday(&ms_setting
.servers
[srv_idx
].reconn_time
, NULL
);
975 (int)(ms_setting
.servers
[srv_idx
].reconn_time
.tv_sec
976 - ms_setting
.servers
[srv_idx
].disconn_time
978 fprintf(stderr
, "Server %s:%d reconnect after %ds\n",
979 ms_setting
.servers
[srv_idx
].srv_host_name
,
980 ms_setting
.servers
[srv_idx
].srv_port
, reconn_time
);
985 if (c
->total_sfds
== 1)
987 /* wait a second and reconnect */
991 while (c
->total_sfds
== 1);
994 if ((c
->total_sfds
> 1) && (c
->tcpsfd
[c
->cur_idx
] == 0))
1005 * reconnect several disconnected socks in the connection
1006 * structure, the ever-1-second timer of the thread will check
1007 * whether some socks in the connections disconnect. if
1008 * disconnect, reconnect the sock.
1010 * @param c, pointer of the concurrency
1012 * @return int, if success, return 0, else return -1
1014 int ms_reconn_socks(ms_conn_t
*c
)
1016 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
1019 int srv_conn_cnt
= 0;
1020 struct timeval cur_time
;
1024 if ((c
->total_sfds
== 1) || (c
->total_sfds
== c
->alive_sfds
))
1029 for (int i
= 0; i
< c
->total_sfds
; i
++)
1031 if (c
->tcpsfd
[i
] == 0)
1033 gettimeofday(&cur_time
, NULL
);
1036 * For failover test of replication, reconnect the socks after
1037 * it disconnects more than 5 seconds, Otherwise memslap will
1038 * block at connect() function and the work threads can't work
1042 - ms_setting
.servers
[srv_idx
].disconn_time
.tv_sec
< 5)
1047 if (ms_setting
.rep_write_srv
> 0)
1050 srv_conn_cnt
= ms_setting
.nconns
;
1054 srv_idx
= ms_thread
->thread_ctx
->srv_idx
;
1055 srv_conn_cnt
= ms_setting
.nconns
/ ms_setting
.srv_cnt
;
1058 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
1059 ms_setting
.servers
[srv_idx
].srv_port
,
1060 ms_setting
.udp
, &ret_sfd
) == 0)
1062 c
->tcpsfd
[i
]= ret_sfd
;
1065 if (atomic_add_32_nv(&ms_setting
.servers
[srv_idx
].reconn_cnt
, 1)
1066 % (uint32_t)srv_conn_cnt
== 0)
1068 gettimeofday(&ms_setting
.servers
[srv_idx
].reconn_time
, NULL
);
1070 (int)(ms_setting
.servers
[srv_idx
].reconn_time
.tv_sec
1071 - ms_setting
.servers
[srv_idx
].disconn_time
1073 fprintf(stderr
, "Server %s:%d reconnect after %ds\n",
1074 ms_setting
.servers
[srv_idx
].srv_host_name
,
1075 ms_setting
.servers
[srv_idx
].srv_port
, reconn_time
);
1082 } /* ms_reconn_socks */
1086 * Tokenize the command string by replacing whitespace with '\0' and update
1087 * the token array tokens with pointer to start of each token and length.
1088 * Returns total number of tokens. The last valid token is the terminal
1089 * token (value points to the first unprocessed character of the string and
1094 * while(ms_tokenize_command(command, ncommand, tokens, max_tokens) > 0) {
1095 * for(int ix = 0; tokens[ix].length != 0; ix++) {
1098 * ncommand = tokens[ix].value - command;
1099 * command = tokens[ix].value;
1102 * @param command, the command string to token
1103 * @param tokens, array to store tokens
1104 * @param max_tokens, maximum tokens number
1106 * @return int, the number of tokens
1108 static int ms_tokenize_command(char *command
,
1110 const int max_tokens
)
1115 assert(command
!= NULL
&& tokens
!= NULL
&& max_tokens
> 1);
1117 for (s
= e
= command
; ntokens
< max_tokens
- 1; ++e
)
1123 tokens
[ntokens
].value
= s
;
1124 tokens
[ntokens
].length
= (size_t)(e
- s
);
1130 else if (*e
== '\0')
1134 tokens
[ntokens
].value
= s
;
1135 tokens
[ntokens
].length
= (size_t)(e
- s
);
1139 break; /* string end */
1144 } /* ms_tokenize_command */
1148 * parse the response of server.
1150 * @param c, pointer of the concurrency
1151 * @param command, the string responded by server
1153 * @return int, if the command completed return 0, else return
1156 static int ms_ascii_process_line(ms_conn_t
*c
, char *command
)
1160 char *buffer
= command
;
1165 * for command get, we store the returned value into local buffer
1166 * then continue in ms_complete_nread().
1171 case 'V': /* VALUE || VERSION */
1172 if (buffer
[1] == 'A') /* VALUE */
1174 token_t tokens
[MAX_TOKENS
];
1175 ms_tokenize_command(command
, tokens
, MAX_TOKENS
);
1176 value_len
= strtol(tokens
[VALUELEN_TOKEN
].value
, NULL
, 10);
1177 c
->currcmd
.key_prefix
= *(uint64_t *)tokens
[KEY_TOKEN
].value
;
1180 * We read the \r\n into the string since not doing so is more
1181 * cycles then the waster of memory to do so.
1183 * We are null terminating through, which will most likely make
1184 * some people lazy about using the return length.
1186 c
->rvbytes
= (int)(value_len
+ 2);
1194 c
->currcmd
.retstat
= MCD_SUCCESS
;
1196 case 'S': /* STORED STATS SERVER_ERROR */
1197 if (buffer
[2] == 'A') /* STORED STATS */
1199 c
->currcmd
.retstat
= MCD_STAT
;
1201 else if (buffer
[1] == 'E')
1204 printf("<%d %s\n", c
->sfd
, buffer
);
1206 c
->currcmd
.retstat
= MCD_SERVER_ERROR
;
1208 else if (buffer
[1] == 'T')
1211 c
->currcmd
.retstat
= MCD_STORED
;
1215 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1219 case 'D': /* DELETED DATA */
1220 if (buffer
[1] == 'E')
1222 c
->currcmd
.retstat
= MCD_DELETED
;
1226 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1231 case 'N': /* NOT_FOUND NOT_STORED*/
1232 if (buffer
[4] == 'F')
1234 c
->currcmd
.retstat
= MCD_NOTFOUND
;
1236 else if (buffer
[4] == 'S')
1238 printf("<%d %s\n", c
->sfd
, buffer
);
1239 c
->currcmd
.retstat
= MCD_NOTSTORED
;
1243 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1247 case 'E': /* PROTOCOL ERROR or END */
1248 if (buffer
[1] == 'N')
1251 c
->currcmd
.retstat
= MCD_END
;
1253 else if (buffer
[1] == 'R')
1255 printf("<%d ERROR\n", c
->sfd
);
1256 c
->currcmd
.retstat
= MCD_PROTOCOL_ERROR
;
1258 else if (buffer
[1] == 'X')
1260 c
->currcmd
.retstat
= MCD_DATA_EXISTS
;
1261 printf("<%d %s\n", c
->sfd
, buffer
);
1265 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1269 case 'C': /* CLIENT ERROR */
1270 printf("<%d %s\n", c
->sfd
, buffer
);
1271 c
->currcmd
.retstat
= MCD_CLIENT_ERROR
;
1275 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1280 } /* ms_ascii_process_line */
1284 * after one operation completes, reset the concurrency
1286 * @param c, pointer of the concurrency
1287 * @param timeout, whether it's timeout
1289 void ms_reset_conn(ms_conn_t
*c
, bool timeout
)
1295 if ((c
->packets
> 0) && (c
->packets
< MAX_UDP_PACKET
))
1297 memset(c
->udppkt
, 0, sizeof(ms_udppkt_t
) * (size_t)c
->packets
);
1306 c
->currcmd
.isfinish
= true;
1310 ms_conn_set_state(c
, conn_write
);
1311 memcpy(&c
->precmd
, &c
->currcmd
, sizeof(ms_cmdstat_t
)); /* replicate command state */
1315 ms_drive_machine(c
);
1317 } /* ms_reset_conn */
1321 * if we have a complete line in the buffer, process it.
1323 * @param c, pointer of the concurrency
1325 * @return int, if success, return 0, else return -1
1327 static int ms_try_read_line(ms_conn_t
*c
)
1329 if (c
->protocol
== binary_prot
)
1331 /* Do we have the complete packet header? */
1332 if ((uint64_t)c
->rbytes
< sizeof(c
->binary_header
))
1334 /* need more data! */
1340 if (((long)(c
->rcurr
)) % 8 != 0)
1342 /* must realign input buffer */
1343 memmove(c
->rbuf
, c
->rcurr
, c
->rbytes
);
1345 if (settings
.verbose
)
1347 fprintf(stderr
, "%d: Realign input buffer.\n", c
->sfd
);
1351 protocol_binary_response_header
*rsp
;
1352 rsp
= (protocol_binary_response_header
*)c
->rcurr
;
1354 c
->binary_header
= *rsp
;
1355 c
->binary_header
.response
.extlen
= rsp
->response
.extlen
;
1356 c
->binary_header
.response
.keylen
= ntohs(rsp
->response
.keylen
);
1357 c
->binary_header
.response
.bodylen
= ntohl(rsp
->response
.bodylen
);
1358 c
->binary_header
.response
.status
= ntohs(rsp
->response
.status
);
1360 if (c
->binary_header
.response
.magic
!= PROTOCOL_BINARY_RES
)
1362 fprintf(stderr
, "Invalid magic: %x\n",
1363 c
->binary_header
.response
.magic
);
1364 ms_conn_set_state(c
, conn_closing
);
1368 /* process this complete response */
1369 if (ms_bin_process_response(c
) == 0)
1371 /* current operation completed */
1372 ms_reset_conn(c
, false);
1377 c
->rbytes
-= (int32_t)sizeof(c
->binary_header
);
1378 c
->rcurr
+= sizeof(c
->binary_header
);
1387 assert(c
->rcurr
<= (c
->rbuf
+ c
->rsize
));
1392 el
= memchr(c
->rcurr
, '\n', (size_t)c
->rbytes
);
1397 if (((el
- c
->rcurr
) > 1) && (*(el
- 1) == '\r'))
1403 assert(cont
<= (c
->rcurr
+ c
->rbytes
));
1405 /* process this complete line */
1406 if (ms_ascii_process_line(c
, c
->rcurr
) == 0)
1408 /* current operation completed */
1409 ms_reset_conn(c
, false);
1414 /* current operation didn't complete */
1415 c
->rbytes
-= (int32_t)(cont
- c
->rcurr
);
1419 assert(c
->rcurr
<= (c
->rbuf
+ c
->rsize
));
1423 } /* ms_try_read_line */
1427 * because the packet of UDP can't ensure the order, the
1428 * function is used to sort the received udp packet.
1430 * @param c, pointer of the concurrency
1431 * @param buf, the buffer to store the ordered packages data
1432 * @param rbytes, the maximum capacity of the buffer
1434 * @return int, if success, return the copy bytes, else return
1437 static int ms_sort_udp_packet(ms_conn_t
*c
, char *buf
, int rbytes
)
1442 uint16_t seq_num
= 0;
1443 uint16_t packets
= 0;
1444 unsigned char *header
= NULL
;
1446 /* no enough data */
1448 assert(buf
!= NULL
);
1449 assert(c
->rudpbytes
>= UDP_HEADER_SIZE
);
1451 /* calculate received packets count */
1452 if (c
->rudpbytes
% UDP_MAX_PAYLOAD_SIZE
>= UDP_HEADER_SIZE
)
1454 /* the last packet has some data */
1455 c
->recvpkt
= c
->rudpbytes
/ UDP_MAX_PAYLOAD_SIZE
+ 1;
1459 c
->recvpkt
= c
->rudpbytes
/ UDP_MAX_PAYLOAD_SIZE
;
1462 /* get the total packets count if necessary */
1463 if (c
->packets
== 0)
1465 c
->packets
= HEADER_TO_PACKETS((unsigned char *)c
->rudpbuf
);
1468 /* build the ordered packet array */
1469 for (int i
= c
->pktcurr
; i
< c
->recvpkt
; i
++)
1471 header
= (unsigned char *)c
->rudpbuf
+ i
* UDP_MAX_PAYLOAD_SIZE
;
1472 req_id
= (uint16_t)HEADER_TO_REQID(header
);
1473 assert(req_id
== c
->request_id
% (1 << 16));
1475 packets
= (uint16_t)HEADER_TO_PACKETS(header
);
1476 assert(c
->packets
== HEADER_TO_PACKETS(header
));
1478 seq_num
= (uint16_t)HEADER_TO_SEQNUM(header
);
1479 c
->udppkt
[seq_num
].header
= header
;
1480 c
->udppkt
[seq_num
].data
= (char *)header
+ UDP_HEADER_SIZE
;
1482 if (i
== c
->recvpkt
- 1)
1484 /* last received packet */
1485 if (c
->rudpbytes
% UDP_MAX_PAYLOAD_SIZE
== 0)
1487 c
->udppkt
[seq_num
].rbytes
= UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
;
1492 c
->udppkt
[seq_num
].rbytes
= c
->rudpbytes
% UDP_MAX_PAYLOAD_SIZE
1498 c
->udppkt
[seq_num
].rbytes
= UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
;
1503 for (int i
= c
->ordcurr
; i
< c
->recvpkt
; i
++)
1505 /* there is some data to copy */
1506 if ((c
->udppkt
[i
].data
!= NULL
)
1507 && (c
->udppkt
[i
].copybytes
< c
->udppkt
[i
].rbytes
))
1509 header
= c
->udppkt
[i
].header
;
1510 len
= c
->udppkt
[i
].rbytes
- c
->udppkt
[i
].copybytes
;
1511 if (len
> rbytes
- wbytes
)
1513 len
= rbytes
- wbytes
;
1516 assert(len
<= rbytes
- wbytes
);
1517 assert(i
== HEADER_TO_SEQNUM(header
));
1519 memcpy(buf
+ wbytes
, c
->udppkt
[i
].data
+ c
->udppkt
[i
].copybytes
,
1522 c
->udppkt
[i
].copybytes
+= len
;
1524 if ((c
->udppkt
[i
].copybytes
== c
->udppkt
[i
].rbytes
)
1525 && (c
->udppkt
[i
].rbytes
== UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
))
1527 /* finish copying all the data of this packet, next */
1531 /* last received packet, and finish copying all the data */
1532 if ((c
->recvpkt
== c
->packets
) && (i
== c
->recvpkt
- 1)
1533 && (c
->udppkt
[i
].copybytes
== c
->udppkt
[i
].rbytes
))
1538 /* no space to copy data */
1539 if (wbytes
>= rbytes
)
1544 /* it doesn't finish reading all the data of the packet from network */
1545 if ((i
!= c
->recvpkt
- 1)
1546 && (c
->udppkt
[i
].rbytes
< UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
))
1553 /* no data to copy */
1558 return wbytes
== 0 ? -1 : wbytes
;
1559 } /* ms_sort_udp_packet */
1563 * encapsulate upd read like tcp read
1565 * @param c, pointer of the concurrency
1566 * @param buf, read buffer
1567 * @param len, length to read
1569 * @return int, if success, return the read bytes, else return
1572 static int ms_udp_read(ms_conn_t
*c
, char *buf
, int len
)
1583 if (c
->rudpbytes
+ UDP_MAX_PAYLOAD_SIZE
> c
->rudpsize
)
1585 char *new_rbuf
= realloc(c
->rudpbuf
, (size_t)c
->rudpsize
* 2);
1588 fprintf(stderr
, "Couldn't realloc input buffer.\n");
1589 c
->rudpbytes
= 0; /* ignore what we read */
1592 c
->rudpbuf
= new_rbuf
;
1596 avail
= c
->rudpsize
- c
->rudpbytes
;
1597 /* UDP each time read a packet, 1400 bytes */
1598 res
= (int)read(c
->sfd
, c
->rudpbuf
+ c
->rudpbytes
, (size_t)avail
);
1602 atomic_add_size(&ms_stats
.bytes_read
, res
);
1617 /* "connection" closed */
1623 /* no data to read */
1628 /* copy data to read buffer */
1631 copybytes
= ms_sort_udp_packet(c
, buf
, len
);
1634 if (copybytes
== -1)
1636 atomic_add_size(&ms_stats
.pkt_disorder
, 1);
1644 * read from network as much as we can, handle buffer overflow and connection
1646 * before reading, move the remaining incomplete fragment of a command
1647 * (if any) to the beginning of the buffer.
1648 * return 0 if there's nothing to read on the first read.
1652 * read from network as much as we can, handle buffer overflow and connection
1653 * close. before reading, move the remaining incomplete fragment of a command
1654 * (if any) to the beginning of the buffer.
1656 * @param c, pointer of the concurrency
1659 * return 0 if there's nothing to read on the first read.
1660 * return 1 if get data
1661 * return -1 if error happens
1663 static int ms_try_read_network(ms_conn_t
*c
)
1671 if ((c
->rcurr
!= c
->rbuf
)
1672 && (! c
->readval
|| (c
->rvbytes
> c
->rsize
- (c
->rcurr
- c
->rbuf
))
1673 || (c
->readval
&& (c
->rcurr
- c
->rbuf
> c
->rbytes
))))
1675 if (c
->rbytes
!= 0) /* otherwise there's nothing to copy */
1676 memmove(c
->rbuf
, c
->rcurr
, (size_t)c
->rbytes
);
1682 if (c
->rbytes
>= c
->rsize
)
1684 char *new_rbuf
= realloc(c
->rbuf
, (size_t)c
->rsize
* 2);
1687 fprintf(stderr
, "Couldn't realloc input buffer.\n");
1688 c
->rbytes
= 0; /* ignore what we read */
1691 c
->rcurr
= c
->rbuf
= new_rbuf
;
1695 avail
= c
->rsize
- c
->rbytes
- (c
->rcurr
- c
->rbuf
);
1703 res
= (int32_t)ms_udp_read(c
, c
->rcurr
+ c
->rbytes
, (int32_t)avail
);
1707 res
= (int)read(c
->sfd
, c
->rcurr
+ c
->rbytes
, (size_t)avail
);
1714 atomic_add_size(&ms_stats
.bytes_read
, res
);
1729 /* connection closed */
1730 ms_conn_set_state(c
, conn_closing
);
1735 if ((errno
== EAGAIN
) || (errno
== EWOULDBLOCK
))
1737 /* Should close on unhandled errors. */
1738 ms_conn_set_state(c
, conn_closing
);
1744 } /* ms_try_read_network */
1748 * after get the object from server, verify the value if
1751 * @param c, pointer of the concurrency
1752 * @param mlget_item, pointer of mulit-get task item structure
1753 * @param value, received value string
1754 * @param vlen, received value string length
1756 static void ms_verify_value(ms_conn_t
*c
,
1757 ms_mlget_task_item_t
*mlget_item
,
1761 if (c
->curr_task
.verify
)
1763 assert(c
->curr_task
.item
->value_offset
!= INVALID_OFFSET
);
1764 char *orignval
= &ms_setting
.char_block
[c
->curr_task
.item
->value_offset
];
1766 &ms_setting
.char_block
[c
->curr_task
.item
->key_suffix_offset
];
1768 /* verify expire time if necessary */
1769 if (c
->curr_task
.item
->exp_time
> 0)
1771 struct timeval curr_time
;
1772 gettimeofday(&curr_time
, NULL
);
1774 /* object expired but get it now */
1775 if (curr_time
.tv_sec
- c
->curr_task
.item
->client_time
1776 > c
->curr_task
.item
->exp_time
+ EXPIRE_TIME_ERROR
)
1778 atomic_add_size(&ms_stats
.exp_get
, 1);
1780 if (ms_setting
.verbose
)
1784 strftime(set_time
, 64, "%Y-%m-%d %H:%M:%S",
1785 localtime(&c
->curr_task
.item
->client_time
));
1786 strftime(cur_time
, 64, "%Y-%m-%d %H:%M:%S",
1787 localtime(&curr_time
.tv_sec
));
1789 "\n<%d expire time verification failed, "
1790 "object expired but get it now\n"
1792 "\tkey: %" PRIx64
" %.*s\n"
1793 "\tset time: %s current time: %s "
1794 "diff time: %d expire time: %d\n"
1795 "\texpected data: \n"
1796 "\treceived data len: %d\n"
1797 "\treceived data: %.*s\n",
1799 c
->curr_task
.item
->key_size
,
1800 c
->curr_task
.item
->key_prefix
,
1801 c
->curr_task
.item
->key_size
- (int)KEY_PREFIX_SIZE
,
1805 (int)(curr_time
.tv_sec
- c
->curr_task
.item
->client_time
),
1806 c
->curr_task
.item
->exp_time
,
1816 if ((c
->curr_task
.item
->value_size
!= vlen
)
1817 || (memcmp(orignval
, value
, (size_t)vlen
) != 0))
1819 atomic_add_size(&ms_stats
.vef_failed
, 1);
1821 if (ms_setting
.verbose
)
1824 "\n<%d data verification failed\n"
1826 "\tkey: %" PRIx64
" %.*s\n"
1827 "\texpected data len: %d\n"
1828 "\texpected data: %.*s\n"
1829 "\treceived data len: %d\n"
1830 "\treceived data: %.*s\n",
1832 c
->curr_task
.item
->key_size
,
1833 c
->curr_task
.item
->key_prefix
,
1834 c
->curr_task
.item
->key_size
- (int)KEY_PREFIX_SIZE
,
1836 c
->curr_task
.item
->value_size
,
1837 c
->curr_task
.item
->value_size
,
1847 c
->curr_task
.finish_verify
= true;
1849 if (mlget_item
!= NULL
)
1851 mlget_item
->finish_verify
= true;
1854 } /* ms_verify_value */
1858 * For ASCII protocol, after store the data into the local
1859 * buffer, run this function to handle the data.
1861 * @param c, pointer of the concurrency
1863 static void ms_ascii_complete_nread(ms_conn_t
*c
)
1866 assert(c
->rbytes
>= c
->rvbytes
);
1867 assert(c
->protocol
== ascii_udp_prot
|| c
->protocol
== ascii_prot
);
1871 c
->rcurr
[c
->rvbytes
- 1] == '\n' && c
->rcurr
[c
->rvbytes
- 2] == '\r');
1875 ms_mlget_task_item_t
*mlget_item
= NULL
;
1876 if (((ms_setting
.mult_key_num
> 1)
1877 && (c
->mlget_task
.mlget_num
>= ms_setting
.mult_key_num
))
1878 || ((c
->remain_exec_num
== 0) && (c
->mlget_task
.mlget_num
> 0)))
1880 c
->mlget_task
.value_index
++;
1881 mlget_item
= &c
->mlget_task
.mlget_item
[c
->mlget_task
.value_index
];
1883 if (mlget_item
->item
->key_prefix
== c
->currcmd
.key_prefix
)
1885 c
->curr_task
.item
= mlget_item
->item
;
1886 c
->curr_task
.verify
= mlget_item
->verify
;
1887 c
->curr_task
.finish_verify
= mlget_item
->finish_verify
;
1888 mlget_item
->get_miss
= false;
1892 /* Try to find the task item in multi-get task array */
1893 for (int i
= 0; i
< c
->mlget_task
.mlget_num
; i
++)
1895 mlget_item
= &c
->mlget_task
.mlget_item
[i
];
1896 if (mlget_item
->item
->key_prefix
== c
->currcmd
.key_prefix
)
1898 c
->curr_task
.item
= mlget_item
->item
;
1899 c
->curr_task
.verify
= mlget_item
->verify
;
1900 c
->curr_task
.finish_verify
= mlget_item
->finish_verify
;
1901 mlget_item
->get_miss
= false;
1909 ms_verify_value(c
, mlget_item
, c
->rcurr
, c
->rvbytes
- 2);
1911 c
->curr_task
.get_miss
= false;
1912 c
->rbytes
-= c
->rvbytes
;
1913 c
->rcurr
= c
->rcurr
+ c
->rvbytes
;
1914 assert(c
->rcurr
<= (c
->rbuf
+ c
->rsize
));
1917 } /* ms_ascii_complete_nread */
1921 * For binary protocol, after store the data into the local
1922 * buffer, run this function to handle the data.
1924 * @param c, pointer of the concurrency
1926 static void ms_bin_complete_nread(ms_conn_t
*c
)
1929 assert(c
->rbytes
>= c
->rvbytes
);
1930 assert(c
->protocol
== binary_prot
);
1932 int extlen
= c
->binary_header
.response
.extlen
;
1933 int keylen
= c
->binary_header
.response
.keylen
;
1934 uint8_t opcode
= c
->binary_header
.response
.opcode
;
1936 /* not get command or not include value, just return */
1937 if (((opcode
!= PROTOCOL_BINARY_CMD_GET
)
1938 && (opcode
!= PROTOCOL_BINARY_CMD_GETQ
))
1939 || (c
->rvbytes
<= extlen
+ keylen
))
1942 if (c
->binary_header
.response
.opcode
== PROTOCOL_BINARY_CMD_GET
)
1944 c
->currcmd
.retstat
= MCD_END
;
1945 c
->curr_task
.get_miss
= true;
1950 ms_reset_conn(c
, false);
1955 ms_mlget_task_item_t
*mlget_item
= NULL
;
1956 if (((ms_setting
.mult_key_num
> 1)
1957 && (c
->mlget_task
.mlget_num
>= ms_setting
.mult_key_num
))
1958 || ((c
->remain_exec_num
== 0) && (c
->mlget_task
.mlget_num
> 0)))
1960 c
->mlget_task
.value_index
++;
1961 mlget_item
= &c
->mlget_task
.mlget_item
[c
->mlget_task
.value_index
];
1963 c
->curr_task
.item
= mlget_item
->item
;
1964 c
->curr_task
.verify
= mlget_item
->verify
;
1965 c
->curr_task
.finish_verify
= mlget_item
->finish_verify
;
1966 mlget_item
->get_miss
= false;
1971 c
->rcurr
+ extlen
+ keylen
,
1972 c
->rvbytes
- extlen
- keylen
);
1974 c
->currcmd
.retstat
= MCD_END
;
1975 c
->curr_task
.get_miss
= false;
1976 c
->rbytes
-= c
->rvbytes
;
1977 c
->rcurr
= c
->rcurr
+ c
->rvbytes
;
1978 assert(c
->rcurr
<= (c
->rbuf
+ c
->rsize
));
1982 if (ms_setting
.mult_key_num
> 1)
1984 /* multi-get have check all the item */
1985 if (c
->mlget_task
.value_index
== c
->mlget_task
.mlget_num
- 1)
1987 ms_reset_conn(c
, false);
1993 ms_reset_conn(c
, false);
1995 } /* ms_bin_complete_nread */
1999 * we get here after reading the value of get commands.
2001 * @param c, pointer of the concurrency
2003 static void ms_complete_nread(ms_conn_t
*c
)
2006 assert(c
->rbytes
>= c
->rvbytes
);
2007 assert(c
->protocol
== ascii_udp_prot
2008 || c
->protocol
== ascii_prot
2009 || c
->protocol
== binary_prot
);
2011 if (c
->protocol
== binary_prot
)
2013 ms_bin_complete_nread(c
);
2017 ms_ascii_complete_nread(c
);
2019 } /* ms_complete_nread */
2023 * Adds a message header to a connection.
2025 * @param c, pointer of the concurrency
2027 * @return int, if success, return 0, else return -1
2029 static int ms_add_msghdr(ms_conn_t
*c
)
2035 if (c
->msgsize
== c
->msgused
)
2038 realloc(c
->msglist
, (size_t)c
->msgsize
* 2 * sizeof(struct msghdr
));
2046 msg
= c
->msglist
+ c
->msgused
;
2049 * this wipes msg_iovlen, msg_control, msg_controllen, and
2050 * msg_flags, the last 3 of which aren't defined on solaris:
2052 memset(msg
, 0, sizeof(struct msghdr
));
2054 msg
->msg_iov
= &c
->iov
[c
->iovused
];
2056 if (c
->udp
&& (c
->srv_recv_addr_size
> 0))
2058 msg
->msg_name
= &c
->srv_recv_addr
;
2059 msg
->msg_namelen
= c
->srv_recv_addr_size
;
2067 /* Leave room for the UDP header, which we'll fill in later. */
2068 return ms_add_iov(c
, NULL
, UDP_HEADER_SIZE
);
2072 } /* ms_add_msghdr */
2076 * Ensures that there is room for another structure iovec in a connection's
2079 * @param c, pointer of the concurrency
2081 * @return int, if success, return 0, else return -1
2083 static int ms_ensure_iov_space(ms_conn_t
*c
)
2087 if (c
->iovused
>= c
->iovsize
)
2090 struct iovec
*new_iov
= (struct iovec
*)realloc(c
->iov
,
2093 * sizeof(struct iovec
));
2100 /* Point all the msghdr structures at the new list. */
2101 for (i
= 0, iovnum
= 0; i
< c
->msgused
; i
++)
2103 c
->msglist
[i
].msg_iov
= &c
->iov
[iovnum
];
2104 iovnum
+= (int)c
->msglist
[i
].msg_iovlen
;
2109 } /* ms_ensure_iov_space */
2113 * Adds data to the list of pending data that will be written out to a
2116 * @param c, pointer of the concurrency
2117 * @param buf, the buffer includes data to send
2118 * @param len, the data length in the buffer
2120 * @return int, if success, return 0, else return -1
2122 static int ms_add_iov(ms_conn_t
*c
, const void *buf
, int len
)
2132 m
= &c
->msglist
[c
->msgused
- 1];
2135 * Limit UDP packets, to UDP_MAX_PAYLOAD_SIZE bytes.
2137 limit_to_mtu
= c
->udp
;
2139 /* We may need to start a new msghdr if this one is full. */
2140 if ((m
->msg_iovlen
== IOV_MAX
)
2141 || (limit_to_mtu
&& (c
->msgbytes
>= UDP_MAX_SEND_PAYLOAD_SIZE
)))
2144 m
= &c
->msglist
[c
->msgused
- 1];
2147 if (ms_ensure_iov_space(c
) != 0)
2150 /* If the fragment is too big to fit in the datagram, split it up */
2151 if (limit_to_mtu
&& (len
+ c
->msgbytes
> UDP_MAX_SEND_PAYLOAD_SIZE
))
2153 leftover
= len
+ c
->msgbytes
- UDP_MAX_SEND_PAYLOAD_SIZE
;
2161 m
= &c
->msglist
[c
->msgused
- 1];
2162 m
->msg_iov
[m
->msg_iovlen
].iov_base
= (void *)buf
;
2163 m
->msg_iov
[m
->msg_iovlen
].iov_len
= (size_t)len
;
2169 buf
= ((char *)buf
) + len
;
2172 while (leftover
> 0);
2179 * Constructs a set of UDP headers and attaches them to the outgoing messages.
2181 * @param c, pointer of the concurrency
2183 * @return int, if success, return 0, else return -1
2185 static int ms_build_udp_headers(ms_conn_t
*c
)
2192 c
->request_id
= ms_get_udp_request_id();
2194 if (c
->msgused
> c
->hdrsize
)
2198 new_hdrbuf
= realloc(c
->hdrbuf
,
2199 (size_t)c
->msgused
* 2 * UDP_HEADER_SIZE
);
2201 new_hdrbuf
= malloc((size_t)c
->msgused
* 2 * UDP_HEADER_SIZE
);
2205 c
->hdrbuf
= (unsigned char *)new_hdrbuf
;
2206 c
->hdrsize
= c
->msgused
* 2;
2209 /* If this is a multi-packet request, drop it. */
2210 if (c
->udp
&& (c
->msgused
> 1))
2212 fprintf(stderr
, "multi-packet request for UDP not supported.\n");
2217 for (i
= 0; i
< c
->msgused
; i
++)
2219 c
->msglist
[i
].msg_iov
[0].iov_base
= (void *)hdr
;
2220 c
->msglist
[i
].msg_iov
[0].iov_len
= UDP_HEADER_SIZE
;
2221 *hdr
++= (unsigned char)(c
->request_id
/ 256);
2222 *hdr
++= (unsigned char)(c
->request_id
% 256);
2223 *hdr
++= (unsigned char)(i
/ 256);
2224 *hdr
++= (unsigned char)(i
% 256);
2225 *hdr
++= (unsigned char)(c
->msgused
/ 256);
2226 *hdr
++= (unsigned char)(c
->msgused
% 256);
2227 *hdr
++= (unsigned char)1; /* support facebook memcached */
2228 *hdr
++= (unsigned char)0;
2230 ((unsigned char *)c
->msglist
[i
].msg_iov
[0].iov_base
2231 + UDP_HEADER_SIZE
));
2235 } /* ms_build_udp_headers */
2239 * Transmit the next chunk of data from our list of msgbuf structures.
2241 * @param c, pointer of the concurrency
2243 * @return TRANSMIT_COMPLETE All done writing.
2244 * TRANSMIT_INCOMPLETE More data remaining to write.
2245 * TRANSMIT_SOFT_ERROR Can't write any more right now.
2246 * TRANSMIT_HARD_ERROR Can't write (c->state is set to conn_closing)
2248 static int ms_transmit(ms_conn_t
*c
)
2252 if ((c
->msgcurr
< c
->msgused
)
2253 && (c
->msglist
[c
->msgcurr
].msg_iovlen
== 0))
2255 /* Finished writing the current msg; advance to the next. */
2259 if (c
->msgcurr
< c
->msgused
)
2262 struct msghdr
*m
= &c
->msglist
[c
->msgcurr
];
2264 res
= sendmsg(c
->sfd
, m
, 0);
2267 atomic_add_size(&ms_stats
.bytes_written
, res
);
2269 /* We've written some of the data. Remove the completed
2270 * iovec entries from the list of pending writes. */
2271 while (m
->msg_iovlen
> 0 && res
>= (ssize_t
)m
->msg_iov
->iov_len
)
2273 res
-= (ssize_t
)m
->msg_iov
->iov_len
;
2278 /* Might have written just part of the last iovec entry;
2279 * adjust it so the next write will do the rest. */
2282 m
->msg_iov
->iov_base
= (void *)((unsigned char *)m
->msg_iov
->iov_base
+ res
);
2283 m
->msg_iov
->iov_len
-= (size_t)res
;
2285 return TRANSMIT_INCOMPLETE
;
2287 if ((res
== -1) && ((errno
== EAGAIN
) || (errno
== EWOULDBLOCK
)))
2289 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2291 fprintf(stderr
, "Couldn't update event.\n");
2292 ms_conn_set_state(c
, conn_closing
);
2293 return TRANSMIT_HARD_ERROR
;
2295 return TRANSMIT_SOFT_ERROR
;
2298 /* if res==0 or res==-1 and error is not EAGAIN or EWOULDBLOCK,
2299 * we have a real error, on which we close the connection */
2300 fprintf(stderr
, "Failed to write, and not due to blocking.\n");
2302 ms_conn_set_state(c
, conn_closing
);
2303 return TRANSMIT_HARD_ERROR
;
2307 return TRANSMIT_COMPLETE
;
2313 * Shrinks a connection's buffers if they're too big. This prevents
2314 * periodic large "mget" response from server chewing lots of client
2317 * This should only be called in between requests since it can wipe output
2320 * @param c, pointer of the concurrency
2322 static void ms_conn_shrink(ms_conn_t
*c
)
2329 if ((c
->rsize
> READ_BUFFER_HIGHWAT
) && (c
->rbytes
< DATA_BUFFER_SIZE
))
2333 if (c
->rcurr
!= c
->rbuf
)
2334 memmove(c
->rbuf
, c
->rcurr
, (size_t)c
->rbytes
);
2336 newbuf
= (char *)realloc((void *)c
->rbuf
, DATA_BUFFER_SIZE
);
2341 c
->rsize
= DATA_BUFFER_SIZE
;
2346 if (c
->udp
&& (c
->rudpsize
> UDP_DATA_BUFFER_HIGHWAT
)
2347 && (c
->rudpbytes
+ UDP_MAX_PAYLOAD_SIZE
< UDP_DATA_BUFFER_SIZE
))
2349 char *new_rbuf
= (char *)realloc(c
->rudpbuf
, (size_t)c
->rudpsize
* 2);
2352 c
->rudpbuf
= new_rbuf
;
2353 c
->rudpsize
= UDP_DATA_BUFFER_SIZE
;
2355 /* TODO check error condition? */
2358 if (c
->msgsize
> MSG_LIST_HIGHWAT
)
2360 struct msghdr
*newbuf
= (struct msghdr
*)realloc(
2363 * sizeof(c
->msglist
[0]));
2367 c
->msgsize
= MSG_LIST_INITIAL
;
2369 /* TODO check error condition? */
2372 if (c
->iovsize
> IOV_LIST_HIGHWAT
)
2374 struct iovec
*newbuf
= (struct iovec
*)realloc((void *)c
->iov
,
2376 * sizeof(c
->iov
[0]));
2380 c
->iovsize
= IOV_LIST_INITIAL
;
2382 /* TODO check return value */
2384 } /* ms_conn_shrink */
2388 * Sets a connection's current state in the state machine. Any special
2389 * processing that needs to happen on certain state transitions can
2392 * @param c, pointer of the concurrency
2393 * @param state, connection state
2395 static void ms_conn_set_state(ms_conn_t
*c
, int state
)
2399 if (state
!= c
->state
)
2401 if (state
== conn_read
)
2407 } /* ms_conn_set_state */
2411 * update the event if socks change state. for example: when
2412 * change the listen scoket read event to sock write event, or
2413 * change socket handler, we could call this function.
2415 * @param c, pointer of the concurrency
2416 * @param new_flags, new event flags
2418 * @return bool, if success, return true, else return false
2420 static bool ms_update_event(ms_conn_t
*c
, const int new_flags
)
2422 /* default event timeout 10 seconds */
2425 .tv_sec
= EVENT_TIMEOUT
, .tv_usec
= 0
2430 struct event_base
*base
= c
->event
.ev_base
;
2431 if ((c
->ev_flags
== new_flags
) && (ms_setting
.rep_write_srv
== 0)
2432 && (! ms_setting
.facebook_test
|| (c
->total_sfds
== 1)))
2437 if (event_del(&c
->event
) == -1)
2439 /* try to delete the event again */
2440 if (event_del(&c
->event
) == -1)
2446 event_set(&c
->event
,
2451 event_base_set(base
, &c
->event
);
2452 c
->ev_flags
= (short)new_flags
;
2454 if (c
->total_sfds
== 1)
2456 if (event_add(&c
->event
, NULL
) == -1)
2463 if (event_add(&c
->event
, &t
) == -1)
2470 } /* ms_update_event */
2474 * If user want to get the expected throughput, we could limit
2475 * the performance of memslap. we could give up some work and
2476 * just wait a short time. The function is used to check this
2479 * @param c, pointer of the concurrency
2481 * @return bool, if success, return true, else return false
2483 static bool ms_need_yield(ms_conn_t
*c
)
2485 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
2487 int64_t time_diff
= 0;
2488 struct timeval curr_time
;
2489 ms_task_t
*task
= &c
->curr_task
;
2491 if (ms_setting
.expected_tps
> 0)
2493 gettimeofday(&curr_time
, NULL
);
2494 time_diff
= ms_time_diff(&ms_thread
->startup_time
, &curr_time
);
2496 (int64_t)((task
->get_opt
2497 + task
->set_opt
) / ((uint64_t)time_diff
/ 1000000));
2499 /* current throughput is greater than expected throughput */
2500 if (tps
> ms_thread
->thread_ctx
->tps_perconn
)
2507 } /* ms_need_yield */
2511 * used to update the start time of each operation
2513 * @param c, pointer of the concurrency
2515 static void ms_update_start_time(ms_conn_t
*c
)
2517 ms_task_item_t
*item
= c
->curr_task
.item
;
2519 if ((ms_setting
.stat_freq
> 0) || c
->udp
2520 || ((c
->currcmd
.cmd
== CMD_SET
) && (item
->exp_time
> 0)))
2522 gettimeofday(&c
->start_time
, NULL
);
2523 if ((c
->currcmd
.cmd
== CMD_SET
) && (item
->exp_time
> 0))
2525 /* record the current time */
2526 item
->client_time
= c
->start_time
.tv_sec
;
2529 } /* ms_update_start_time */
2533 * run the state machine
2535 * @param c, pointer of the concurrency
2537 static void ms_drive_machine(ms_conn_t
*c
)
2550 if (c
->rbytes
>= c
->rvbytes
)
2552 ms_complete_nread(c
);
2558 if (ms_try_read_line(c
) != 0)
2564 if (ms_try_read_network(c
) != 0)
2569 /* doesn't read all the response data, wait event wake up */
2570 if (! c
->currcmd
.isfinish
)
2572 if (! ms_update_event(c
, EV_READ
| EV_PERSIST
))
2574 fprintf(stderr
, "Couldn't update event.\n");
2575 ms_conn_set_state(c
, conn_closing
);
2582 /* we have no command line and no data to read from network, next write */
2583 ms_conn_set_state(c
, conn_write
);
2584 memcpy(&c
->precmd
, &c
->currcmd
, sizeof(ms_cmdstat_t
)); /* replicate command state */
2589 if (! c
->ctnwrite
&& ms_need_yield(c
))
2593 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2595 fprintf(stderr
, "Couldn't update event.\n");
2596 ms_conn_set_state(c
, conn_closing
);
2603 if (! c
->ctnwrite
&& (ms_exec_task(c
) != 0))
2605 ms_conn_set_state(c
, conn_closing
);
2609 /* record the start time before starting to send data if necessary */
2610 if (! c
->ctnwrite
|| (c
->change_sfd
&& c
->ctnwrite
))
2614 c
->change_sfd
= false;
2616 ms_update_start_time(c
);
2619 /* change sfd if necessary */
2627 /* execute task until nothing need be written to network */
2628 if (! c
->ctnwrite
&& (c
->msgcurr
== c
->msgused
))
2630 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2632 fprintf(stderr
, "Couldn't update event.\n");
2633 ms_conn_set_state(c
, conn_closing
);
2640 switch (ms_transmit(c
))
2642 case TRANSMIT_COMPLETE
:
2643 /* we have no data to write to network, next wait repose */
2644 if (! ms_update_event(c
, EV_READ
| EV_PERSIST
))
2646 fprintf(stderr
, "Couldn't update event.\n");
2647 ms_conn_set_state(c
, conn_closing
);
2651 ms_conn_set_state(c
, conn_read
);
2656 case TRANSMIT_INCOMPLETE
:
2658 break; /* Continue in state machine. */
2660 case TRANSMIT_HARD_ERROR
:
2664 case TRANSMIT_SOFT_ERROR
:
2676 /* recovery mode, need reconnect if connection close */
2677 if (ms_setting
.reconnect
&& (! ms_global
.time_out
2678 || ((ms_setting
.run_time
== 0)
2679 && (c
->remain_exec_num
> 0))))
2681 if (ms_reconn(c
) != 0)
2688 ms_reset_conn(c
, false);
2690 if (c
->total_sfds
== 1)
2692 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2694 fprintf(stderr
, "Couldn't update event.\n");
2695 ms_conn_set_state(c
, conn_closing
);
2713 } /* ms_drive_machine */
2717 * the event handler of each thread
2719 * @param fd, the file descriptor of socket
2720 * @param which, event flag
2721 * @param arg, argument
2723 void ms_event_handler(const int fd
, const short which
, void *arg
)
2725 ms_conn_t
*c
= (ms_conn_t
*)arg
;
2735 "Catastrophic: event fd: %d doesn't match conn fd: %d\n",
2741 assert(fd
== c
->sfd
);
2743 /* event timeout, close the current connection */
2744 if (c
->which
== EV_TIMEOUT
)
2746 ms_conn_set_state(c
, conn_closing
);
2749 ms_drive_machine(c
);
2751 /* wait for next event */
2752 } /* ms_event_handler */
2756 * get the next socket descriptor index to run for replication
2758 * @param c, pointer of the concurrency
2759 * @param cmd, command(get or set )
2761 * @return int, if success, return the index, else return 0
2763 static int ms_get_rep_sock_index(ms_conn_t
*c
, int cmd
)
2768 if (c
->total_sfds
== 1)
2773 if (ms_setting
.rep_write_srv
== 0)
2782 for (i
= 0; i
< ms_setting
.rep_write_srv
; i
++)
2784 if (c
->tcpsfd
[i
] > 0)
2790 if (i
== ms_setting
.rep_write_srv
)
2792 /* random get one replication server to read */
2793 sock_index
= (int)(random() % c
->total_sfds
);
2797 /* random get one replication writing server to write */
2798 sock_index
= (int)(random() % ms_setting
.rep_write_srv
);
2801 else if (cmd
== CMD_GET
)
2803 /* random get one replication server to read */
2804 sock_index
= (int)(random() % c
->total_sfds
);
2807 while (c
->tcpsfd
[sock_index
] == 0);
2810 } /* ms_get_rep_sock_index */
2814 * get the next socket descriptor index to run
2816 * @param c, pointer of the concurrency
2818 * @return int, return the index
2820 static int ms_get_next_sock_index(ms_conn_t
*c
)
2826 sock_index
= (++c
->cur_idx
== c
->total_sfds
) ? 0 : c
->cur_idx
;
2828 while (c
->tcpsfd
[sock_index
] == 0);
2831 } /* ms_get_next_sock_index */
2835 * update socket event of the connections
2837 * @param c, pointer of the concurrency
2839 * @return int, if success, return 0, else return -1
2841 static int ms_update_conn_sock_event(ms_conn_t
*c
)
2845 switch (c
->currcmd
.cmd
)
2848 if (ms_setting
.facebook_test
&& c
->udp
)
2850 c
->sfd
= c
->tcpsfd
[0];
2852 c
->change_sfd
= true;
2857 if (ms_setting
.facebook_test
&& ! c
->udp
)
2861 c
->change_sfd
= true;
2869 if (! c
->udp
&& (c
->total_sfds
> 1))
2871 if (c
->cur_idx
!= c
->total_sfds
)
2873 if (ms_setting
.rep_write_srv
== 0)
2875 c
->cur_idx
= ms_get_next_sock_index(c
);
2879 c
->cur_idx
= ms_get_rep_sock_index(c
, c
->currcmd
.cmd
);
2884 /* must select the first sock of the connection at the beginning */
2888 c
->sfd
= c
->tcpsfd
[c
->cur_idx
];
2889 assert(c
->sfd
!= 0);
2890 c
->change_sfd
= true;
2895 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2897 fprintf(stderr
, "Couldn't update event.\n");
2898 ms_conn_set_state(c
, conn_closing
);
2904 } /* ms_update_conn_sock_event */
2908 * for ASCII protocol, this function build the set command
2909 * string and send the command.
2911 * @param c, pointer of the concurrency
2912 * @param item, pointer of task item which includes the object
2915 * @return int, if success, return 0, else return -1
2917 static int ms_build_ascii_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
)
2921 char *buffer
= c
->wbuf
;
2923 write_len
= sprintf(buffer
,
2929 if (write_len
> c
->wsize
)
2931 /* ought to be always enough. just fail for simplicity */
2932 fprintf(stderr
, "output command line too long.\n");
2936 if (item
->value_offset
== INVALID_OFFSET
)
2938 value_offset
= item
->key_suffix_offset
;
2942 value_offset
= item
->value_offset
;
2945 if ((ms_add_iov(c
, "set ", 4) != 0)
2946 || (ms_add_iov(c
, (char *)&item
->key_prefix
,
2947 (int)KEY_PREFIX_SIZE
) != 0)
2948 || (ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
2949 item
->key_size
- (int)KEY_PREFIX_SIZE
) != 0)
2950 || (ms_add_iov(c
, buffer
, write_len
) != 0)
2951 || (ms_add_iov(c
, &ms_setting
.char_block
[value_offset
],
2952 item
->value_size
) != 0)
2953 || (ms_add_iov(c
, "\r\n", 2) != 0)
2954 || (c
->udp
&& (ms_build_udp_headers(c
) != 0)))
2960 } /* ms_build_ascii_write_buf_set */
2964 * used to send set command to server
2966 * @param c, pointer of the concurrency
2967 * @param item, pointer of task item which includes the object
2970 * @return int, if success, return 0, else return -1
2972 int ms_mcd_set(ms_conn_t
*c
, ms_task_item_t
*item
)
2976 c
->currcmd
.cmd
= CMD_SET
;
2977 c
->currcmd
.isfinish
= false;
2978 c
->currcmd
.retstat
= MCD_FAILURE
;
2980 if (ms_update_conn_sock_event(c
) != 0)
2988 if (ms_add_msghdr(c
) != 0)
2990 fprintf(stderr
, "Out of memory preparing request.");
2994 /* binary protocol */
2995 if (c
->protocol
== binary_prot
)
2997 if (ms_build_bin_write_buf_set(c
, item
) != 0)
3004 if (ms_build_ascii_write_buf_set(c
, item
) != 0)
3010 atomic_add_size(&ms_stats
.obj_bytes
,
3011 item
->key_size
+ item
->value_size
);
3012 atomic_add_size(&ms_stats
.cmd_set
, 1);
3019 * for ASCII protocol, this function build the get command
3020 * string and send the command.
3022 * @param c, pointer of the concurrency
3023 * @param item, pointer of task item which includes the object
3026 * @return int, if success, return 0, else return -1
3028 static int ms_build_ascii_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
)
3030 if ((ms_add_iov(c
, "get ", 4) != 0)
3031 || (ms_add_iov(c
, (char *)&item
->key_prefix
,
3032 (int)KEY_PREFIX_SIZE
) != 0)
3033 || (ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
3034 item
->key_size
- (int)KEY_PREFIX_SIZE
) != 0)
3035 || (ms_add_iov(c
, "\r\n", 2) != 0)
3036 || (c
->udp
&& (ms_build_udp_headers(c
) != 0)))
3042 } /* ms_build_ascii_write_buf_get */
3046 * used to send the get command to server
3048 * @param c, pointer of the concurrency
3049 * @param item, pointer of task item which includes the object
3051 * @param verify, whether do verification
3053 * @return int, if success, return 0, else return -1
3055 int ms_mcd_get(ms_conn_t
*c
, ms_task_item_t
*item
, bool verify
)
3057 /* verify not supported yet */
3058 UNUSED_ARGUMENT(verify
);
3062 c
->currcmd
.cmd
= CMD_GET
;
3063 c
->currcmd
.isfinish
= false;
3064 c
->currcmd
.retstat
= MCD_FAILURE
;
3066 if (ms_update_conn_sock_event(c
) != 0)
3074 if (ms_add_msghdr(c
) != 0)
3076 fprintf(stderr
, "Out of memory preparing request.");
3080 /* binary protocol */
3081 if (c
->protocol
== binary_prot
)
3083 if (ms_build_bin_write_buf_get(c
, item
) != 0)
3090 if (ms_build_ascii_write_buf_get(c
, item
) != 0)
3096 atomic_add_size(&ms_stats
.cmd_get
, 1);
3103 * for ASCII protocol, this function build the multi-get command
3104 * string and send the command.
3106 * @param c, pointer of the concurrency
3108 * @return int, if success, return 0, else return -1
3110 static int ms_build_ascii_write_buf_mlget(ms_conn_t
*c
)
3112 ms_task_item_t
*item
;
3114 if (ms_add_iov(c
, "get", 3) != 0)
3119 for (int i
= 0; i
< c
->mlget_task
.mlget_num
; i
++)
3121 item
= c
->mlget_task
.mlget_item
[i
].item
;
3122 assert(item
!= NULL
);
3123 if ((ms_add_iov(c
, " ", 1) != 0)
3124 || (ms_add_iov(c
, (char *)&item
->key_prefix
,
3125 (int)KEY_PREFIX_SIZE
) != 0)
3126 || (ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
3127 item
->key_size
- (int)KEY_PREFIX_SIZE
) != 0))
3133 if ((ms_add_iov(c
, "\r\n", 2) != 0)
3134 || (c
->udp
&& (ms_build_udp_headers(c
) != 0)))
3140 } /* ms_build_ascii_write_buf_mlget */
3144 * used to send the multi-get command to server
3146 * @param c, pointer of the concurrency
3148 * @return int, if success, return 0, else return -1
3150 int ms_mcd_mlget(ms_conn_t
*c
)
3152 ms_task_item_t
*item
;
3155 assert(c
->mlget_task
.mlget_num
>= 1);
3157 c
->currcmd
.cmd
= CMD_GET
;
3158 c
->currcmd
.isfinish
= false;
3159 c
->currcmd
.retstat
= MCD_FAILURE
;
3161 if (ms_update_conn_sock_event(c
) != 0)
3169 if (ms_add_msghdr(c
) != 0)
3171 fprintf(stderr
, "Out of memory preparing request.");
3175 /* binary protocol */
3176 if (c
->protocol
== binary_prot
)
3178 if (ms_build_bin_write_buf_mlget(c
) != 0)
3185 if (ms_build_ascii_write_buf_mlget(c
) != 0)
3191 /* decrease operation time of each item */
3192 for (int i
= 0; i
< c
->mlget_task
.mlget_num
; i
++)
3194 item
= c
->mlget_task
.mlget_item
[i
].item
;
3195 atomic_add_size(&ms_stats
.cmd_get
, 1);
3199 } /* ms_mcd_mlget */
3203 * binary protocol support
3207 * for binary protocol, parse the response of server
3209 * @param c, pointer of the concurrency
3211 * @return int, if success, return 0, else return -1
3213 static int ms_bin_process_response(ms_conn_t
*c
)
3215 const char *errstr
= NULL
;
3219 uint32_t bodylen
= c
->binary_header
.response
.bodylen
;
3220 uint8_t opcode
= c
->binary_header
.response
.opcode
;
3221 uint16_t status
= c
->binary_header
.response
.status
;
3225 c
->rvbytes
= (int32_t)bodylen
;
3233 case PROTOCOL_BINARY_RESPONSE_SUCCESS
:
3234 if (opcode
== PROTOCOL_BINARY_CMD_SET
)
3236 c
->currcmd
.retstat
= MCD_STORED
;
3238 else if (opcode
== PROTOCOL_BINARY_CMD_DELETE
)
3240 c
->currcmd
.retstat
= MCD_DELETED
;
3242 else if (opcode
== PROTOCOL_BINARY_CMD_GET
)
3244 c
->currcmd
.retstat
= MCD_END
;
3248 case PROTOCOL_BINARY_RESPONSE_ENOMEM
:
3249 errstr
= "Out of memory";
3250 c
->currcmd
.retstat
= MCD_SERVER_ERROR
;
3253 case PROTOCOL_BINARY_RESPONSE_UNKNOWN_COMMAND
:
3254 errstr
= "Unknown command";
3255 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
3258 case PROTOCOL_BINARY_RESPONSE_KEY_ENOENT
:
3259 errstr
= "Not found";
3260 c
->currcmd
.retstat
= MCD_NOTFOUND
;
3263 case PROTOCOL_BINARY_RESPONSE_EINVAL
:
3264 errstr
= "Invalid arguments";
3265 c
->currcmd
.retstat
= MCD_PROTOCOL_ERROR
;
3268 case PROTOCOL_BINARY_RESPONSE_KEY_EEXISTS
:
3269 errstr
= "Data exists for key.";
3272 case PROTOCOL_BINARY_RESPONSE_E2BIG
:
3273 errstr
= "Too large.";
3274 c
->currcmd
.retstat
= MCD_SERVER_ERROR
;
3277 case PROTOCOL_BINARY_RESPONSE_NOT_STORED
:
3278 errstr
= "Not stored.";
3279 c
->currcmd
.retstat
= MCD_NOTSTORED
;
3283 errstr
= "Unknown error";
3284 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
3290 fprintf(stderr
, "%s\n", errstr
);
3295 } /* ms_bin_process_response */
3298 /* build binary header and add the header to the buffer to send */
3301 * build binary header and add the header to the buffer to send
3303 * @param c, pointer of the concurrency
3304 * @param opcode, operation code
3305 * @param hdr_len, length of header
3306 * @param key_len, length of key
3307 * @param body_len. length of body
3309 static void ms_add_bin_header(ms_conn_t
*c
,
3315 protocol_binary_request_header
*header
;
3319 header
= (protocol_binary_request_header
*)c
->wcurr
;
3321 header
->request
.magic
= (uint8_t)PROTOCOL_BINARY_REQ
;
3322 header
->request
.opcode
= (uint8_t)opcode
;
3323 header
->request
.keylen
= htons(key_len
);
3325 header
->request
.extlen
= (uint8_t)hdr_len
;
3326 header
->request
.datatype
= (uint8_t)PROTOCOL_BINARY_RAW_BYTES
;
3327 header
->request
.reserved
= 0;
3329 header
->request
.bodylen
= htonl(body_len
);
3330 header
->request
.opaque
= 0;
3331 header
->request
.cas
= 0;
3333 ms_add_iov(c
, c
->wcurr
, sizeof(header
->request
));
3334 } /* ms_add_bin_header */
3338 * add the key to the socket write buffer array
3340 * @param c, pointer of the concurrency
3341 * @param item, pointer of task item which includes the object
3344 static void ms_add_key_to_iov(ms_conn_t
*c
, ms_task_item_t
*item
)
3346 ms_add_iov(c
, (char *)&item
->key_prefix
, (int)KEY_PREFIX_SIZE
);
3347 ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
3348 item
->key_size
- (int)KEY_PREFIX_SIZE
);
3353 * for binary protocol, this function build the set command
3354 * and add the command to send buffer array.
3356 * @param c, pointer of the concurrency
3357 * @param item, pointer of task item which includes the object
3360 * @return int, if success, return 0, else return -1
3362 static int ms_build_bin_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
)
3364 assert(c
->wbuf
== c
->wcurr
);
3367 protocol_binary_request_set
*rep
= (protocol_binary_request_set
*)c
->wcurr
;
3368 uint16_t keylen
= (uint16_t)item
->key_size
;
3369 uint32_t bodylen
= (uint32_t)sizeof(rep
->message
.body
)
3370 + (uint32_t)keylen
+ (uint32_t)item
->value_size
;
3372 ms_add_bin_header(c
,
3373 PROTOCOL_BINARY_CMD_SET
,
3374 sizeof(rep
->message
.body
),
3377 rep
->message
.body
.flags
= 0;
3378 rep
->message
.body
.expiration
= htonl((uint32_t)item
->exp_time
);
3379 ms_add_iov(c
, &rep
->message
.body
, sizeof(rep
->message
.body
));
3380 ms_add_key_to_iov(c
, item
);
3382 if (item
->value_offset
== INVALID_OFFSET
)
3384 value_offset
= item
->key_suffix_offset
;
3388 value_offset
= item
->value_offset
;
3390 ms_add_iov(c
, &ms_setting
.char_block
[value_offset
], item
->value_size
);
3393 } /* ms_build_bin_write_buf_set */
3397 * for binary protocol, this function build the get command and
3398 * add the command to send buffer array.
3400 * @param c, pointer of the concurrency
3401 * @param item, pointer of task item which includes the object
3404 * @return int, if success, return 0, else return -1
3406 static int ms_build_bin_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
)
3408 assert(c
->wbuf
== c
->wcurr
);
3410 ms_add_bin_header(c
, PROTOCOL_BINARY_CMD_GET
, 0, (uint16_t)item
->key_size
,
3411 (uint32_t)item
->key_size
);
3412 ms_add_key_to_iov(c
, item
);
3415 } /* ms_build_bin_write_buf_get */
3419 * for binary protocol, this function build the multi-get
3420 * command and add the command to send buffer array.
3422 * @param c, pointer of the concurrency
3423 * @param item, pointer of task item which includes the object
3426 * @return int, if success, return 0, else return -1
3428 static int ms_build_bin_write_buf_mlget(ms_conn_t
*c
)
3430 ms_task_item_t
*item
;
3432 assert(c
->wbuf
== c
->wcurr
);
3434 for (int i
= 0; i
< c
->mlget_task
.mlget_num
; i
++)
3436 item
= c
->mlget_task
.mlget_item
[i
].item
;
3437 assert(item
!= NULL
);
3439 ms_add_bin_header(c
,
3440 PROTOCOL_BINARY_CMD_GET
,
3442 (uint16_t)item
->key_size
,
3443 (uint32_t)item
->key_size
);
3444 ms_add_key_to_iov(c
, item
);
3445 c
->wcurr
+= sizeof(protocol_binary_request_get
);
3451 } /* ms_build_bin_write_buf_mlget */