3 * Author: Mingqiang Zhuang
5 * Created on February 10, 2009
7 * (c) Copyright 2009, Schooner Information Technology, Inc.
8 * http://www.schoonerinfotech.com/
12 #include "mem_config.h"
20 #include <netinet/tcp.h>
21 #include <netinet/in.h>
23 #if defined(HAVE_ARPA_INET_H)
24 # include <arpa/inet.h>
27 #if defined(HAVE_SYS_TIME_H)
28 # include <sys/time.h>
31 #if defined(HAVE_TIME_H)
35 #include "ms_setting.h"
36 #include "ms_thread.h"
37 #include "ms_atomic.h"
40 /* /usr/include/netinet/in.h defines macros from ntohs() to _bswap_nn to
41 * optimize the conversion functions, but the prototypes generate warnings
42 * from gcc. The conversion methods isn't the bottleneck for my app, so
43 * just remove the warnings by undef'ing the optimization ..
51 /* for network write */
52 #define TRANSMIT_COMPLETE 0
53 #define TRANSMIT_INCOMPLETE 1
54 #define TRANSMIT_SOFT_ERROR 2
55 #define TRANSMIT_HARD_ERROR 3
57 /* for generating key */
58 #define KEY_PREFIX_BASE 0x1010101010101010 /* not include ' ' '\r' '\n' '\0' */
59 #define KEY_PREFIX_MASK 0x1010101010101010
61 /* For parse the value length return by server */
63 #define VALUELEN_TOKEN 3
65 /* global increasing counter, to ensure the key prefix unique */
66 static uint64_t key_prefix_seq
= KEY_PREFIX_BASE
;
68 /* global increasing counter, generating request id for UDP */
69 static ATOMIC
uint32_t udp_request_id
= 0;
71 extern pthread_key_t ms_thread_key
;
73 /* generate upd request id */
74 static uint32_t ms_get_udp_request_id(void);
77 /* connect initialize */
78 static void ms_task_init(ms_conn_t
*c
);
79 static int ms_conn_udp_init(ms_conn_t
*c
, const bool is_udp
);
80 static int ms_conn_sock_init(ms_conn_t
*c
);
81 static int ms_conn_event_init(ms_conn_t
*c
);
82 static int ms_conn_init(ms_conn_t
*c
,
84 const int read_buffer_size
,
86 static void ms_warmup_num_init(ms_conn_t
*c
);
87 static int ms_item_win_init(ms_conn_t
*c
);
90 /* connection close */
91 void ms_conn_free(ms_conn_t
*c
);
92 static void ms_conn_close(ms_conn_t
*c
);
95 /* create network connection */
96 static int ms_new_socket(struct addrinfo
*ai
);
97 static void ms_maximize_sndbuf(const int sfd
);
98 static int ms_network_connect(ms_conn_t
*c
,
103 static int ms_reconn(ms_conn_t
*c
);
107 static int ms_tokenize_command(char *command
,
109 const int max_tokens
);
110 static int ms_ascii_process_line(ms_conn_t
*c
, char *command
);
111 static int ms_try_read_line(ms_conn_t
*c
);
112 static int ms_sort_udp_packet(ms_conn_t
*c
, char *buf
, int rbytes
);
113 static int ms_udp_read(ms_conn_t
*c
, char *buf
, int len
);
114 static int ms_try_read_network(ms_conn_t
*c
);
115 static void ms_verify_value(ms_conn_t
*c
,
116 ms_mlget_task_item_t
*mlget_item
,
119 static void ms_ascii_complete_nread(ms_conn_t
*c
);
120 static void ms_bin_complete_nread(ms_conn_t
*c
);
121 static void ms_complete_nread(ms_conn_t
*c
);
125 static int ms_add_msghdr(ms_conn_t
*c
);
126 static int ms_ensure_iov_space(ms_conn_t
*c
);
127 static int ms_add_iov(ms_conn_t
*c
, const void *buf
, int len
);
128 static int ms_build_udp_headers(ms_conn_t
*c
);
129 static int ms_transmit(ms_conn_t
*c
);
132 /* status adjustment */
133 static void ms_conn_shrink(ms_conn_t
*c
);
134 static void ms_conn_set_state(ms_conn_t
*c
, int state
);
135 static bool ms_update_event(ms_conn_t
*c
, const int new_flags
);
136 static uint32_t ms_get_rep_sock_index(ms_conn_t
*c
, int cmd
);
137 static uint32_t ms_get_next_sock_index(ms_conn_t
*c
);
138 static int ms_update_conn_sock_event(ms_conn_t
*c
);
139 static bool ms_need_yield(ms_conn_t
*c
);
140 static void ms_update_start_time(ms_conn_t
*c
);
144 static void ms_drive_machine(ms_conn_t
*c
);
145 void ms_event_handler(const int fd
, const short which
, void *arg
);
149 static int ms_build_ascii_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
);
150 static int ms_build_ascii_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
);
151 static int ms_build_ascii_write_buf_mlget(ms_conn_t
*c
);
154 /* binary protocol */
155 static int ms_bin_process_response(ms_conn_t
*c
);
156 static void ms_add_bin_header(ms_conn_t
*c
,
161 static void ms_add_key_to_iov(ms_conn_t
*c
, ms_task_item_t
*item
);
162 static int ms_build_bin_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
);
163 static int ms_build_bin_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
);
164 static int ms_build_bin_write_buf_mlget(ms_conn_t
*c
);
168 * each key has two parts, prefix and suffix. The suffix is a
169 * string random get form the character table. The prefix is a
170 * uint64_t variable. And the prefix must be unique. we use the
171 * prefix to identify a key. And the prefix can't include
172 * character ' ' '\r' '\n' '\0'.
176 uint64_t ms_get_key_prefix(void)
180 pthread_mutex_lock(&ms_global
.seq_mutex
);
181 key_prefix_seq
|= KEY_PREFIX_MASK
;
182 key_prefix
= key_prefix_seq
;
184 pthread_mutex_unlock(&ms_global
.seq_mutex
);
187 } /* ms_get_key_prefix */
191 * get an unique udp request id
193 * @return an unique UDP request id
195 static uint32_t ms_get_udp_request_id(void)
197 return atomic_add_32_nv(&udp_request_id
, 1);
202 * initialize current task structure
204 * @param c, pointer of the concurrency
206 static void ms_task_init(ms_conn_t
*c
)
208 c
->curr_task
.cmd
= CMD_NULL
;
209 c
->curr_task
.item
= 0;
210 c
->curr_task
.verify
= false;
211 c
->curr_task
.finish_verify
= true;
212 c
->curr_task
.get_miss
= true;
214 c
->curr_task
.get_opt
= 0;
215 c
->curr_task
.set_opt
= 0;
216 c
->curr_task
.cycle_undo_get
= 0;
217 c
->curr_task
.cycle_undo_set
= 0;
218 c
->curr_task
.verified_get
= 0;
219 c
->curr_task
.overwrite_set
= 0;
224 * initialize udp for the connection structure
226 * @param c, pointer of the concurrency
227 * @param is_udp, whether it's udp
229 * @return int, if success, return EXIT_SUCCESS, else return -1
231 static int ms_conn_udp_init(ms_conn_t
*c
, const bool is_udp
)
237 c
->rudpsize
= UDP_DATA_BUFFER_SIZE
;
248 if (c
->udp
|| (! c
->udp
&& ms_setting
.facebook_test
))
250 c
->rudpbuf
= (char *)malloc((size_t)c
->rudpsize
);
251 c
->udppkt
= (ms_udppkt_t
*)malloc(MAX_UDP_PACKET
* sizeof(ms_udppkt_t
));
253 if ((c
->rudpbuf
== NULL
) || (c
->udppkt
== NULL
))
255 if (c
->rudpbuf
!= NULL
)
257 if (c
->udppkt
!= NULL
)
259 fprintf(stderr
, "malloc()\n");
262 memset(c
->udppkt
, 0, MAX_UDP_PACKET
* sizeof(ms_udppkt_t
));
266 } /* ms_conn_udp_init */
270 * initialize the connection structure
272 * @param c, pointer of the concurrency
273 * @param init_state, (conn_read, conn_write, conn_closing)
274 * @param read_buffer_size
275 * @param is_udp, whether it's udp
277 * @return int, if success, return EXIT_SUCCESS, else return -1
279 static int ms_conn_init(ms_conn_t
*c
,
280 const int init_state
,
281 const int read_buffer_size
,
290 c
->rsize
= read_buffer_size
;
291 c
->wsize
= WRITE_BUFFER_SIZE
;
292 c
->iovsize
= IOV_LIST_INITIAL
;
293 c
->msgsize
= MSG_LIST_INITIAL
;
295 /* for replication, each connection need connect all the server */
296 if (ms_setting
.rep_write_srv
> 0)
298 c
->total_sfds
= ms_setting
.srv_cnt
* ms_setting
.sock_per_conn
;
302 c
->total_sfds
= ms_setting
.sock_per_conn
;
306 c
->rbuf
= (char *)malloc((size_t)c
->rsize
);
307 c
->wbuf
= (char *)malloc((size_t)c
->wsize
);
308 c
->iov
= (struct iovec
*)malloc(sizeof(struct iovec
) * (size_t)c
->iovsize
);
309 c
->msglist
= (struct msghdr
*)malloc(
310 sizeof(struct msghdr
) * (size_t)c
->msgsize
);
311 if (ms_setting
.mult_key_num
> 1)
313 c
->mlget_task
.mlget_item
= (ms_mlget_task_item_t
*)
315 sizeof(ms_mlget_task_item_t
) * (size_t)ms_setting
.mult_key_num
);
317 c
->tcpsfd
= (int *)malloc((size_t)c
->total_sfds
* sizeof(int));
319 if ((c
->rbuf
== NULL
) || (c
->wbuf
== NULL
) || (c
->iov
== NULL
)
320 || (c
->msglist
== NULL
) || (c
->tcpsfd
== NULL
)
321 || ((ms_setting
.mult_key_num
> 1)
322 && (c
->mlget_task
.mlget_item
== NULL
)))
330 if (c
->msglist
!= NULL
)
332 if (c
->mlget_task
.mlget_item
!= NULL
)
333 free(c
->mlget_task
.mlget_item
);
334 if (c
->tcpsfd
!= NULL
)
336 fprintf(stderr
, "malloc()\n");
340 c
->state
= init_state
;
348 c
->cur_idx
= c
->total_sfds
; /* default index is a invalid value */
352 c
->change_sfd
= false;
354 c
->precmd
.cmd
= c
->currcmd
.cmd
= CMD_NULL
;
355 c
->precmd
.isfinish
= true; /* default the previous command finished */
356 c
->currcmd
.isfinish
= false;
357 c
->precmd
.retstat
= c
->currcmd
.retstat
= MCD_FAILURE
;
358 c
->precmd
.key_prefix
= c
->currcmd
.key_prefix
= 0;
360 c
->mlget_task
.mlget_num
= 0;
361 c
->mlget_task
.value_index
= -1; /* default invalid value */
363 if (ms_setting
.binary_prot_
)
365 c
->protocol
= binary_prot
;
369 c
->protocol
= ascii_prot
;
373 if (ms_conn_udp_init(c
, is_udp
) != 0)
378 /* initialize task */
381 if (! (ms_setting
.facebook_test
&& is_udp
))
383 atomic_add_32(&ms_stats
.active_conns
, 1);
391 * when doing 100% get operation, it could preset some objects
392 * to warmup the server. this function is used to initialize the
393 * number of the objects to preset.
395 * @param c, pointer of the concurrency
397 static void ms_warmup_num_init(ms_conn_t
*c
)
399 /* no set operation, preset all the items in the window */
400 if (ms_setting
.cmd_distr
[CMD_SET
].cmd_prop
< PROP_ERROR
)
402 c
->warmup_num
= c
->win_size
;
403 c
->remain_warmup_num
= c
->warmup_num
;
408 c
->remain_warmup_num
= c
->warmup_num
;
410 } /* ms_warmup_num_init */
414 * each connection has an item window, this function initialize
415 * the window. The window is used to generate task.
417 * @param c, pointer of the concurrency
419 * @return int, if success, return EXIT_SUCCESS, else return -1
421 static int ms_item_win_init(ms_conn_t
*c
)
423 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
426 c
->win_size
= (int)ms_setting
.win_size
;
428 c
->exec_num
= ms_thread
->thread_ctx
->exec_num_perconn
;
429 c
->remain_exec_num
= c
->exec_num
;
431 c
->item_win
= (ms_task_item_t
*)malloc(
432 sizeof(ms_task_item_t
) * (size_t)c
->win_size
);
433 if (c
->item_win
== NULL
)
435 fprintf(stderr
, "Can't allocate task item array for conn.\n");
438 memset(c
->item_win
, 0, sizeof(ms_task_item_t
) * (size_t)c
->win_size
);
440 for (int i
= 0; i
< c
->win_size
; i
++)
442 c
->item_win
[i
].key_size
= (int)ms_setting
.distr
[i
].key_size
;
443 c
->item_win
[i
].key_prefix
= ms_get_key_prefix();
444 c
->item_win
[i
].key_suffix_offset
= ms_setting
.distr
[i
].key_offset
;
445 c
->item_win
[i
].value_size
= (int)ms_setting
.distr
[i
].value_size
;
446 c
->item_win
[i
].value_offset
= INVALID_OFFSET
; /* default in invalid offset */
447 c
->item_win
[i
].client_time
= 0;
449 /* set expire time base on the proportion */
450 if (exp_cnt
< ms_setting
.exp_ver_per
* i
)
452 c
->item_win
[i
].exp_time
= FIXED_EXPIRE_TIME
;
457 c
->item_win
[i
].exp_time
= 0;
461 ms_warmup_num_init(c
);
464 } /* ms_item_win_init */
468 * each connection structure can include one or more sock
469 * handlers. this function create these socks and connect the
472 * @param c, pointer of the concurrency
474 * @return int, if success, return EXIT_SUCCESS, else return -1
476 static int ms_conn_sock_init(ms_conn_t
*c
)
478 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
484 assert(c
->tcpsfd
!= NULL
);
486 for (i
= 0; i
< c
->total_sfds
; i
++)
489 if (ms_setting
.rep_write_srv
> 0)
491 /* for replication, each connection need connect all the server */
492 srv_idx
= i
% ms_setting
.srv_cnt
;
496 /* all the connections in a thread connects the same server */
497 srv_idx
= ms_thread
->thread_ctx
->srv_idx
;
500 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
501 ms_setting
.servers
[srv_idx
].srv_port
,
502 ms_setting
.udp
, &ret_sfd
) != 0)
512 if (! ms_setting
.udp
)
514 c
->tcpsfd
[i
]= ret_sfd
;
520 /* initialize udp sock handler if necessary */
521 if (ms_setting
.facebook_test
)
524 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
525 ms_setting
.servers
[srv_idx
].srv_port
,
526 true, &ret_sfd
) != 0)
536 if ((i
!= c
->total_sfds
) || (ms_setting
.facebook_test
&& (c
->udpsfd
== 0)))
544 for (uint32_t j
= 0; j
< i
; j
++)
559 } /* ms_conn_sock_init */
563 * each connection is managed by libevent, this function
564 * initialize the event of the connection structure.
566 * @param c, pointer of the concurrency
568 * @return int, if success, return EXIT_SUCCESS, else return -1
570 static int ms_conn_event_init(ms_conn_t
*c
)
572 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
573 short event_flags
= EV_WRITE
| EV_PERSIST
;
575 event_set(&c
->event
, c
->sfd
, event_flags
, ms_event_handler
, (void *)c
);
576 event_base_set(ms_thread
->base
, &c
->event
);
577 c
->ev_flags
= event_flags
;
579 if (event_add(&c
->event
, NULL
) == -1)
585 } /* ms_conn_event_init */
589 * setup a connection, each connection structure of each
590 * thread must call this function to initialize.
592 * @param c, pointer of the concurrency
594 * @return int, if success, return EXIT_SUCCESS, else return -1
596 int ms_setup_conn(ms_conn_t
*c
)
598 if (ms_item_win_init(c
) != 0)
603 if (ms_conn_init(c
, conn_write
, DATA_BUFFER_SIZE
, ms_setting
.udp
) != 0)
608 if (ms_conn_sock_init(c
) != 0)
613 if (ms_conn_event_init(c
) != 0)
619 } /* ms_setup_conn */
623 * Frees a connection.
625 * @param c, pointer of the concurrency
627 void ms_conn_free(ms_conn_t
*c
)
629 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
632 if (c
->hdrbuf
!= NULL
)
634 if (c
->msglist
!= NULL
)
642 if (c
->mlget_task
.mlget_item
!= NULL
)
643 free(c
->mlget_task
.mlget_item
);
644 if (c
->rudpbuf
!= NULL
)
646 if (c
->udppkt
!= NULL
)
648 if (c
->item_win
!= NULL
)
650 if (c
->tcpsfd
!= NULL
)
653 if (--ms_thread
->nactive_conn
== 0)
655 free(ms_thread
->conn
);
664 * @param c, pointer of the concurrency
666 static void ms_conn_close(ms_conn_t
*c
)
668 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
671 /* delete the event, the socket and the connection */
672 event_del(&c
->event
);
674 for (uint32_t i
= 0; i
< c
->total_sfds
; i
++)
676 if (c
->tcpsfd
[i
] > 0)
683 if (ms_setting
.facebook_test
)
688 atomic_dec_32(&ms_stats
.active_conns
);
692 if (ms_setting
.run_time
== 0)
694 pthread_mutex_lock(&ms_global
.run_lock
.lock
);
695 ms_global
.run_lock
.count
++;
696 pthread_cond_signal(&ms_global
.run_lock
.cond
);
697 pthread_mutex_unlock(&ms_global
.run_lock
.lock
);
700 if (ms_thread
->nactive_conn
== 0)
704 } /* ms_conn_close */
710 * @param ai, server address information
712 * @return int, if success, return EXIT_SUCCESS, else return -1
714 static int ms_new_socket(struct addrinfo
*ai
)
718 if ((sfd
= socket(ai
->ai_family
, ai
->ai_socktype
, ai
->ai_protocol
)) == -1)
720 fprintf(stderr
, "socket() error: %s.\n", strerror(errno
));
725 } /* ms_new_socket */
729 * Sets a socket's send buffer size to the maximum allowed by the system.
731 * @param sfd, file descriptor of socket
733 static void ms_maximize_sndbuf(const int sfd
)
735 socklen_t intsize
= sizeof(int);
736 unsigned int last_good
= 0;
737 unsigned int min
, max
, avg
;
738 unsigned int old_size
;
740 /* Start with the default size. */
741 if (getsockopt(sfd
, SOL_SOCKET
, SO_SNDBUF
, &old_size
, &intsize
) != 0)
743 fprintf(stderr
, "getsockopt(SO_SNDBUF)\n");
747 /* Binary-search for the real maximum. */
749 max
= MAX_SENDBUF_SIZE
;
753 avg
= ((unsigned int)(min
+ max
)) / 2;
754 if (setsockopt(sfd
, SOL_SOCKET
, SO_SNDBUF
, (void *)&avg
, intsize
) == 0)
765 } /* ms_maximize_sndbuf */
769 * socket connects the server
771 * @param c, pointer of the concurrency
772 * @param srv_host_name, the host name of the server
773 * @param srv_port, port of server
774 * @param is_udp, whether it's udp
775 * @param ret_sfd, the connected socket file descriptor
777 * @return int, if success, return EXIT_SUCCESS, else return -1
779 static int ms_network_connect(ms_conn_t
*c
,
791 struct addrinfo
*next
;
792 struct addrinfo hints
;
793 char port_buf
[NI_MAXSERV
];
800 * the memset call clears nonstandard fields in some impementations
801 * that otherwise mess things up.
803 memset(&hints
, 0, sizeof(hints
));
805 hints
.ai_flags
= AI_PASSIVE
| AI_ADDRCONFIG
;
807 hints
.ai_flags
= AI_PASSIVE
;
808 #endif /* AI_ADDRCONFIG */
811 hints
.ai_protocol
= IPPROTO_UDP
;
812 hints
.ai_socktype
= SOCK_DGRAM
;
813 hints
.ai_family
= AF_INET
; /* This left here because of issues with OSX 10.5 */
817 hints
.ai_family
= AF_UNSPEC
;
818 hints
.ai_protocol
= IPPROTO_TCP
;
819 hints
.ai_socktype
= SOCK_STREAM
;
822 snprintf(port_buf
, NI_MAXSERV
, "%d", srv_port
);
823 error
= getaddrinfo(srv_host_name
, port_buf
, &hints
, &ai
);
826 if (error
!= EAI_SYSTEM
)
827 fprintf(stderr
, "getaddrinfo(): %s.\n", gai_strerror(error
));
829 perror("getaddrinfo()\n");
834 for (next
= ai
; next
; next
= next
->ai_next
)
836 if ((sfd
= ms_new_socket(next
)) == -1)
842 setsockopt(sfd
, SOL_SOCKET
, SO_REUSEADDR
, (void *)&flags
, sizeof(flags
));
845 ms_maximize_sndbuf(sfd
);
849 setsockopt(sfd
, SOL_SOCKET
, SO_KEEPALIVE
, (void *)&flags
,
851 setsockopt(sfd
, SOL_SOCKET
, SO_LINGER
, (void *)&ling
, sizeof(ling
));
852 setsockopt(sfd
, IPPROTO_TCP
, TCP_NODELAY
, (void *)&flags
,
858 c
->srv_recv_addr_size
= sizeof(struct sockaddr
);
859 memcpy(&c
->srv_recv_addr
, next
->ai_addr
, c
->srv_recv_addr_size
);
863 if (connect(sfd
, next
->ai_addr
, next
->ai_addrlen
) == -1)
871 if (((flags
= fcntl(sfd
, F_GETFL
, 0)) < 0)
872 || (fcntl(sfd
, F_SETFL
, flags
| O_NONBLOCK
) < 0))
874 fprintf(stderr
, "setting O_NONBLOCK\n");
890 /* Return zero if we detected no errors in starting up connections */
892 } /* ms_network_connect */
896 * reconnect a disconnected sock
898 * @param c, pointer of the concurrency
900 * @return int, if success, return EXIT_SUCCESS, else return -1
902 static int ms_reconn(ms_conn_t
*c
)
904 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
906 uint32_t srv_conn_cnt
= 0;
908 if (ms_setting
.rep_write_srv
> 0)
910 srv_idx
= c
->cur_idx
% ms_setting
.srv_cnt
;
911 srv_conn_cnt
= ms_setting
.sock_per_conn
* ms_setting
.nconns
;
915 srv_idx
= ms_thread
->thread_ctx
->srv_idx
;
916 srv_conn_cnt
= ms_setting
.nconns
/ ms_setting
.srv_cnt
;
919 /* close the old socket handler */
921 c
->tcpsfd
[c
->cur_idx
]= 0;
923 if (atomic_add_32_nv(&ms_setting
.servers
[srv_idx
].disconn_cnt
, 1)
926 gettimeofday(&ms_setting
.servers
[srv_idx
].disconn_time
, NULL
);
927 fprintf(stderr
, "Server %s:%d disconnect\n",
928 ms_setting
.servers
[srv_idx
].srv_host_name
,
929 ms_setting
.servers
[srv_idx
].srv_port
);
932 if (ms_setting
.rep_write_srv
> 0)
936 for (i
= 0; i
< c
->total_sfds
; i
++)
938 if (c
->tcpsfd
[i
] != 0)
944 /* all socks disconnect */
945 if (i
== c
->total_sfds
)
954 /* reconnect success, break the loop */
955 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
956 ms_setting
.servers
[srv_idx
].srv_port
,
957 ms_setting
.udp
, &c
->sfd
) == 0)
959 c
->tcpsfd
[c
->cur_idx
]= c
->sfd
;
960 if (atomic_add_32_nv(&ms_setting
.servers
[srv_idx
].reconn_cnt
, 1)
961 % (uint32_t)srv_conn_cnt
== 0)
963 gettimeofday(&ms_setting
.servers
[srv_idx
].reconn_time
, NULL
);
965 (int)(ms_setting
.servers
[srv_idx
].reconn_time
.tv_sec
966 - ms_setting
.servers
[srv_idx
].disconn_time
968 fprintf(stderr
, "Server %s:%d reconnect after %ds\n",
969 ms_setting
.servers
[srv_idx
].srv_host_name
,
970 ms_setting
.servers
[srv_idx
].srv_port
, reconn_time
);
975 if (ms_setting
.rep_write_srv
== 0 && c
->total_sfds
> 0)
977 /* wait a second and reconnect */
981 while (ms_setting
.rep_write_srv
== 0 && c
->total_sfds
> 0);
984 if ((c
->total_sfds
> 1) && (c
->tcpsfd
[c
->cur_idx
] == 0))
995 * reconnect several disconnected socks in the connection
996 * structure, the ever-1-second timer of the thread will check
997 * whether some socks in the connections disconnect. if
998 * disconnect, reconnect the sock.
1000 * @param c, pointer of the concurrency
1002 * @return int, if success, return EXIT_SUCCESS, else return -1
1004 int ms_reconn_socks(ms_conn_t
*c
)
1006 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
1007 uint32_t srv_idx
= 0;
1009 uint32_t srv_conn_cnt
= 0;
1010 struct timeval cur_time
;
1014 if ((c
->total_sfds
== 1) || (c
->total_sfds
== c
->alive_sfds
))
1016 return EXIT_SUCCESS
;
1019 for (uint32_t i
= 0; i
< c
->total_sfds
; i
++)
1021 if (c
->tcpsfd
[i
] == 0)
1023 gettimeofday(&cur_time
, NULL
);
1026 * For failover test of replication, reconnect the socks after
1027 * it disconnects more than 5 seconds, Otherwise memslap will
1028 * block at connect() function and the work threads can't work
1032 - ms_setting
.servers
[srv_idx
].disconn_time
.tv_sec
< 5)
1037 if (ms_setting
.rep_write_srv
> 0)
1039 srv_idx
= i
% ms_setting
.srv_cnt
;
1040 srv_conn_cnt
= ms_setting
.sock_per_conn
* ms_setting
.nconns
;
1044 srv_idx
= ms_thread
->thread_ctx
->srv_idx
;
1045 srv_conn_cnt
= ms_setting
.nconns
/ ms_setting
.srv_cnt
;
1048 if (ms_network_connect(c
, ms_setting
.servers
[srv_idx
].srv_host_name
,
1049 ms_setting
.servers
[srv_idx
].srv_port
,
1050 ms_setting
.udp
, &ret_sfd
) == 0)
1052 c
->tcpsfd
[i
]= ret_sfd
;
1055 if (atomic_add_32_nv(&ms_setting
.servers
[srv_idx
].reconn_cnt
, 1)
1056 % (uint32_t)srv_conn_cnt
== 0)
1058 gettimeofday(&ms_setting
.servers
[srv_idx
].reconn_time
, NULL
);
1060 (int)(ms_setting
.servers
[srv_idx
].reconn_time
.tv_sec
1061 - ms_setting
.servers
[srv_idx
].disconn_time
1063 fprintf(stderr
, "Server %s:%d reconnect after %ds\n",
1064 ms_setting
.servers
[srv_idx
].srv_host_name
,
1065 ms_setting
.servers
[srv_idx
].srv_port
, reconn_time
);
1071 return EXIT_SUCCESS
;
1072 } /* ms_reconn_socks */
1076 * Tokenize the command string by replacing whitespace with '\0' and update
1077 * the token array tokens with pointer to start of each token and length.
1078 * Returns total number of tokens. The last valid token is the terminal
1079 * token (value points to the first unprocessed character of the string and
1084 * while(ms_tokenize_command(command, ncommand, tokens, max_tokens) > 0) {
1085 * for(int ix = 0; tokens[ix].length != 0; ix++) {
1088 * ncommand = tokens[ix].value - command;
1089 * command = tokens[ix].value;
1092 * @param command, the command string to token
1093 * @param tokens, array to store tokens
1094 * @param max_tokens, maximum tokens number
1096 * @return int, the number of tokens
1098 static int ms_tokenize_command(char *command
,
1100 const int max_tokens
)
1105 assert(command
!= NULL
&& tokens
!= NULL
&& max_tokens
> 1);
1107 for (s
= e
= command
; ntokens
< max_tokens
- 1; ++e
)
1113 tokens
[ntokens
].value
= s
;
1114 tokens
[ntokens
].length
= (size_t)(e
- s
);
1120 else if (*e
== '\0')
1124 tokens
[ntokens
].value
= s
;
1125 tokens
[ntokens
].length
= (size_t)(e
- s
);
1129 break; /* string end */
1134 } /* ms_tokenize_command */
1138 * parse the response of server.
1140 * @param c, pointer of the concurrency
1141 * @param command, the string responded by server
1143 * @return int, if the command completed return EXIT_SUCCESS, else return
1146 static int ms_ascii_process_line(ms_conn_t
*c
, char *command
)
1150 char *buffer
= command
;
1155 * for command get, we store the returned value into local buffer
1156 * then continue in ms_complete_nread().
1161 case 'V': /* VALUE || VERSION */
1162 if (buffer
[1] == 'A') /* VALUE */
1164 token_t tokens
[MAX_TOKENS
];
1165 ms_tokenize_command(command
, tokens
, MAX_TOKENS
);
1167 value_len
= strtol(tokens
[VALUELEN_TOKEN
].value
, NULL
, 10);
1170 printf("<%d ERROR %s\n", c
->sfd
, strerror(errno
));
1172 c
->currcmd
.key_prefix
= *(uint64_t *)tokens
[KEY_TOKEN
].value
;
1175 * We read the \r\n into the string since not doing so is more
1176 * cycles then the waster of memory to do so.
1178 * We are null terminating through, which will most likely make
1179 * some people lazy about using the return length.
1181 c
->rvbytes
= (int)(value_len
+ 2);
1189 c
->currcmd
.retstat
= MCD_SUCCESS
;
1192 case 'S': /* STORED STATS SERVER_ERROR */
1193 if (buffer
[2] == 'A') /* STORED STATS */
1195 c
->currcmd
.retstat
= MCD_STAT
;
1197 else if (buffer
[1] == 'E')
1200 printf("<%d %s\n", c
->sfd
, buffer
);
1202 c
->currcmd
.retstat
= MCD_SERVER_ERROR
;
1204 else if (buffer
[1] == 'T')
1207 c
->currcmd
.retstat
= MCD_STORED
;
1211 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1215 case 'D': /* DELETED DATA */
1216 if (buffer
[1] == 'E')
1218 c
->currcmd
.retstat
= MCD_DELETED
;
1222 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1227 case 'N': /* NOT_FOUND NOT_STORED*/
1228 if (buffer
[4] == 'F')
1230 c
->currcmd
.retstat
= MCD_NOTFOUND
;
1232 else if (buffer
[4] == 'S')
1234 printf("<%d %s\n", c
->sfd
, buffer
);
1235 c
->currcmd
.retstat
= MCD_NOTSTORED
;
1239 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1243 case 'E': /* PROTOCOL ERROR or END */
1244 if (buffer
[1] == 'N')
1247 c
->currcmd
.retstat
= MCD_END
;
1249 else if (buffer
[1] == 'R')
1251 printf("<%d ERROR\n", c
->sfd
);
1252 c
->currcmd
.retstat
= MCD_PROTOCOL_ERROR
;
1254 else if (buffer
[1] == 'X')
1256 c
->currcmd
.retstat
= MCD_DATA_EXISTS
;
1257 printf("<%d %s\n", c
->sfd
, buffer
);
1261 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1265 case 'C': /* CLIENT ERROR */
1266 printf("<%d %s\n", c
->sfd
, buffer
);
1267 c
->currcmd
.retstat
= MCD_CLIENT_ERROR
;
1271 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
1276 } /* ms_ascii_process_line */
1280 * after one operation completes, reset the concurrency
1282 * @param c, pointer of the concurrency
1283 * @param timeout, whether it's timeout
1285 void ms_reset_conn(ms_conn_t
*c
, bool timeout
)
1291 if ((c
->packets
> 0) && (c
->packets
< MAX_UDP_PACKET
))
1293 memset(c
->udppkt
, 0, sizeof(ms_udppkt_t
) * (size_t)c
->packets
);
1302 c
->currcmd
.isfinish
= true;
1309 ms_conn_set_state(c
, conn_write
);
1310 memcpy(&c
->precmd
, &c
->currcmd
, sizeof(ms_cmdstat_t
)); /* replicate command state */
1314 ms_drive_machine(c
);
1316 } /* ms_reset_conn */
1320 * if we have a complete line in the buffer, process it.
1322 * @param c, pointer of the concurrency
1324 * @return int, if success, return EXIT_SUCCESS, else return -1
1326 static int ms_try_read_line(ms_conn_t
*c
)
1328 if (c
->protocol
== binary_prot
)
1330 /* Do we have the complete packet header? */
1331 if ((uint64_t)c
->rbytes
< sizeof(c
->binary_header
))
1333 /* need more data! */
1334 return EXIT_SUCCESS
;
1339 if (((long)(c
->rcurr
)) % 8 != 0)
1341 /* must realign input buffer */
1342 memmove(c
->rbuf
, c
->rcurr
, c
->rbytes
);
1344 if (settings
.verbose
)
1346 fprintf(stderr
, "%d: Realign input buffer.\n", c
->sfd
);
1350 protocol_binary_response_header
*rsp
;
1351 rsp
= (protocol_binary_response_header
*)c
->rcurr
;
1353 c
->binary_header
= *rsp
;
1354 c
->binary_header
.response
.extlen
= rsp
->response
.extlen
;
1355 c
->binary_header
.response
.keylen
= ntohs(rsp
->response
.keylen
);
1356 c
->binary_header
.response
.bodylen
= ntohl(rsp
->response
.bodylen
);
1357 c
->binary_header
.response
.status
= ntohs(rsp
->response
.status
);
1359 if (c
->binary_header
.response
.magic
!= PROTOCOL_BINARY_RES
)
1361 fprintf(stderr
, "Invalid magic: %x\n",
1362 c
->binary_header
.response
.magic
);
1363 ms_conn_set_state(c
, conn_closing
);
1364 return EXIT_SUCCESS
;
1367 /* process this complete response */
1368 if (ms_bin_process_response(c
) == 0)
1370 /* current operation completed */
1371 ms_reset_conn(c
, false);
1376 c
->rbytes
-= (int32_t)sizeof(c
->binary_header
);
1377 c
->rcurr
+= sizeof(c
->binary_header
);
1386 assert(c
->rcurr
<= (c
->rbuf
+ c
->rsize
));
1389 return EXIT_SUCCESS
;
1391 el
= memchr(c
->rcurr
, '\n', (size_t)c
->rbytes
);
1393 return EXIT_SUCCESS
;
1396 if (((el
- c
->rcurr
) > 1) && (*(el
- 1) == '\r'))
1402 assert(cont
<= (c
->rcurr
+ c
->rbytes
));
1404 /* process this complete line */
1405 if (ms_ascii_process_line(c
, c
->rcurr
) == 0)
1407 /* current operation completed */
1408 ms_reset_conn(c
, false);
1413 /* current operation didn't complete */
1414 c
->rbytes
-= (int32_t)(cont
- c
->rcurr
);
1418 assert(c
->rcurr
<= (c
->rbuf
+ c
->rsize
));
1422 } /* ms_try_read_line */
1426 * because the packet of UDP can't ensure the order, the
1427 * function is used to sort the received udp packet.
1429 * @param c, pointer of the concurrency
1430 * @param buf, the buffer to store the ordered packages data
1431 * @param rbytes, the maximum capacity of the buffer
1433 * @return int, if success, return the copy bytes, else return
1436 static int ms_sort_udp_packet(ms_conn_t
*c
, char *buf
, int rbytes
)
1441 uint16_t seq_num
= 0;
1442 uint16_t packets
= 0;
1443 unsigned char *header
= NULL
;
1445 /* no enough data */
1447 assert(buf
!= NULL
);
1448 assert(c
->rudpbytes
>= UDP_HEADER_SIZE
);
1450 /* calculate received packets count */
1451 if (c
->rudpbytes
% UDP_MAX_PAYLOAD_SIZE
>= UDP_HEADER_SIZE
)
1453 /* the last packet has some data */
1454 c
->recvpkt
= c
->rudpbytes
/ UDP_MAX_PAYLOAD_SIZE
+ 1;
1458 c
->recvpkt
= c
->rudpbytes
/ UDP_MAX_PAYLOAD_SIZE
;
1461 /* get the total packets count if necessary */
1462 if (c
->packets
== 0)
1464 c
->packets
= HEADER_TO_PACKETS((unsigned char *)c
->rudpbuf
);
1467 /* build the ordered packet array */
1468 for (int i
= c
->pktcurr
; i
< c
->recvpkt
; i
++)
1470 header
= (unsigned char *)c
->rudpbuf
+ i
* UDP_MAX_PAYLOAD_SIZE
;
1471 req_id
= (uint16_t)HEADER_TO_REQID(header
);
1472 assert(req_id
== c
->request_id
% (1 << 16));
1474 packets
= (uint16_t)HEADER_TO_PACKETS(header
);
1475 assert(c
->packets
== HEADER_TO_PACKETS(header
));
1477 seq_num
= (uint16_t)HEADER_TO_SEQNUM(header
);
1478 c
->udppkt
[seq_num
].header
= header
;
1479 c
->udppkt
[seq_num
].data
= (char *)header
+ UDP_HEADER_SIZE
;
1481 if (i
== c
->recvpkt
- 1)
1483 /* last received packet */
1484 if (c
->rudpbytes
% UDP_MAX_PAYLOAD_SIZE
== 0)
1486 c
->udppkt
[seq_num
].rbytes
= UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
;
1491 c
->udppkt
[seq_num
].rbytes
= c
->rudpbytes
% UDP_MAX_PAYLOAD_SIZE
1497 c
->udppkt
[seq_num
].rbytes
= UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
;
1502 for (int i
= c
->ordcurr
; i
< c
->recvpkt
; i
++)
1504 /* there is some data to copy */
1505 if ((c
->udppkt
[i
].data
!= NULL
)
1506 && (c
->udppkt
[i
].copybytes
< c
->udppkt
[i
].rbytes
))
1508 header
= c
->udppkt
[i
].header
;
1509 len
= c
->udppkt
[i
].rbytes
- c
->udppkt
[i
].copybytes
;
1510 if (len
> rbytes
- wbytes
)
1512 len
= rbytes
- wbytes
;
1515 assert(len
<= rbytes
- wbytes
);
1516 assert(i
== HEADER_TO_SEQNUM(header
));
1518 memcpy(buf
+ wbytes
, c
->udppkt
[i
].data
+ c
->udppkt
[i
].copybytes
,
1521 c
->udppkt
[i
].copybytes
+= len
;
1523 if ((c
->udppkt
[i
].copybytes
== c
->udppkt
[i
].rbytes
)
1524 && (c
->udppkt
[i
].rbytes
== UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
))
1526 /* finish copying all the data of this packet, next */
1530 /* last received packet, and finish copying all the data */
1531 if ((c
->recvpkt
== c
->packets
) && (i
== c
->recvpkt
- 1)
1532 && (c
->udppkt
[i
].copybytes
== c
->udppkt
[i
].rbytes
))
1537 /* no space to copy data */
1538 if (wbytes
>= rbytes
)
1543 /* it doesn't finish reading all the data of the packet from network */
1544 if ((i
!= c
->recvpkt
- 1)
1545 && (c
->udppkt
[i
].rbytes
< UDP_MAX_PAYLOAD_SIZE
- UDP_HEADER_SIZE
))
1552 /* 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 EXIT_SUCCESS 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 EXIT_SUCCESS if there's nothing to read on the first read.
1660 * return EXIT_FAILURE 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_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_prot
2008 || c
->protocol
== binary_prot
);
2010 if (c
->protocol
== binary_prot
)
2012 ms_bin_complete_nread(c
);
2016 ms_ascii_complete_nread(c
);
2018 } /* ms_complete_nread */
2022 * Adds a message header to a connection.
2024 * @param c, pointer of the concurrency
2026 * @return int, if success, return EXIT_SUCCESS, else return -1
2028 static int ms_add_msghdr(ms_conn_t
*c
)
2034 if (c
->msgsize
== c
->msgused
)
2037 realloc(c
->msglist
, (size_t)c
->msgsize
* 2 * sizeof(struct msghdr
));
2045 msg
= c
->msglist
+ c
->msgused
;
2048 * this wipes msg_iovlen, msg_control, msg_controllen, and
2049 * msg_flags, the last 3 of which aren't defined on solaris:
2051 memset(msg
, 0, sizeof(struct msghdr
));
2053 msg
->msg_iov
= &c
->iov
[c
->iovused
];
2055 if (c
->udp
&& (c
->srv_recv_addr_size
> 0))
2057 msg
->msg_name
= &c
->srv_recv_addr
;
2058 msg
->msg_namelen
= c
->srv_recv_addr_size
;
2066 /* Leave room for the UDP header, which we'll fill in later. */
2067 return ms_add_iov(c
, NULL
, UDP_HEADER_SIZE
);
2070 return EXIT_SUCCESS
;
2071 } /* ms_add_msghdr */
2075 * Ensures that there is room for another structure iovec in a connection's
2078 * @param c, pointer of the concurrency
2080 * @return int, if success, return EXIT_SUCCESS, else return -1
2082 static int ms_ensure_iov_space(ms_conn_t
*c
)
2086 if (c
->iovused
>= c
->iovsize
)
2089 struct iovec
*new_iov
= (struct iovec
*)realloc(c
->iov
,
2092 * sizeof(struct iovec
));
2099 /* Point all the msghdr structures at the new list. */
2100 for (i
= 0, iovnum
= 0; i
< c
->msgused
; i
++)
2102 c
->msglist
[i
].msg_iov
= &c
->iov
[iovnum
];
2103 iovnum
+= (int)c
->msglist
[i
].msg_iovlen
;
2107 return EXIT_SUCCESS
;
2108 } /* ms_ensure_iov_space */
2112 * Adds data to the list of pending data that will be written out to a
2115 * @param c, pointer of the concurrency
2116 * @param buf, the buffer includes data to send
2117 * @param len, the data length in the buffer
2119 * @return int, if success, return EXIT_SUCCESS, else return -1
2121 static int ms_add_iov(ms_conn_t
*c
, const void *buf
, int len
)
2131 m
= &c
->msglist
[c
->msgused
- 1];
2134 * Limit UDP packets, to UDP_MAX_PAYLOAD_SIZE bytes.
2136 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];
2148 if (ms_ensure_iov_space(c
) != 0)
2151 /* If the fragment is too big to fit in the datagram, split it up */
2152 if (limit_to_mtu
&& (len
+ c
->msgbytes
> UDP_MAX_SEND_PAYLOAD_SIZE
))
2154 leftover
= len
+ c
->msgbytes
- UDP_MAX_SEND_PAYLOAD_SIZE
;
2162 m
= &c
->msglist
[c
->msgused
- 1];
2163 m
->msg_iov
[m
->msg_iovlen
].iov_base
= (void *)buf
;
2164 m
->msg_iov
[m
->msg_iovlen
].iov_len
= (size_t)len
;
2170 buf
= ((char *)buf
) + len
;
2173 while (leftover
> 0);
2175 return EXIT_SUCCESS
;
2180 * Constructs a set of UDP headers and attaches them to the outgoing messages.
2182 * @param c, pointer of the concurrency
2184 * @return int, if success, return EXIT_SUCCESS, else return -1
2186 static int ms_build_udp_headers(ms_conn_t
*c
)
2193 c
->request_id
= ms_get_udp_request_id();
2195 if (c
->msgused
> c
->hdrsize
)
2199 new_hdrbuf
= realloc(c
->hdrbuf
,
2200 (size_t)c
->msgused
* 2 * UDP_HEADER_SIZE
);
2202 new_hdrbuf
= malloc((size_t)c
->msgused
* 2 * UDP_HEADER_SIZE
);
2206 c
->hdrbuf
= (unsigned char *)new_hdrbuf
;
2207 c
->hdrsize
= c
->msgused
* 2;
2210 /* If this is a multi-packet request, drop it. */
2211 if (c
->udp
&& (c
->msgused
> 1))
2213 fprintf(stderr
, "multi-packet request for UDP not supported.\n");
2218 for (i
= 0; i
< c
->msgused
; i
++)
2220 c
->msglist
[i
].msg_iov
[0].iov_base
= (void *)hdr
;
2221 c
->msglist
[i
].msg_iov
[0].iov_len
= UDP_HEADER_SIZE
;
2222 *hdr
++= (unsigned char)(c
->request_id
/ 256);
2223 *hdr
++= (unsigned char)(c
->request_id
% 256);
2224 *hdr
++= (unsigned char)(i
/ 256);
2225 *hdr
++= (unsigned char)(i
% 256);
2226 *hdr
++= (unsigned char)(c
->msgused
/ 256);
2227 *hdr
++= (unsigned char)(c
->msgused
% 256);
2228 *hdr
++= (unsigned char)1; /* support facebook memcached */
2229 *hdr
++= (unsigned char)0;
2231 ((unsigned char *)c
->msglist
[i
].msg_iov
[0].iov_base
2232 + UDP_HEADER_SIZE
));
2235 return EXIT_SUCCESS
;
2236 } /* ms_build_udp_headers */
2240 * Transmit the next chunk of data from our list of msgbuf structures.
2242 * @param c, pointer of the concurrency
2244 * @return TRANSMIT_COMPLETE All done writing.
2245 * TRANSMIT_INCOMPLETE More data remaining to write.
2246 * TRANSMIT_SOFT_ERROR Can't write any more right now.
2247 * TRANSMIT_HARD_ERROR Can't write (c->state is set to conn_closing)
2249 static int ms_transmit(ms_conn_t
*c
)
2253 if ((c
->msgcurr
< c
->msgused
)
2254 && (c
->msglist
[c
->msgcurr
].msg_iovlen
== 0))
2256 /* Finished writing the current msg; advance to the next. */
2260 if (c
->msgcurr
< c
->msgused
)
2263 struct msghdr
*m
= &c
->msglist
[c
->msgcurr
];
2265 res
= sendmsg(c
->sfd
, m
, 0);
2268 atomic_add_size(&ms_stats
.bytes_written
, res
);
2270 /* We've written some of the data. Remove the completed
2271 * iovec entries from the list of pending writes. */
2272 while (m
->msg_iovlen
> 0 && res
>= (ssize_t
)m
->msg_iov
->iov_len
)
2274 res
-= (ssize_t
)m
->msg_iov
->iov_len
;
2279 /* Might have written just part of the last iovec entry;
2280 * adjust it so the next write will do the rest. */
2283 m
->msg_iov
->iov_base
= (void *)((unsigned char *)m
->msg_iov
->iov_base
+ res
);
2284 m
->msg_iov
->iov_len
-= (size_t)res
;
2286 return TRANSMIT_INCOMPLETE
;
2288 if ((res
== -1) && ((errno
== EAGAIN
) || (errno
== EWOULDBLOCK
)))
2290 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2292 fprintf(stderr
, "Couldn't update event.\n");
2293 ms_conn_set_state(c
, conn_closing
);
2294 return TRANSMIT_HARD_ERROR
;
2296 return TRANSMIT_SOFT_ERROR
;
2299 /* if res==0 or res==-1 and error is not EAGAIN or EWOULDBLOCK,
2300 * we have a real error, on which we close the connection */
2301 fprintf(stderr
, "Failed to write, and not due to blocking.\n");
2303 ms_conn_set_state(c
, conn_closing
);
2304 return TRANSMIT_HARD_ERROR
;
2308 return TRANSMIT_COMPLETE
;
2314 * Shrinks a connection's buffers if they're too big. This prevents
2315 * periodic large "mget" response from server chewing lots of client
2318 * This should only be called in between requests since it can wipe output
2321 * @param c, pointer of the concurrency
2323 static void ms_conn_shrink(ms_conn_t
*c
)
2330 if ((c
->rsize
> READ_BUFFER_HIGHWAT
) && (c
->rbytes
< DATA_BUFFER_SIZE
))
2334 if (c
->rcurr
!= c
->rbuf
)
2335 memmove(c
->rbuf
, c
->rcurr
, (size_t)c
->rbytes
);
2337 newbuf
= (char *)realloc((void *)c
->rbuf
, DATA_BUFFER_SIZE
);
2342 c
->rsize
= DATA_BUFFER_SIZE
;
2347 if (c
->udp
&& (c
->rudpsize
> UDP_DATA_BUFFER_HIGHWAT
)
2348 && (c
->rudpbytes
+ UDP_MAX_PAYLOAD_SIZE
< UDP_DATA_BUFFER_SIZE
))
2350 char *new_rbuf
= (char *)realloc(c
->rudpbuf
, (size_t)c
->rudpsize
* 2);
2353 c
->rudpbuf
= new_rbuf
;
2354 c
->rudpsize
= UDP_DATA_BUFFER_SIZE
;
2356 /* TODO check error condition? */
2359 if (c
->msgsize
> MSG_LIST_HIGHWAT
)
2361 struct msghdr
*newbuf
= (struct msghdr
*)realloc(
2364 * sizeof(c
->msglist
[0]));
2368 c
->msgsize
= MSG_LIST_INITIAL
;
2370 /* TODO check error condition? */
2373 if (c
->iovsize
> IOV_LIST_HIGHWAT
)
2375 struct iovec
*newbuf
= (struct iovec
*)realloc((void *)c
->iov
,
2377 * sizeof(c
->iov
[0]));
2381 c
->iovsize
= IOV_LIST_INITIAL
;
2383 /* TODO check return value */
2385 } /* ms_conn_shrink */
2389 * Sets a connection's current state in the state machine. Any special
2390 * processing that needs to happen on certain state transitions can
2393 * @param c, pointer of the concurrency
2394 * @param state, connection state
2396 static void ms_conn_set_state(ms_conn_t
*c
, int state
)
2400 if (state
!= c
->state
)
2402 if (state
== conn_read
)
2408 } /* ms_conn_set_state */
2412 * update the event if socks change state. for example: when
2413 * change the listen scoket read event to sock write event, or
2414 * change socket handler, we could call this function.
2416 * @param c, pointer of the concurrency
2417 * @param new_flags, new event flags
2419 * @return bool, if success, return true, else return false
2421 static bool ms_update_event(ms_conn_t
*c
, const int new_flags
)
2425 struct event_base
*base
= c
->event
.ev_base
;
2426 if ((c
->ev_flags
== new_flags
) && (ms_setting
.rep_write_srv
== 0)
2427 && (! ms_setting
.facebook_test
|| (c
->total_sfds
== 1)))
2432 if (event_del(&c
->event
) == -1)
2434 /* try to delete the event again */
2435 if (event_del(&c
->event
) == -1)
2441 event_set(&c
->event
,
2446 event_base_set(base
, &c
->event
);
2447 c
->ev_flags
= (short)new_flags
;
2449 if (event_add(&c
->event
, NULL
) == -1)
2455 } /* ms_update_event */
2459 * If user want to get the expected throughput, we could limit
2460 * the performance of memslap. we could give up some work and
2461 * just wait a short time. The function is used to check this
2464 * @param c, pointer of the concurrency
2466 * @return bool, if success, return true, else return false
2468 static bool ms_need_yield(ms_conn_t
*c
)
2470 ms_thread_t
*ms_thread
= pthread_getspecific(ms_thread_key
);
2472 int64_t time_diff
= 0;
2473 struct timeval curr_time
;
2474 ms_task_t
*task
= &c
->curr_task
;
2476 if (ms_setting
.expected_tps
> 0)
2478 gettimeofday(&curr_time
, NULL
);
2479 time_diff
= ms_time_diff(&ms_thread
->startup_time
, &curr_time
);
2480 tps
= (int64_t)(((task
->get_opt
+ task
->set_opt
) / (uint64_t)time_diff
) * 1000000);
2482 /* current throughput is greater than expected throughput */
2483 if (tps
> ms_thread
->thread_ctx
->tps_perconn
)
2490 } /* ms_need_yield */
2494 * used to update the start time of each operation
2496 * @param c, pointer of the concurrency
2498 static void ms_update_start_time(ms_conn_t
*c
)
2500 ms_task_item_t
*item
= c
->curr_task
.item
;
2502 if ((ms_setting
.stat_freq
> 0) || c
->udp
2503 || ((c
->currcmd
.cmd
== CMD_SET
) && (item
->exp_time
> 0)))
2505 gettimeofday(&c
->start_time
, NULL
);
2506 if ((c
->currcmd
.cmd
== CMD_SET
) && (item
->exp_time
> 0))
2508 /* record the current time */
2509 item
->client_time
= c
->start_time
.tv_sec
;
2512 } /* ms_update_start_time */
2516 * run the state machine
2518 * @param c, pointer of the concurrency
2520 static void ms_drive_machine(ms_conn_t
*c
)
2533 if (c
->rbytes
>= c
->rvbytes
)
2535 ms_complete_nread(c
);
2541 if (ms_try_read_line(c
) != 0)
2547 if (ms_try_read_network(c
) != 0)
2552 /* doesn't read all the response data, wait event wake up */
2553 if (! c
->currcmd
.isfinish
)
2555 if (! ms_update_event(c
, EV_READ
| EV_PERSIST
))
2557 fprintf(stderr
, "Couldn't update event.\n");
2558 ms_conn_set_state(c
, conn_closing
);
2565 /* we have no command line and no data to read from network, next write */
2566 ms_conn_set_state(c
, conn_write
);
2567 memcpy(&c
->precmd
, &c
->currcmd
, sizeof(ms_cmdstat_t
)); /* replicate command state */
2572 if (! c
->ctnwrite
&& ms_need_yield(c
))
2576 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2578 fprintf(stderr
, "Couldn't update event.\n");
2579 ms_conn_set_state(c
, conn_closing
);
2586 if (! c
->ctnwrite
&& (ms_exec_task(c
) != 0))
2588 ms_conn_set_state(c
, conn_closing
);
2592 /* record the start time before starting to send data if necessary */
2593 if (! c
->ctnwrite
|| (c
->change_sfd
&& c
->ctnwrite
))
2597 c
->change_sfd
= false;
2599 ms_update_start_time(c
);
2602 /* change sfd if necessary */
2610 /* execute task until nothing need be written to network */
2611 if (! c
->ctnwrite
&& (c
->msgcurr
== c
->msgused
))
2613 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2615 fprintf(stderr
, "Couldn't update event.\n");
2616 ms_conn_set_state(c
, conn_closing
);
2623 switch (ms_transmit(c
))
2625 case TRANSMIT_COMPLETE
:
2626 /* we have no data to write to network, next wait repose */
2627 if (! ms_update_event(c
, EV_READ
| EV_PERSIST
))
2629 fprintf(stderr
, "Couldn't update event.\n");
2630 ms_conn_set_state(c
, conn_closing
);
2634 ms_conn_set_state(c
, conn_read
);
2639 case TRANSMIT_INCOMPLETE
:
2641 break; /* Continue in state machine. */
2643 case TRANSMIT_HARD_ERROR
:
2647 case TRANSMIT_SOFT_ERROR
:
2659 /* recovery mode, need reconnect if connection close */
2660 if (ms_setting
.reconnect
&& (! ms_global
.time_out
2661 || ((ms_setting
.run_time
== 0)
2662 && (c
->remain_exec_num
> 0))))
2664 if (ms_reconn(c
) != 0)
2671 ms_reset_conn(c
, false);
2673 if (c
->total_sfds
== 1)
2675 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2677 fprintf(stderr
, "Couldn't update event.\n");
2678 ms_conn_set_state(c
, conn_closing
);
2696 } /* ms_drive_machine */
2700 * the event handler of each thread
2702 * @param fd, the file descriptor of socket
2703 * @param which, event flag
2704 * @param arg, argument
2706 void ms_event_handler(const int fd
, const short which
, void *arg
)
2708 ms_conn_t
*c
= (ms_conn_t
*)arg
;
2718 "Catastrophic: event fd: %d doesn't match conn fd: %d\n",
2724 assert(fd
== c
->sfd
);
2726 ms_drive_machine(c
);
2728 /* wait for next event */
2729 } /* ms_event_handler */
2733 * get the next socket descriptor index to run for replication
2735 * @param c, pointer of the concurrency
2736 * @param cmd, command(get or set )
2738 * @return int, if success, return the index, else return EXIT_SUCCESS
2740 static uint32_t ms_get_rep_sock_index(ms_conn_t
*c
, int cmd
)
2742 uint32_t sock_index
= 0;
2745 if (c
->total_sfds
== 1)
2747 return EXIT_SUCCESS
;
2750 if (ms_setting
.rep_write_srv
== 0)
2759 for (i
= 0; i
< ms_setting
.rep_write_srv
; i
++)
2761 if (c
->tcpsfd
[i
] > 0)
2767 if (i
== ms_setting
.rep_write_srv
)
2769 /* random get one replication server to read */
2770 sock_index
= (uint32_t)random() % c
->total_sfds
;
2774 /* random get one replication writing server to write */
2775 sock_index
= (uint32_t)random() % ms_setting
.rep_write_srv
;
2778 else if (cmd
== CMD_GET
)
2780 /* random get one replication server to read */
2781 sock_index
= (uint32_t)random() % c
->total_sfds
;
2784 while (c
->tcpsfd
[sock_index
] == 0);
2787 } /* ms_get_rep_sock_index */
2791 * get the next socket descriptor index to run
2793 * @param c, pointer of the concurrency
2795 * @return int, return the index
2797 static uint32_t ms_get_next_sock_index(ms_conn_t
*c
)
2799 uint32_t sock_index
= 0;
2803 sock_index
= (++c
->cur_idx
== c
->total_sfds
) ? 0 : c
->cur_idx
;
2805 while (c
->tcpsfd
[sock_index
] == 0);
2808 } /* ms_get_next_sock_index */
2812 * update socket event of the connections
2814 * @param c, pointer of the concurrency
2816 * @return int, if success, return EXIT_SUCCESS, else return -1
2818 static int ms_update_conn_sock_event(ms_conn_t
*c
)
2822 switch (c
->currcmd
.cmd
)
2825 if (ms_setting
.facebook_test
&& c
->udp
)
2827 c
->sfd
= c
->tcpsfd
[0];
2829 c
->change_sfd
= true;
2834 if (ms_setting
.facebook_test
&& ! c
->udp
)
2838 c
->change_sfd
= true;
2846 if (! c
->udp
&& (c
->total_sfds
> 1))
2848 if (c
->cur_idx
!= c
->total_sfds
)
2850 if (ms_setting
.rep_write_srv
== 0)
2852 c
->cur_idx
= ms_get_next_sock_index(c
);
2856 c
->cur_idx
= ms_get_rep_sock_index(c
, c
->currcmd
.cmd
);
2861 /* must select the first sock of the connection at the beginning */
2865 c
->sfd
= c
->tcpsfd
[c
->cur_idx
];
2866 assert(c
->sfd
!= 0);
2867 c
->change_sfd
= true;
2872 if (! ms_update_event(c
, EV_WRITE
| EV_PERSIST
))
2874 fprintf(stderr
, "Couldn't update event.\n");
2875 ms_conn_set_state(c
, conn_closing
);
2880 return EXIT_SUCCESS
;
2881 } /* ms_update_conn_sock_event */
2885 * for ASCII protocol, this function build the set command
2886 * string and send the command.
2888 * @param c, pointer of the concurrency
2889 * @param item, pointer of task item which includes the object
2892 * @return int, if success, return EXIT_SUCCESS, else return -1
2894 static int ms_build_ascii_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
)
2898 char *buffer
= c
->wbuf
;
2900 write_len
= snprintf(buffer
,
2907 if (write_len
> c
->wsize
|| write_len
< 0)
2909 /* ought to be always enough. just fail for simplicity */
2910 fprintf(stderr
, "output command line too long.\n");
2914 if (item
->value_offset
== INVALID_OFFSET
)
2916 value_offset
= item
->key_suffix_offset
;
2920 value_offset
= item
->value_offset
;
2923 if ((ms_add_iov(c
, "set ", 4) != 0)
2924 || (ms_add_iov(c
, (char *)&item
->key_prefix
,
2925 (int)KEY_PREFIX_SIZE
) != 0)
2926 || (ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
2927 item
->key_size
- (int)KEY_PREFIX_SIZE
) != 0)
2928 || (ms_add_iov(c
, buffer
, write_len
) != 0)
2929 || (ms_add_iov(c
, &ms_setting
.char_block
[value_offset
],
2930 item
->value_size
) != 0)
2931 || (ms_add_iov(c
, "\r\n", 2) != 0)
2932 || (c
->udp
&& (ms_build_udp_headers(c
) != 0)))
2937 return EXIT_SUCCESS
;
2938 } /* ms_build_ascii_write_buf_set */
2942 * used to send set command to server
2944 * @param c, pointer of the concurrency
2945 * @param item, pointer of task item which includes the object
2948 * @return int, if success, return EXIT_SUCCESS, else return -1
2950 int ms_mcd_set(ms_conn_t
*c
, ms_task_item_t
*item
)
2954 c
->currcmd
.cmd
= CMD_SET
;
2955 c
->currcmd
.isfinish
= false;
2956 c
->currcmd
.retstat
= MCD_FAILURE
;
2958 if (ms_update_conn_sock_event(c
) != 0)
2966 if (ms_add_msghdr(c
) != 0)
2968 fprintf(stderr
, "Out of memory preparing request.");
2972 /* binary protocol */
2973 if (c
->protocol
== binary_prot
)
2975 if (ms_build_bin_write_buf_set(c
, item
) != 0)
2982 if (ms_build_ascii_write_buf_set(c
, item
) != 0)
2988 atomic_add_size(&ms_stats
.obj_bytes
,
2989 item
->key_size
+ item
->value_size
);
2990 atomic_add_size(&ms_stats
.cmd_set
, 1);
2992 return EXIT_SUCCESS
;
2997 * for ASCII protocol, this function build the get command
2998 * string and send the command.
3000 * @param c, pointer of the concurrency
3001 * @param item, pointer of task item which includes the object
3004 * @return int, if success, return EXIT_SUCCESS, else return -1
3006 static int ms_build_ascii_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
)
3008 if ((ms_add_iov(c
, "get ", 4) != 0)
3009 || (ms_add_iov(c
, (char *)&item
->key_prefix
,
3010 (int)KEY_PREFIX_SIZE
) != 0)
3011 || (ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
3012 item
->key_size
- (int)KEY_PREFIX_SIZE
) != 0)
3013 || (ms_add_iov(c
, "\r\n", 2) != 0)
3014 || (c
->udp
&& (ms_build_udp_headers(c
) != 0)))
3019 return EXIT_SUCCESS
;
3020 } /* ms_build_ascii_write_buf_get */
3024 * used to send the get command to server
3026 * @param c, pointer of the concurrency
3027 * @param item, pointer of task item which includes the object
3030 * @return int, if success, return EXIT_SUCCESS, else return -1
3032 int ms_mcd_get(ms_conn_t
*c
, ms_task_item_t
*item
)
3036 c
->currcmd
.cmd
= CMD_GET
;
3037 c
->currcmd
.isfinish
= false;
3038 c
->currcmd
.retstat
= MCD_FAILURE
;
3040 if (ms_update_conn_sock_event(c
) != 0)
3048 if (ms_add_msghdr(c
) != 0)
3050 fprintf(stderr
, "Out of memory preparing request.");
3054 /* binary protocol */
3055 if (c
->protocol
== binary_prot
)
3057 if (ms_build_bin_write_buf_get(c
, item
) != 0)
3064 if (ms_build_ascii_write_buf_get(c
, item
) != 0)
3070 atomic_add_size(&ms_stats
.cmd_get
, 1);
3072 return EXIT_SUCCESS
;
3077 * for ASCII protocol, this function build the multi-get command
3078 * string and send the command.
3080 * @param c, pointer of the concurrency
3082 * @return int, if success, return EXIT_SUCCESS, else return -1
3084 static int ms_build_ascii_write_buf_mlget(ms_conn_t
*c
)
3086 ms_task_item_t
*item
;
3088 if (ms_add_iov(c
, "get", 3) != 0)
3093 for (int i
= 0; i
< c
->mlget_task
.mlget_num
; i
++)
3095 item
= c
->mlget_task
.mlget_item
[i
].item
;
3096 assert(item
!= NULL
);
3097 if ((ms_add_iov(c
, " ", 1) != 0)
3098 || (ms_add_iov(c
, (char *)&item
->key_prefix
,
3099 (int)KEY_PREFIX_SIZE
) != 0)
3100 || (ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
3101 item
->key_size
- (int)KEY_PREFIX_SIZE
) != 0))
3107 if ((ms_add_iov(c
, "\r\n", 2) != 0)
3108 || (c
->udp
&& (ms_build_udp_headers(c
) != 0)))
3113 return EXIT_SUCCESS
;
3114 } /* ms_build_ascii_write_buf_mlget */
3118 * used to send the multi-get command to server
3120 * @param c, pointer of the concurrency
3122 * @return int, if success, return EXIT_SUCCESS, else return -1
3124 int ms_mcd_mlget(ms_conn_t
*c
)
3126 ms_task_item_t
*item
;
3129 assert(c
->mlget_task
.mlget_num
>= 1);
3131 c
->currcmd
.cmd
= CMD_GET
;
3132 c
->currcmd
.isfinish
= false;
3133 c
->currcmd
.retstat
= MCD_FAILURE
;
3135 if (ms_update_conn_sock_event(c
) != 0)
3143 if (ms_add_msghdr(c
) != 0)
3145 fprintf(stderr
, "Out of memory preparing request.");
3149 /* binary protocol */
3150 if (c
->protocol
== binary_prot
)
3152 if (ms_build_bin_write_buf_mlget(c
) != 0)
3159 if (ms_build_ascii_write_buf_mlget(c
) != 0)
3165 /* decrease operation time of each item */
3166 for (int i
= 0; i
< c
->mlget_task
.mlget_num
; i
++)
3168 item
= c
->mlget_task
.mlget_item
[i
].item
;
3169 atomic_add_size(&ms_stats
.cmd_get
, 1);
3174 return EXIT_SUCCESS
;
3175 } /* ms_mcd_mlget */
3179 * binary protocol support
3183 * for binary protocol, parse the response of server
3185 * @param c, pointer of the concurrency
3187 * @return int, if success, return EXIT_SUCCESS, else return -1
3189 static int ms_bin_process_response(ms_conn_t
*c
)
3191 const char *errstr
= NULL
;
3195 uint32_t bodylen
= c
->binary_header
.response
.bodylen
;
3196 uint8_t opcode
= c
->binary_header
.response
.opcode
;
3197 uint16_t status
= c
->binary_header
.response
.status
;
3201 c
->rvbytes
= (int32_t)bodylen
;
3203 return EXIT_FAILURE
;
3209 case PROTOCOL_BINARY_RESPONSE_SUCCESS
:
3210 if (opcode
== PROTOCOL_BINARY_CMD_SET
)
3212 c
->currcmd
.retstat
= MCD_STORED
;
3214 else if (opcode
== PROTOCOL_BINARY_CMD_DELETE
)
3216 c
->currcmd
.retstat
= MCD_DELETED
;
3218 else if (opcode
== PROTOCOL_BINARY_CMD_GET
)
3220 c
->currcmd
.retstat
= MCD_END
;
3224 case PROTOCOL_BINARY_RESPONSE_ENOMEM
:
3225 errstr
= "Out of memory";
3226 c
->currcmd
.retstat
= MCD_SERVER_ERROR
;
3229 case PROTOCOL_BINARY_RESPONSE_UNKNOWN_COMMAND
:
3230 errstr
= "Unknown command";
3231 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
3234 case PROTOCOL_BINARY_RESPONSE_KEY_ENOENT
:
3235 errstr
= "Not found";
3236 c
->currcmd
.retstat
= MCD_NOTFOUND
;
3239 case PROTOCOL_BINARY_RESPONSE_EINVAL
:
3240 errstr
= "Invalid arguments";
3241 c
->currcmd
.retstat
= MCD_PROTOCOL_ERROR
;
3244 case PROTOCOL_BINARY_RESPONSE_KEY_EEXISTS
:
3245 errstr
= "Data exists for key.";
3248 case PROTOCOL_BINARY_RESPONSE_E2BIG
:
3249 errstr
= "Too large.";
3250 c
->currcmd
.retstat
= MCD_SERVER_ERROR
;
3253 case PROTOCOL_BINARY_RESPONSE_NOT_STORED
:
3254 errstr
= "Not stored.";
3255 c
->currcmd
.retstat
= MCD_NOTSTORED
;
3259 errstr
= "Unknown error";
3260 c
->currcmd
.retstat
= MCD_UNKNOWN_READ_FAILURE
;
3266 fprintf(stderr
, "%s\n", errstr
);
3270 return EXIT_SUCCESS
;
3271 } /* ms_bin_process_response */
3274 /* build binary header and add the header to the buffer to send */
3277 * build binary header and add the header to the buffer to send
3279 * @param c, pointer of the concurrency
3280 * @param opcode, operation code
3281 * @param hdr_len, length of header
3282 * @param key_len, length of key
3283 * @param body_len. length of body
3285 static void ms_add_bin_header(ms_conn_t
*c
,
3291 protocol_binary_request_header
*header
;
3295 header
= (protocol_binary_request_header
*)c
->wcurr
;
3297 header
->request
.magic
= (uint8_t)PROTOCOL_BINARY_REQ
;
3298 header
->request
.opcode
= (uint8_t)opcode
;
3299 header
->request
.keylen
= htons(key_len
);
3301 header
->request
.extlen
= (uint8_t)hdr_len
;
3302 header
->request
.datatype
= (uint8_t)PROTOCOL_BINARY_RAW_BYTES
;
3303 header
->request
.vbucket
= 0;
3305 header
->request
.bodylen
= htonl(body_len
);
3306 header
->request
.opaque
= 0;
3307 header
->request
.cas
= 0;
3309 ms_add_iov(c
, c
->wcurr
, sizeof(header
->request
));
3310 } /* ms_add_bin_header */
3314 * add the key to the socket write buffer array
3316 * @param c, pointer of the concurrency
3317 * @param item, pointer of task item which includes the object
3320 static void ms_add_key_to_iov(ms_conn_t
*c
, ms_task_item_t
*item
)
3322 ms_add_iov(c
, (char *)&item
->key_prefix
, (int)KEY_PREFIX_SIZE
);
3323 ms_add_iov(c
, &ms_setting
.char_block
[item
->key_suffix_offset
],
3324 item
->key_size
- (int)KEY_PREFIX_SIZE
);
3329 * for binary protocol, this function build the set command
3330 * and add the command to send buffer array.
3332 * @param c, pointer of the concurrency
3333 * @param item, pointer of task item which includes the object
3336 * @return int, if success, return EXIT_SUCCESS, else return -1
3338 static int ms_build_bin_write_buf_set(ms_conn_t
*c
, ms_task_item_t
*item
)
3340 assert(c
->wbuf
== c
->wcurr
);
3343 protocol_binary_request_set
*rep
= (protocol_binary_request_set
*)c
->wcurr
;
3344 uint16_t keylen
= (uint16_t)item
->key_size
;
3345 uint32_t bodylen
= (uint32_t)sizeof(rep
->message
.body
)
3346 + (uint32_t)keylen
+ (uint32_t)item
->value_size
;
3348 ms_add_bin_header(c
,
3349 PROTOCOL_BINARY_CMD_SET
,
3350 sizeof(rep
->message
.body
),
3353 rep
->message
.body
.flags
= 0;
3354 rep
->message
.body
.expiration
= htonl((uint32_t)item
->exp_time
);
3355 ms_add_iov(c
, &rep
->message
.body
, sizeof(rep
->message
.body
));
3356 ms_add_key_to_iov(c
, item
);
3358 if (item
->value_offset
== INVALID_OFFSET
)
3360 value_offset
= item
->key_suffix_offset
;
3364 value_offset
= item
->value_offset
;
3366 ms_add_iov(c
, &ms_setting
.char_block
[value_offset
], item
->value_size
);
3368 return EXIT_SUCCESS
;
3369 } /* ms_build_bin_write_buf_set */
3373 * for binary protocol, this function build the get command and
3374 * add the command to send buffer array.
3376 * @param c, pointer of the concurrency
3377 * @param item, pointer of task item which includes the object
3380 * @return int, if success, return EXIT_SUCCESS, else return -1
3382 static int ms_build_bin_write_buf_get(ms_conn_t
*c
, ms_task_item_t
*item
)
3384 assert(c
->wbuf
== c
->wcurr
);
3386 ms_add_bin_header(c
, PROTOCOL_BINARY_CMD_GET
, 0, (uint16_t)item
->key_size
,
3387 (uint32_t)item
->key_size
);
3388 ms_add_key_to_iov(c
, item
);
3390 return EXIT_SUCCESS
;
3391 } /* ms_build_bin_write_buf_get */
3395 * for binary protocol, this function build the multi-get
3396 * command and add the command to send buffer array.
3398 * @param c, pointer of the concurrency
3399 * @param item, pointer of task item which includes the object
3402 * @return int, if success, return EXIT_SUCCESS, else return -1
3404 static int ms_build_bin_write_buf_mlget(ms_conn_t
*c
)
3406 ms_task_item_t
*item
;
3408 assert(c
->wbuf
== c
->wcurr
);
3410 for (int i
= 0; i
< c
->mlget_task
.mlget_num
; i
++)
3412 item
= c
->mlget_task
.mlget_item
[i
].item
;
3413 assert(item
!= NULL
);
3415 ms_add_bin_header(c
,
3416 PROTOCOL_BINARY_CMD_GET
,
3418 (uint16_t)item
->key_size
,
3419 (uint32_t)item
->key_size
);
3420 ms_add_key_to_iov(c
, item
);
3421 c
->wcurr
+= sizeof(protocol_binary_request_get
);
3426 return EXIT_SUCCESS
;
3427 } /* ms_build_bin_write_buf_mlget */