]> git.saurik.com Git - redis.git/blob - src/cluster.c
Minor MIGRATE implementation simplification about ttl handling.
[redis.git] / src / cluster.c
1 #include "redis.h"
2 #include "endianconv.h"
3
4 #include <arpa/inet.h>
5 #include <fcntl.h>
6 #include <unistd.h>
7
8 void clusterAcceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
9 void clusterReadHandler(aeEventLoop *el, int fd, void *privdata, int mask);
10 void clusterSendPing(clusterLink *link, int type);
11 void clusterSendFail(char *nodename);
12 void clusterUpdateState(void);
13 int clusterNodeGetSlotBit(clusterNode *n, int slot);
14 sds clusterGenNodesDescription(void);
15 clusterNode *clusterLookupNode(char *name);
16 int clusterNodeAddSlave(clusterNode *master, clusterNode *slave);
17 int clusterAddSlot(clusterNode *n, int slot);
18
19 /* -----------------------------------------------------------------------------
20 * Initialization
21 * -------------------------------------------------------------------------- */
22
23 int clusterLoadConfig(char *filename) {
24 FILE *fp = fopen(filename,"r");
25 char *line;
26 int maxline, j;
27
28 if (fp == NULL) return REDIS_ERR;
29
30 /* Parse the file. Note that single liens of the cluster config file can
31 * be really long as they include all the hash slots of the node.
32 * This means in the worst possible case REDIS_CLUSTER_SLOTS/2 integers.
33 * To simplify we allocate 1024+REDIS_CLUSTER_SLOTS*16 bytes per line. */
34 maxline = 1024+REDIS_CLUSTER_SLOTS*16;
35 line = zmalloc(maxline);
36 while(fgets(line,maxline,fp) != NULL) {
37 int argc;
38 sds *argv = sdssplitargs(line,&argc);
39 clusterNode *n, *master;
40 char *p, *s;
41
42 /* Create this node if it does not exist */
43 n = clusterLookupNode(argv[0]);
44 if (!n) {
45 n = createClusterNode(argv[0],0);
46 clusterAddNode(n);
47 }
48 /* Address and port */
49 if ((p = strchr(argv[1],':')) == NULL) goto fmterr;
50 *p = '\0';
51 memcpy(n->ip,argv[1],strlen(argv[1])+1);
52 n->port = atoi(p+1);
53
54 /* Parse flags */
55 p = s = argv[2];
56 while(p) {
57 p = strchr(s,',');
58 if (p) *p = '\0';
59 if (!strcasecmp(s,"myself")) {
60 redisAssert(server.cluster.myself == NULL);
61 server.cluster.myself = n;
62 n->flags |= REDIS_NODE_MYSELF;
63 } else if (!strcasecmp(s,"master")) {
64 n->flags |= REDIS_NODE_MASTER;
65 } else if (!strcasecmp(s,"slave")) {
66 n->flags |= REDIS_NODE_SLAVE;
67 } else if (!strcasecmp(s,"fail?")) {
68 n->flags |= REDIS_NODE_PFAIL;
69 } else if (!strcasecmp(s,"fail")) {
70 n->flags |= REDIS_NODE_FAIL;
71 } else if (!strcasecmp(s,"handshake")) {
72 n->flags |= REDIS_NODE_HANDSHAKE;
73 } else if (!strcasecmp(s,"noaddr")) {
74 n->flags |= REDIS_NODE_NOADDR;
75 } else if (!strcasecmp(s,"noflags")) {
76 /* nothing to do */
77 } else {
78 redisPanic("Unknown flag in redis cluster config file");
79 }
80 if (p) s = p+1;
81 }
82
83 /* Get master if any. Set the master and populate master's
84 * slave list. */
85 if (argv[3][0] != '-') {
86 master = clusterLookupNode(argv[3]);
87 if (!master) {
88 master = createClusterNode(argv[3],0);
89 clusterAddNode(master);
90 }
91 n->slaveof = master;
92 clusterNodeAddSlave(master,n);
93 }
94
95 /* Set ping sent / pong received timestamps */
96 if (atoi(argv[4])) n->ping_sent = time(NULL);
97 if (atoi(argv[5])) n->pong_received = time(NULL);
98
99 /* Populate hash slots served by this instance. */
100 for (j = 7; j < argc; j++) {
101 int start, stop;
102
103 if (argv[j][0] == '[') {
104 /* Here we handle migrating / importing slots */
105 int slot;
106 char direction;
107 clusterNode *cn;
108
109 p = strchr(argv[j],'-');
110 redisAssert(p != NULL);
111 *p = '\0';
112 direction = p[1]; /* Either '>' or '<' */
113 slot = atoi(argv[j]+1);
114 p += 3;
115 cn = clusterLookupNode(p);
116 if (!cn) {
117 cn = createClusterNode(p,0);
118 clusterAddNode(cn);
119 }
120 if (direction == '>') {
121 server.cluster.migrating_slots_to[slot] = cn;
122 } else {
123 server.cluster.importing_slots_from[slot] = cn;
124 }
125 continue;
126 } else if ((p = strchr(argv[j],'-')) != NULL) {
127 *p = '\0';
128 start = atoi(argv[j]);
129 stop = atoi(p+1);
130 } else {
131 start = stop = atoi(argv[j]);
132 }
133 while(start <= stop) clusterAddSlot(n, start++);
134 }
135
136 sdssplitargs_free(argv,argc);
137 }
138 zfree(line);
139 fclose(fp);
140
141 /* Config sanity check */
142 redisAssert(server.cluster.myself != NULL);
143 redisLog(REDIS_NOTICE,"Node configuration loaded, I'm %.40s",
144 server.cluster.myself->name);
145 clusterUpdateState();
146 return REDIS_OK;
147
148 fmterr:
149 redisLog(REDIS_WARNING,"Unrecovarable error: corrupted cluster config file.");
150 fclose(fp);
151 exit(1);
152 }
153
154 /* Cluster node configuration is exactly the same as CLUSTER NODES output.
155 *
156 * This function writes the node config and returns 0, on error -1
157 * is returned. */
158 int clusterSaveConfig(void) {
159 sds ci = clusterGenNodesDescription();
160 int fd;
161
162 if ((fd = open(server.cluster.configfile,O_WRONLY|O_CREAT|O_TRUNC,0644))
163 == -1) goto err;
164 if (write(fd,ci,sdslen(ci)) != (ssize_t)sdslen(ci)) goto err;
165 close(fd);
166 sdsfree(ci);
167 return 0;
168
169 err:
170 sdsfree(ci);
171 return -1;
172 }
173
174 void clusterSaveConfigOrDie(void) {
175 if (clusterSaveConfig() == -1) {
176 redisLog(REDIS_WARNING,"Fatal: can't update cluster config file.");
177 exit(1);
178 }
179 }
180
181 void clusterInit(void) {
182 int saveconf = 0;
183
184 server.cluster.myself = NULL;
185 server.cluster.state = REDIS_CLUSTER_FAIL;
186 server.cluster.nodes = dictCreate(&clusterNodesDictType,NULL);
187 server.cluster.node_timeout = 15;
188 memset(server.cluster.migrating_slots_to,0,
189 sizeof(server.cluster.migrating_slots_to));
190 memset(server.cluster.importing_slots_from,0,
191 sizeof(server.cluster.importing_slots_from));
192 memset(server.cluster.slots,0,
193 sizeof(server.cluster.slots));
194 if (clusterLoadConfig(server.cluster.configfile) == REDIS_ERR) {
195 /* No configuration found. We will just use the random name provided
196 * by the createClusterNode() function. */
197 server.cluster.myself = createClusterNode(NULL,REDIS_NODE_MYSELF);
198 redisLog(REDIS_NOTICE,"No cluster configuration found, I'm %.40s",
199 server.cluster.myself->name);
200 clusterAddNode(server.cluster.myself);
201 saveconf = 1;
202 }
203 if (saveconf) clusterSaveConfigOrDie();
204 /* We need a listening TCP port for our cluster messaging needs */
205 server.cfd = anetTcpServer(server.neterr,
206 server.port+REDIS_CLUSTER_PORT_INCR, server.bindaddr);
207 if (server.cfd == -1) {
208 redisLog(REDIS_WARNING, "Opening cluster TCP port: %s", server.neterr);
209 exit(1);
210 }
211 if (aeCreateFileEvent(server.el, server.cfd, AE_READABLE,
212 clusterAcceptHandler, NULL) == AE_ERR) oom("creating file event");
213 server.cluster.slots_to_keys = zslCreate();
214 }
215
216 /* -----------------------------------------------------------------------------
217 * CLUSTER communication link
218 * -------------------------------------------------------------------------- */
219
220 clusterLink *createClusterLink(clusterNode *node) {
221 clusterLink *link = zmalloc(sizeof(*link));
222 link->sndbuf = sdsempty();
223 link->rcvbuf = sdsempty();
224 link->node = node;
225 link->fd = -1;
226 return link;
227 }
228
229 /* Free a cluster link, but does not free the associated node of course.
230 * Just this function will make sure that the original node associated
231 * with this link will have the 'link' field set to NULL. */
232 void freeClusterLink(clusterLink *link) {
233 if (link->fd != -1) {
234 aeDeleteFileEvent(server.el, link->fd, AE_WRITABLE);
235 aeDeleteFileEvent(server.el, link->fd, AE_READABLE);
236 }
237 sdsfree(link->sndbuf);
238 sdsfree(link->rcvbuf);
239 if (link->node)
240 link->node->link = NULL;
241 close(link->fd);
242 zfree(link);
243 }
244
245 void clusterAcceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) {
246 int cport, cfd;
247 char cip[128];
248 clusterLink *link;
249 REDIS_NOTUSED(el);
250 REDIS_NOTUSED(mask);
251 REDIS_NOTUSED(privdata);
252
253 cfd = anetTcpAccept(server.neterr, fd, cip, &cport);
254 if (cfd == AE_ERR) {
255 redisLog(REDIS_VERBOSE,"Accepting cluster node: %s", server.neterr);
256 return;
257 }
258 redisLog(REDIS_VERBOSE,"Accepted cluster node %s:%d", cip, cport);
259 /* We need to create a temporary node in order to read the incoming
260 * packet in a valid contest. This node will be released once we
261 * read the packet and reply. */
262 link = createClusterLink(NULL);
263 link->fd = cfd;
264 aeCreateFileEvent(server.el,cfd,AE_READABLE,clusterReadHandler,link);
265 }
266
267 /* -----------------------------------------------------------------------------
268 * Key space handling
269 * -------------------------------------------------------------------------- */
270
271 /* We have 4096 hash slots. The hash slot of a given key is obtained
272 * as the least significant 12 bits of the crc16 of the key. */
273 unsigned int keyHashSlot(char *key, int keylen) {
274 return crc16(key,keylen) & 0x0FFF;
275 }
276
277 /* -----------------------------------------------------------------------------
278 * CLUSTER node API
279 * -------------------------------------------------------------------------- */
280
281 /* Create a new cluster node, with the specified flags.
282 * If "nodename" is NULL this is considered a first handshake and a random
283 * node name is assigned to this node (it will be fixed later when we'll
284 * receive the first pong).
285 *
286 * The node is created and returned to the user, but it is not automatically
287 * added to the nodes hash table. */
288 clusterNode *createClusterNode(char *nodename, int flags) {
289 clusterNode *node = zmalloc(sizeof(*node));
290
291 if (nodename)
292 memcpy(node->name, nodename, REDIS_CLUSTER_NAMELEN);
293 else
294 getRandomHexChars(node->name, REDIS_CLUSTER_NAMELEN);
295 node->flags = flags;
296 memset(node->slots,0,sizeof(node->slots));
297 node->numslaves = 0;
298 node->slaves = NULL;
299 node->slaveof = NULL;
300 node->ping_sent = node->pong_received = 0;
301 node->configdigest = NULL;
302 node->configdigest_ts = 0;
303 node->link = NULL;
304 return node;
305 }
306
307 int clusterNodeRemoveSlave(clusterNode *master, clusterNode *slave) {
308 int j;
309
310 for (j = 0; j < master->numslaves; j++) {
311 if (master->slaves[j] == slave) {
312 memmove(master->slaves+j,master->slaves+(j+1),
313 (master->numslaves-1)-j);
314 master->numslaves--;
315 return REDIS_OK;
316 }
317 }
318 return REDIS_ERR;
319 }
320
321 int clusterNodeAddSlave(clusterNode *master, clusterNode *slave) {
322 int j;
323
324 /* If it's already a slave, don't add it again. */
325 for (j = 0; j < master->numslaves; j++)
326 if (master->slaves[j] == slave) return REDIS_ERR;
327 master->slaves = zrealloc(master->slaves,
328 sizeof(clusterNode*)*(master->numslaves+1));
329 master->slaves[master->numslaves] = slave;
330 master->numslaves++;
331 return REDIS_OK;
332 }
333
334 void clusterNodeResetSlaves(clusterNode *n) {
335 zfree(n->slaves);
336 n->numslaves = 0;
337 }
338
339 void freeClusterNode(clusterNode *n) {
340 sds nodename;
341
342 nodename = sdsnewlen(n->name, REDIS_CLUSTER_NAMELEN);
343 redisAssert(dictDelete(server.cluster.nodes,nodename) == DICT_OK);
344 sdsfree(nodename);
345 if (n->slaveof) clusterNodeRemoveSlave(n->slaveof, n);
346 if (n->link) freeClusterLink(n->link);
347 zfree(n);
348 }
349
350 /* Add a node to the nodes hash table */
351 int clusterAddNode(clusterNode *node) {
352 int retval;
353
354 retval = dictAdd(server.cluster.nodes,
355 sdsnewlen(node->name,REDIS_CLUSTER_NAMELEN), node);
356 return (retval == DICT_OK) ? REDIS_OK : REDIS_ERR;
357 }
358
359 /* Node lookup by name */
360 clusterNode *clusterLookupNode(char *name) {
361 sds s = sdsnewlen(name, REDIS_CLUSTER_NAMELEN);
362 struct dictEntry *de;
363
364 de = dictFind(server.cluster.nodes,s);
365 sdsfree(s);
366 if (de == NULL) return NULL;
367 return dictGetVal(de);
368 }
369
370 /* This is only used after the handshake. When we connect a given IP/PORT
371 * as a result of CLUSTER MEET we don't have the node name yet, so we
372 * pick a random one, and will fix it when we receive the PONG request using
373 * this function. */
374 void clusterRenameNode(clusterNode *node, char *newname) {
375 int retval;
376 sds s = sdsnewlen(node->name, REDIS_CLUSTER_NAMELEN);
377
378 redisLog(REDIS_DEBUG,"Renaming node %.40s into %.40s",
379 node->name, newname);
380 retval = dictDelete(server.cluster.nodes, s);
381 sdsfree(s);
382 redisAssert(retval == DICT_OK);
383 memcpy(node->name, newname, REDIS_CLUSTER_NAMELEN);
384 clusterAddNode(node);
385 }
386
387 /* -----------------------------------------------------------------------------
388 * CLUSTER messages exchange - PING/PONG and gossip
389 * -------------------------------------------------------------------------- */
390
391 /* Process the gossip section of PING or PONG packets.
392 * Note that this function assumes that the packet is already sanity-checked
393 * by the caller, not in the content of the gossip section, but in the
394 * length. */
395 void clusterProcessGossipSection(clusterMsg *hdr, clusterLink *link) {
396 uint16_t count = ntohs(hdr->count);
397 clusterMsgDataGossip *g = (clusterMsgDataGossip*) hdr->data.ping.gossip;
398 clusterNode *sender = link->node ? link->node : clusterLookupNode(hdr->sender);
399
400 while(count--) {
401 sds ci = sdsempty();
402 uint16_t flags = ntohs(g->flags);
403 clusterNode *node;
404
405 if (flags == 0) ci = sdscat(ci,"noflags,");
406 if (flags & REDIS_NODE_MYSELF) ci = sdscat(ci,"myself,");
407 if (flags & REDIS_NODE_MASTER) ci = sdscat(ci,"master,");
408 if (flags & REDIS_NODE_SLAVE) ci = sdscat(ci,"slave,");
409 if (flags & REDIS_NODE_PFAIL) ci = sdscat(ci,"fail?,");
410 if (flags & REDIS_NODE_FAIL) ci = sdscat(ci,"fail,");
411 if (flags & REDIS_NODE_HANDSHAKE) ci = sdscat(ci,"handshake,");
412 if (flags & REDIS_NODE_NOADDR) ci = sdscat(ci,"noaddr,");
413 if (ci[sdslen(ci)-1] == ',') ci[sdslen(ci)-1] = ' ';
414
415 redisLog(REDIS_DEBUG,"GOSSIP %.40s %s:%d %s",
416 g->nodename,
417 g->ip,
418 ntohs(g->port),
419 ci);
420 sdsfree(ci);
421
422 /* Update our state accordingly to the gossip sections */
423 node = clusterLookupNode(g->nodename);
424 if (node != NULL) {
425 /* We already know this node. Let's start updating the last
426 * time PONG figure if it is newer than our figure.
427 * Note that it's not a problem if we have a PING already
428 * in progress against this node. */
429 if (node->pong_received < (signed) ntohl(g->pong_received)) {
430 redisLog(REDIS_DEBUG,"Node pong_received updated by gossip");
431 node->pong_received = ntohl(g->pong_received);
432 }
433 /* Mark this node as FAILED if we think it is possibly failing
434 * and another node also thinks it's failing. */
435 if (node->flags & REDIS_NODE_PFAIL &&
436 (flags & (REDIS_NODE_FAIL|REDIS_NODE_PFAIL)))
437 {
438 redisLog(REDIS_NOTICE,"Received a PFAIL acknowledge from node %.40s, marking node %.40s as FAIL!", hdr->sender, node->name);
439 node->flags &= ~REDIS_NODE_PFAIL;
440 node->flags |= REDIS_NODE_FAIL;
441 /* Broadcast the failing node name to everybody */
442 clusterSendFail(node->name);
443 clusterUpdateState();
444 clusterSaveConfigOrDie();
445 }
446 } else {
447 /* If it's not in NOADDR state and we don't have it, we
448 * start an handshake process against this IP/PORT pairs.
449 *
450 * Note that we require that the sender of this gossip message
451 * is a well known node in our cluster, otherwise we risk
452 * joining another cluster. */
453 if (sender && !(flags & REDIS_NODE_NOADDR)) {
454 clusterNode *newnode;
455
456 redisLog(REDIS_DEBUG,"Adding the new node");
457 newnode = createClusterNode(NULL,REDIS_NODE_HANDSHAKE);
458 memcpy(newnode->ip,g->ip,sizeof(g->ip));
459 newnode->port = ntohs(g->port);
460 clusterAddNode(newnode);
461 }
462 }
463
464 /* Next node */
465 g++;
466 }
467 }
468
469 /* IP -> string conversion. 'buf' is supposed to at least be 16 bytes. */
470 void nodeIp2String(char *buf, clusterLink *link) {
471 struct sockaddr_in sa;
472 socklen_t salen = sizeof(sa);
473
474 if (getpeername(link->fd, (struct sockaddr*) &sa, &salen) == -1)
475 redisPanic("getpeername() failed.");
476 strncpy(buf,inet_ntoa(sa.sin_addr),sizeof(link->node->ip));
477 }
478
479
480 /* Update the node address to the IP address that can be extracted
481 * from link->fd, and at the specified port. */
482 void nodeUpdateAddress(clusterNode *node, clusterLink *link, int port) {
483 /* TODO */
484 }
485
486 /* When this function is called, there is a packet to process starting
487 * at node->rcvbuf. Releasing the buffer is up to the caller, so this
488 * function should just handle the higher level stuff of processing the
489 * packet, modifying the cluster state if needed.
490 *
491 * The function returns 1 if the link is still valid after the packet
492 * was processed, otherwise 0 if the link was freed since the packet
493 * processing lead to some inconsistency error (for instance a PONG
494 * received from the wrong sender ID). */
495 int clusterProcessPacket(clusterLink *link) {
496 clusterMsg *hdr = (clusterMsg*) link->rcvbuf;
497 uint32_t totlen = ntohl(hdr->totlen);
498 uint16_t type = ntohs(hdr->type);
499 clusterNode *sender;
500
501 redisLog(REDIS_DEBUG,"--- Processing packet of type %d, %lu bytes",
502 type, (unsigned long) totlen);
503
504 /* Perform sanity checks */
505 if (totlen < 8) return 1;
506 if (totlen > sdslen(link->rcvbuf)) return 1;
507 if (type == CLUSTERMSG_TYPE_PING || type == CLUSTERMSG_TYPE_PONG ||
508 type == CLUSTERMSG_TYPE_MEET)
509 {
510 uint16_t count = ntohs(hdr->count);
511 uint32_t explen; /* expected length of this packet */
512
513 explen = sizeof(clusterMsg)-sizeof(union clusterMsgData);
514 explen += (sizeof(clusterMsgDataGossip)*count);
515 if (totlen != explen) return 1;
516 }
517 if (type == CLUSTERMSG_TYPE_FAIL) {
518 uint32_t explen = sizeof(clusterMsg)-sizeof(union clusterMsgData);
519
520 explen += sizeof(clusterMsgDataFail);
521 if (totlen != explen) return 1;
522 }
523 if (type == CLUSTERMSG_TYPE_PUBLISH) {
524 uint32_t explen = sizeof(clusterMsg)-sizeof(union clusterMsgData);
525
526 explen += sizeof(clusterMsgDataPublish) +
527 ntohl(hdr->data.publish.msg.channel_len) +
528 ntohl(hdr->data.publish.msg.message_len);
529 if (totlen != explen) return 1;
530 }
531
532 /* Ready to process the packet. Dispatch by type. */
533 sender = clusterLookupNode(hdr->sender);
534 if (type == CLUSTERMSG_TYPE_PING || type == CLUSTERMSG_TYPE_MEET) {
535 int update_config = 0;
536 redisLog(REDIS_DEBUG,"Ping packet received: %p", link->node);
537
538 /* Add this node if it is new for us and the msg type is MEET.
539 * In this stage we don't try to add the node with the right
540 * flags, slaveof pointer, and so forth, as this details will be
541 * resolved when we'll receive PONGs from the server. */
542 if (!sender && type == CLUSTERMSG_TYPE_MEET) {
543 clusterNode *node;
544
545 node = createClusterNode(NULL,REDIS_NODE_HANDSHAKE);
546 nodeIp2String(node->ip,link);
547 node->port = ntohs(hdr->port);
548 clusterAddNode(node);
549 update_config = 1;
550 }
551
552 /* Get info from the gossip section */
553 clusterProcessGossipSection(hdr,link);
554
555 /* Anyway reply with a PONG */
556 clusterSendPing(link,CLUSTERMSG_TYPE_PONG);
557
558 /* Update config if needed */
559 if (update_config) clusterSaveConfigOrDie();
560 } else if (type == CLUSTERMSG_TYPE_PONG) {
561 int update_state = 0;
562 int update_config = 0;
563
564 redisLog(REDIS_DEBUG,"Pong packet received: %p", link->node);
565 if (link->node) {
566 if (link->node->flags & REDIS_NODE_HANDSHAKE) {
567 /* If we already have this node, try to change the
568 * IP/port of the node with the new one. */
569 if (sender) {
570 redisLog(REDIS_WARNING,
571 "Handshake error: we already know node %.40s, updating the address if needed.", sender->name);
572 nodeUpdateAddress(sender,link,ntohs(hdr->port));
573 freeClusterNode(link->node); /* will free the link too */
574 return 0;
575 }
576
577 /* First thing to do is replacing the random name with the
578 * right node name if this was an handshake stage. */
579 clusterRenameNode(link->node, hdr->sender);
580 redisLog(REDIS_DEBUG,"Handshake with node %.40s completed.",
581 link->node->name);
582 link->node->flags &= ~REDIS_NODE_HANDSHAKE;
583 update_config = 1;
584 } else if (memcmp(link->node->name,hdr->sender,
585 REDIS_CLUSTER_NAMELEN) != 0)
586 {
587 /* If the reply has a non matching node ID we
588 * disconnect this node and set it as not having an associated
589 * address. */
590 redisLog(REDIS_DEBUG,"PONG contains mismatching sender ID");
591 link->node->flags |= REDIS_NODE_NOADDR;
592 freeClusterLink(link);
593 update_config = 1;
594 /* FIXME: remove this node if we already have it.
595 *
596 * If we already have it but the IP is different, use
597 * the new one if the old node is in FAIL, PFAIL, or NOADDR
598 * status... */
599 return 0;
600 }
601 }
602 /* Update our info about the node */
603 if (link->node) link->node->pong_received = time(NULL);
604
605 /* Update master/slave info */
606 if (sender) {
607 if (!memcmp(hdr->slaveof,REDIS_NODE_NULL_NAME,
608 sizeof(hdr->slaveof)))
609 {
610 sender->flags &= ~REDIS_NODE_SLAVE;
611 sender->flags |= REDIS_NODE_MASTER;
612 sender->slaveof = NULL;
613 } else {
614 clusterNode *master = clusterLookupNode(hdr->slaveof);
615
616 sender->flags &= ~REDIS_NODE_MASTER;
617 sender->flags |= REDIS_NODE_SLAVE;
618 if (sender->numslaves) clusterNodeResetSlaves(sender);
619 if (master) clusterNodeAddSlave(master,sender);
620 }
621 }
622
623 /* Update our info about served slots if this new node is serving
624 * slots that are not served from our point of view. */
625 if (sender && sender->flags & REDIS_NODE_MASTER) {
626 int newslots, j;
627
628 newslots =
629 memcmp(sender->slots,hdr->myslots,sizeof(hdr->myslots)) != 0;
630 memcpy(sender->slots,hdr->myslots,sizeof(hdr->myslots));
631 if (newslots) {
632 for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) {
633 if (clusterNodeGetSlotBit(sender,j)) {
634 if (server.cluster.slots[j] == sender) continue;
635 if (server.cluster.slots[j] == NULL ||
636 server.cluster.slots[j]->flags & REDIS_NODE_FAIL)
637 {
638 server.cluster.slots[j] = sender;
639 update_state = update_config = 1;
640 }
641 }
642 }
643 }
644 }
645
646 /* Get info from the gossip section */
647 clusterProcessGossipSection(hdr,link);
648
649 /* Update the cluster state if needed */
650 if (update_state) clusterUpdateState();
651 if (update_config) clusterSaveConfigOrDie();
652 } else if (type == CLUSTERMSG_TYPE_FAIL && sender) {
653 clusterNode *failing;
654
655 failing = clusterLookupNode(hdr->data.fail.about.nodename);
656 if (failing && !(failing->flags & (REDIS_NODE_FAIL|REDIS_NODE_MYSELF)))
657 {
658 redisLog(REDIS_NOTICE,
659 "FAIL message received from %.40s about %.40s",
660 hdr->sender, hdr->data.fail.about.nodename);
661 failing->flags |= REDIS_NODE_FAIL;
662 failing->flags &= ~REDIS_NODE_PFAIL;
663 clusterUpdateState();
664 clusterSaveConfigOrDie();
665 }
666 } else if (type == CLUSTERMSG_TYPE_PUBLISH) {
667 robj *channel, *message;
668 uint32_t channel_len, message_len;
669
670 /* Don't bother creating useless objects if there are no Pub/Sub subscribers. */
671 if (dictSize(server.pubsub_channels) || listLength(server.pubsub_patterns)) {
672 channel_len = ntohl(hdr->data.publish.msg.channel_len);
673 message_len = ntohl(hdr->data.publish.msg.message_len);
674 channel = createStringObject(
675 (char*)hdr->data.publish.msg.bulk_data,channel_len);
676 message = createStringObject(
677 (char*)hdr->data.publish.msg.bulk_data+channel_len, message_len);
678 pubsubPublishMessage(channel,message);
679 decrRefCount(channel);
680 decrRefCount(message);
681 }
682 } else {
683 redisLog(REDIS_WARNING,"Received unknown packet type: %d", type);
684 }
685 return 1;
686 }
687
688 /* This function is called when we detect the link with this node is lost.
689 We set the node as no longer connected. The Cluster Cron will detect
690 this connection and will try to get it connected again.
691
692 Instead if the node is a temporary node used to accept a query, we
693 completely free the node on error. */
694 void handleLinkIOError(clusterLink *link) {
695 freeClusterLink(link);
696 }
697
698 /* Send data. This is handled using a trivial send buffer that gets
699 * consumed by write(). We don't try to optimize this for speed too much
700 * as this is a very low traffic channel. */
701 void clusterWriteHandler(aeEventLoop *el, int fd, void *privdata, int mask) {
702 clusterLink *link = (clusterLink*) privdata;
703 ssize_t nwritten;
704 REDIS_NOTUSED(el);
705 REDIS_NOTUSED(mask);
706
707 nwritten = write(fd, link->sndbuf, sdslen(link->sndbuf));
708 if (nwritten <= 0) {
709 redisLog(REDIS_NOTICE,"I/O error writing to node link: %s",
710 strerror(errno));
711 handleLinkIOError(link);
712 return;
713 }
714 link->sndbuf = sdsrange(link->sndbuf,nwritten,-1);
715 if (sdslen(link->sndbuf) == 0)
716 aeDeleteFileEvent(server.el, link->fd, AE_WRITABLE);
717 }
718
719 /* Read data. Try to read the first field of the header first to check the
720 * full length of the packet. When a whole packet is in memory this function
721 * will call the function to process the packet. And so forth. */
722 void clusterReadHandler(aeEventLoop *el, int fd, void *privdata, int mask) {
723 char buf[1024];
724 ssize_t nread;
725 clusterMsg *hdr;
726 clusterLink *link = (clusterLink*) privdata;
727 int readlen;
728 REDIS_NOTUSED(el);
729 REDIS_NOTUSED(mask);
730
731 again:
732 if (sdslen(link->rcvbuf) >= 4) {
733 hdr = (clusterMsg*) link->rcvbuf;
734 readlen = ntohl(hdr->totlen) - sdslen(link->rcvbuf);
735 } else {
736 readlen = 4 - sdslen(link->rcvbuf);
737 }
738
739 nread = read(fd,buf,readlen);
740 if (nread == -1 && errno == EAGAIN) return; /* Just no data */
741
742 if (nread <= 0) {
743 /* I/O error... */
744 redisLog(REDIS_NOTICE,"I/O error reading from node link: %s",
745 (nread == 0) ? "connection closed" : strerror(errno));
746 handleLinkIOError(link);
747 return;
748 } else {
749 /* Read data and recast the pointer to the new buffer. */
750 link->rcvbuf = sdscatlen(link->rcvbuf,buf,nread);
751 hdr = (clusterMsg*) link->rcvbuf;
752 }
753
754 /* Total length obtained? read the payload now instead of burning
755 * cycles waiting for a new event to fire. */
756 if (sdslen(link->rcvbuf) == 4) goto again;
757
758 /* Whole packet in memory? We can process it. */
759 if (sdslen(link->rcvbuf) == ntohl(hdr->totlen)) {
760 if (clusterProcessPacket(link)) {
761 sdsfree(link->rcvbuf);
762 link->rcvbuf = sdsempty();
763 }
764 }
765 }
766
767 /* Put stuff into the send buffer. */
768 void clusterSendMessage(clusterLink *link, unsigned char *msg, size_t msglen) {
769 if (sdslen(link->sndbuf) == 0 && msglen != 0)
770 aeCreateFileEvent(server.el,link->fd,AE_WRITABLE,
771 clusterWriteHandler,link);
772
773 link->sndbuf = sdscatlen(link->sndbuf, msg, msglen);
774 }
775
776 /* Send a message to all the nodes with a reliable link */
777 void clusterBroadcastMessage(void *buf, size_t len) {
778 dictIterator *di;
779 dictEntry *de;
780
781 di = dictGetIterator(server.cluster.nodes);
782 while((de = dictNext(di)) != NULL) {
783 clusterNode *node = dictGetVal(de);
784
785 if (!node->link) continue;
786 if (node->flags & (REDIS_NODE_MYSELF|REDIS_NODE_NOADDR)) continue;
787 clusterSendMessage(node->link,buf,len);
788 }
789 dictReleaseIterator(di);
790 }
791
792 /* Build the message header */
793 void clusterBuildMessageHdr(clusterMsg *hdr, int type) {
794 int totlen = 0;
795
796 memset(hdr,0,sizeof(*hdr));
797 hdr->type = htons(type);
798 memcpy(hdr->sender,server.cluster.myself->name,REDIS_CLUSTER_NAMELEN);
799 memcpy(hdr->myslots,server.cluster.myself->slots,
800 sizeof(hdr->myslots));
801 memset(hdr->slaveof,0,REDIS_CLUSTER_NAMELEN);
802 if (server.cluster.myself->slaveof != NULL) {
803 memcpy(hdr->slaveof,server.cluster.myself->slaveof->name,
804 REDIS_CLUSTER_NAMELEN);
805 }
806 hdr->port = htons(server.port);
807 hdr->state = server.cluster.state;
808 memset(hdr->configdigest,0,32); /* FIXME: set config digest */
809
810 if (type == CLUSTERMSG_TYPE_FAIL) {
811 totlen = sizeof(clusterMsg)-sizeof(union clusterMsgData);
812 totlen += sizeof(clusterMsgDataFail);
813 }
814 hdr->totlen = htonl(totlen);
815 /* For PING, PONG, and MEET, fixing the totlen field is up to the caller */
816 }
817
818 /* Send a PING or PONG packet to the specified node, making sure to add enough
819 * gossip informations. */
820 void clusterSendPing(clusterLink *link, int type) {
821 unsigned char buf[1024];
822 clusterMsg *hdr = (clusterMsg*) buf;
823 int gossipcount = 0, totlen;
824 /* freshnodes is the number of nodes we can still use to populate the
825 * gossip section of the ping packet. Basically we start with the nodes
826 * we have in memory minus two (ourself and the node we are sending the
827 * message to). Every time we add a node we decrement the counter, so when
828 * it will drop to <= zero we know there is no more gossip info we can
829 * send. */
830 int freshnodes = dictSize(server.cluster.nodes)-2;
831
832 if (link->node && type == CLUSTERMSG_TYPE_PING)
833 link->node->ping_sent = time(NULL);
834 clusterBuildMessageHdr(hdr,type);
835
836 /* Populate the gossip fields */
837 while(freshnodes > 0 && gossipcount < 3) {
838 struct dictEntry *de = dictGetRandomKey(server.cluster.nodes);
839 clusterNode *this = dictGetVal(de);
840 clusterMsgDataGossip *gossip;
841 int j;
842
843 /* Not interesting to gossip about ourself.
844 * Nor to send gossip info about HANDSHAKE state nodes (zero info). */
845 if (this == server.cluster.myself ||
846 this->flags & REDIS_NODE_HANDSHAKE) {
847 freshnodes--; /* otherwise we may loop forever. */
848 continue;
849 }
850
851 /* Check if we already added this node */
852 for (j = 0; j < gossipcount; j++) {
853 if (memcmp(hdr->data.ping.gossip[j].nodename,this->name,
854 REDIS_CLUSTER_NAMELEN) == 0) break;
855 }
856 if (j != gossipcount) continue;
857
858 /* Add it */
859 freshnodes--;
860 gossip = &(hdr->data.ping.gossip[gossipcount]);
861 memcpy(gossip->nodename,this->name,REDIS_CLUSTER_NAMELEN);
862 gossip->ping_sent = htonl(this->ping_sent);
863 gossip->pong_received = htonl(this->pong_received);
864 memcpy(gossip->ip,this->ip,sizeof(this->ip));
865 gossip->port = htons(this->port);
866 gossip->flags = htons(this->flags);
867 gossipcount++;
868 }
869 totlen = sizeof(clusterMsg)-sizeof(union clusterMsgData);
870 totlen += (sizeof(clusterMsgDataGossip)*gossipcount);
871 hdr->count = htons(gossipcount);
872 hdr->totlen = htonl(totlen);
873 clusterSendMessage(link,buf,totlen);
874 }
875
876 /* Send a PUBLISH message.
877 *
878 * If link is NULL, then the message is broadcasted to the whole cluster. */
879 void clusterSendPublish(clusterLink *link, robj *channel, robj *message) {
880 unsigned char buf[4096], *payload;
881 clusterMsg *hdr = (clusterMsg*) buf;
882 uint32_t totlen;
883 uint32_t channel_len, message_len;
884
885 channel = getDecodedObject(channel);
886 message = getDecodedObject(message);
887 channel_len = sdslen(channel->ptr);
888 message_len = sdslen(message->ptr);
889
890 clusterBuildMessageHdr(hdr,CLUSTERMSG_TYPE_PUBLISH);
891 totlen = sizeof(clusterMsg)-sizeof(union clusterMsgData);
892 totlen += sizeof(clusterMsgDataPublish) + channel_len + message_len;
893
894 hdr->data.publish.msg.channel_len = htonl(channel_len);
895 hdr->data.publish.msg.message_len = htonl(message_len);
896 hdr->totlen = htonl(totlen);
897
898 /* Try to use the local buffer if possible */
899 if (totlen < sizeof(buf)) {
900 payload = buf;
901 } else {
902 payload = zmalloc(totlen);
903 hdr = (clusterMsg*) payload;
904 memcpy(payload,hdr,sizeof(hdr));
905 }
906 memcpy(hdr->data.publish.msg.bulk_data,channel->ptr,sdslen(channel->ptr));
907 memcpy(hdr->data.publish.msg.bulk_data+sdslen(channel->ptr),
908 message->ptr,sdslen(message->ptr));
909
910 if (link)
911 clusterSendMessage(link,payload,totlen);
912 else
913 clusterBroadcastMessage(payload,totlen);
914
915 decrRefCount(channel);
916 decrRefCount(message);
917 if (payload != buf) zfree(payload);
918 }
919
920 /* Send a FAIL message to all the nodes we are able to contact.
921 * The FAIL message is sent when we detect that a node is failing
922 * (REDIS_NODE_PFAIL) and we also receive a gossip confirmation of this:
923 * we switch the node state to REDIS_NODE_FAIL and ask all the other
924 * nodes to do the same ASAP. */
925 void clusterSendFail(char *nodename) {
926 unsigned char buf[1024];
927 clusterMsg *hdr = (clusterMsg*) buf;
928
929 clusterBuildMessageHdr(hdr,CLUSTERMSG_TYPE_FAIL);
930 memcpy(hdr->data.fail.about.nodename,nodename,REDIS_CLUSTER_NAMELEN);
931 clusterBroadcastMessage(buf,ntohl(hdr->totlen));
932 }
933
934 /* -----------------------------------------------------------------------------
935 * CLUSTER Pub/Sub support
936 *
937 * For now we do very little, just propagating PUBLISH messages across the whole
938 * cluster. In the future we'll try to get smarter and avoiding propagating those
939 * messages to hosts without receives for a given channel.
940 * -------------------------------------------------------------------------- */
941 void clusterPropagatePublish(robj *channel, robj *message) {
942 clusterSendPublish(NULL, channel, message);
943 }
944
945 /* -----------------------------------------------------------------------------
946 * CLUSTER cron job
947 * -------------------------------------------------------------------------- */
948
949 /* This is executed 1 time every second */
950 void clusterCron(void) {
951 dictIterator *di;
952 dictEntry *de;
953 int j;
954 time_t min_ping_sent = 0;
955 clusterNode *min_ping_node = NULL;
956
957 /* Check if we have disconnected nodes and reestablish the connection. */
958 di = dictGetIterator(server.cluster.nodes);
959 while((de = dictNext(di)) != NULL) {
960 clusterNode *node = dictGetVal(de);
961
962 if (node->flags & (REDIS_NODE_MYSELF|REDIS_NODE_NOADDR)) continue;
963 if (node->link == NULL) {
964 int fd;
965 clusterLink *link;
966
967 fd = anetTcpNonBlockConnect(server.neterr, node->ip,
968 node->port+REDIS_CLUSTER_PORT_INCR);
969 if (fd == -1) continue;
970 link = createClusterLink(node);
971 link->fd = fd;
972 node->link = link;
973 aeCreateFileEvent(server.el,link->fd,AE_READABLE,clusterReadHandler,link);
974 /* If the node is flagged as MEET, we send a MEET message instead
975 * of a PING one, to force the receiver to add us in its node
976 * table. */
977 clusterSendPing(link, node->flags & REDIS_NODE_MEET ?
978 CLUSTERMSG_TYPE_MEET : CLUSTERMSG_TYPE_PING);
979 /* We can clear the flag after the first packet is sent.
980 * If we'll never receive a PONG, we'll never send new packets
981 * to this node. Instead after the PONG is received and we
982 * are no longer in meet/handshake status, we want to send
983 * normal PING packets. */
984 node->flags &= ~REDIS_NODE_MEET;
985
986 redisLog(REDIS_NOTICE,"Connecting with Node %.40s at %s:%d", node->name, node->ip, node->port+REDIS_CLUSTER_PORT_INCR);
987 }
988 }
989 dictReleaseIterator(di);
990
991 /* Ping some random node. Check a few random nodes and ping the one with
992 * the oldest ping_sent time */
993 for (j = 0; j < 5; j++) {
994 de = dictGetRandomKey(server.cluster.nodes);
995 clusterNode *this = dictGetVal(de);
996
997 if (this->link == NULL) continue;
998 if (this->flags & (REDIS_NODE_MYSELF|REDIS_NODE_HANDSHAKE)) continue;
999 if (min_ping_node == NULL || min_ping_sent > this->ping_sent) {
1000 min_ping_node = this;
1001 min_ping_sent = this->ping_sent;
1002 }
1003 }
1004 if (min_ping_node) {
1005 redisLog(REDIS_DEBUG,"Pinging node %40s", min_ping_node->name);
1006 clusterSendPing(min_ping_node->link, CLUSTERMSG_TYPE_PING);
1007 }
1008
1009 /* Iterate nodes to check if we need to flag something as failing */
1010 di = dictGetIterator(server.cluster.nodes);
1011 while((de = dictNext(di)) != NULL) {
1012 clusterNode *node = dictGetVal(de);
1013 int delay;
1014
1015 if (node->flags &
1016 (REDIS_NODE_MYSELF|REDIS_NODE_NOADDR|REDIS_NODE_HANDSHAKE))
1017 continue;
1018 /* Check only if we already sent a ping and did not received
1019 * a reply yet. */
1020 if (node->ping_sent == 0 ||
1021 node->ping_sent <= node->pong_received) continue;
1022
1023 delay = time(NULL) - node->pong_received;
1024 if (delay < server.cluster.node_timeout) {
1025 /* The PFAIL condition can be reversed without external
1026 * help if it is not transitive (that is, if it does not
1027 * turn into a FAIL state).
1028 *
1029 * The FAIL condition is also reversible if there are no slaves
1030 * for this host, so no slave election should be in progress.
1031 *
1032 * TODO: consider all the implications of resurrecting a
1033 * FAIL node. */
1034 if (node->flags & REDIS_NODE_PFAIL) {
1035 node->flags &= ~REDIS_NODE_PFAIL;
1036 } else if (node->flags & REDIS_NODE_FAIL && !node->numslaves) {
1037 node->flags &= ~REDIS_NODE_FAIL;
1038 clusterUpdateState();
1039 }
1040 } else {
1041 /* Timeout reached. Set the noad se possibly failing if it is
1042 * not already in this state. */
1043 if (!(node->flags & (REDIS_NODE_PFAIL|REDIS_NODE_FAIL))) {
1044 redisLog(REDIS_DEBUG,"*** NODE %.40s possibly failing",
1045 node->name);
1046 node->flags |= REDIS_NODE_PFAIL;
1047 }
1048 }
1049 }
1050 dictReleaseIterator(di);
1051 }
1052
1053 /* -----------------------------------------------------------------------------
1054 * Slots management
1055 * -------------------------------------------------------------------------- */
1056
1057 /* Set the slot bit and return the old value. */
1058 int clusterNodeSetSlotBit(clusterNode *n, int slot) {
1059 off_t byte = slot/8;
1060 int bit = slot&7;
1061 int old = (n->slots[byte] & (1<<bit)) != 0;
1062 n->slots[byte] |= 1<<bit;
1063 return old;
1064 }
1065
1066 /* Clear the slot bit and return the old value. */
1067 int clusterNodeClearSlotBit(clusterNode *n, int slot) {
1068 off_t byte = slot/8;
1069 int bit = slot&7;
1070 int old = (n->slots[byte] & (1<<bit)) != 0;
1071 n->slots[byte] &= ~(1<<bit);
1072 return old;
1073 }
1074
1075 /* Return the slot bit from the cluster node structure. */
1076 int clusterNodeGetSlotBit(clusterNode *n, int slot) {
1077 off_t byte = slot/8;
1078 int bit = slot&7;
1079 return (n->slots[byte] & (1<<bit)) != 0;
1080 }
1081
1082 /* Add the specified slot to the list of slots that node 'n' will
1083 * serve. Return REDIS_OK if the operation ended with success.
1084 * If the slot is already assigned to another instance this is considered
1085 * an error and REDIS_ERR is returned. */
1086 int clusterAddSlot(clusterNode *n, int slot) {
1087 if (clusterNodeSetSlotBit(n,slot) != 0)
1088 return REDIS_ERR;
1089 server.cluster.slots[slot] = n;
1090 return REDIS_OK;
1091 }
1092
1093 /* Delete the specified slot marking it as unassigned.
1094 * Returns REDIS_OK if the slot was assigned, otherwise if the slot was
1095 * already unassigned REDIS_ERR is returned. */
1096 int clusterDelSlot(int slot) {
1097 clusterNode *n = server.cluster.slots[slot];
1098
1099 if (!n) return REDIS_ERR;
1100 redisAssert(clusterNodeClearSlotBit(n,slot) == 1);
1101 server.cluster.slots[slot] = NULL;
1102 return REDIS_OK;
1103 }
1104
1105 /* -----------------------------------------------------------------------------
1106 * Cluster state evaluation function
1107 * -------------------------------------------------------------------------- */
1108 void clusterUpdateState(void) {
1109 int ok = 1;
1110 int j;
1111
1112 for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) {
1113 if (server.cluster.slots[j] == NULL ||
1114 server.cluster.slots[j]->flags & (REDIS_NODE_FAIL))
1115 {
1116 ok = 0;
1117 break;
1118 }
1119 }
1120 if (ok) {
1121 if (server.cluster.state == REDIS_CLUSTER_NEEDHELP) {
1122 server.cluster.state = REDIS_CLUSTER_NEEDHELP;
1123 } else {
1124 server.cluster.state = REDIS_CLUSTER_OK;
1125 }
1126 } else {
1127 server.cluster.state = REDIS_CLUSTER_FAIL;
1128 }
1129 }
1130
1131 /* -----------------------------------------------------------------------------
1132 * CLUSTER command
1133 * -------------------------------------------------------------------------- */
1134
1135 sds clusterGenNodesDescription(void) {
1136 sds ci = sdsempty();
1137 dictIterator *di;
1138 dictEntry *de;
1139 int j, start;
1140
1141 di = dictGetIterator(server.cluster.nodes);
1142 while((de = dictNext(di)) != NULL) {
1143 clusterNode *node = dictGetVal(de);
1144
1145 /* Node coordinates */
1146 ci = sdscatprintf(ci,"%.40s %s:%d ",
1147 node->name,
1148 node->ip,
1149 node->port);
1150
1151 /* Flags */
1152 if (node->flags == 0) ci = sdscat(ci,"noflags,");
1153 if (node->flags & REDIS_NODE_MYSELF) ci = sdscat(ci,"myself,");
1154 if (node->flags & REDIS_NODE_MASTER) ci = sdscat(ci,"master,");
1155 if (node->flags & REDIS_NODE_SLAVE) ci = sdscat(ci,"slave,");
1156 if (node->flags & REDIS_NODE_PFAIL) ci = sdscat(ci,"fail?,");
1157 if (node->flags & REDIS_NODE_FAIL) ci = sdscat(ci,"fail,");
1158 if (node->flags & REDIS_NODE_HANDSHAKE) ci =sdscat(ci,"handshake,");
1159 if (node->flags & REDIS_NODE_NOADDR) ci = sdscat(ci,"noaddr,");
1160 if (ci[sdslen(ci)-1] == ',') ci[sdslen(ci)-1] = ' ';
1161
1162 /* Slave of... or just "-" */
1163 if (node->slaveof)
1164 ci = sdscatprintf(ci,"%.40s ",node->slaveof->name);
1165 else
1166 ci = sdscatprintf(ci,"- ");
1167
1168 /* Latency from the POV of this node, link status */
1169 ci = sdscatprintf(ci,"%ld %ld %s",
1170 (long) node->ping_sent,
1171 (long) node->pong_received,
1172 (node->link || node->flags & REDIS_NODE_MYSELF) ?
1173 "connected" : "disconnected");
1174
1175 /* Slots served by this instance */
1176 start = -1;
1177 for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) {
1178 int bit;
1179
1180 if ((bit = clusterNodeGetSlotBit(node,j)) != 0) {
1181 if (start == -1) start = j;
1182 }
1183 if (start != -1 && (!bit || j == REDIS_CLUSTER_SLOTS-1)) {
1184 if (j == REDIS_CLUSTER_SLOTS-1) j++;
1185
1186 if (start == j-1) {
1187 ci = sdscatprintf(ci," %d",start);
1188 } else {
1189 ci = sdscatprintf(ci," %d-%d",start,j-1);
1190 }
1191 start = -1;
1192 }
1193 }
1194
1195 /* Just for MYSELF node we also dump info about slots that
1196 * we are migrating to other instances or importing from other
1197 * instances. */
1198 if (node->flags & REDIS_NODE_MYSELF) {
1199 for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) {
1200 if (server.cluster.migrating_slots_to[j]) {
1201 ci = sdscatprintf(ci," [%d->-%.40s]",j,
1202 server.cluster.migrating_slots_to[j]->name);
1203 } else if (server.cluster.importing_slots_from[j]) {
1204 ci = sdscatprintf(ci," [%d-<-%.40s]",j,
1205 server.cluster.importing_slots_from[j]->name);
1206 }
1207 }
1208 }
1209 ci = sdscatlen(ci,"\n",1);
1210 }
1211 dictReleaseIterator(di);
1212 return ci;
1213 }
1214
1215 int getSlotOrReply(redisClient *c, robj *o) {
1216 long long slot;
1217
1218 if (getLongLongFromObject(o,&slot) != REDIS_OK ||
1219 slot < 0 || slot > REDIS_CLUSTER_SLOTS)
1220 {
1221 addReplyError(c,"Invalid or out of range slot");
1222 return -1;
1223 }
1224 return (int) slot;
1225 }
1226
1227 void clusterCommand(redisClient *c) {
1228 if (server.cluster_enabled == 0) {
1229 addReplyError(c,"This instance has cluster support disabled");
1230 return;
1231 }
1232
1233 if (!strcasecmp(c->argv[1]->ptr,"meet") && c->argc == 4) {
1234 clusterNode *n;
1235 struct sockaddr_in sa;
1236 long port;
1237
1238 /* Perform sanity checks on IP/port */
1239 if (inet_aton(c->argv[2]->ptr,&sa.sin_addr) == 0) {
1240 addReplyError(c,"Invalid IP address in MEET");
1241 return;
1242 }
1243 if (getLongFromObjectOrReply(c, c->argv[3], &port, NULL) != REDIS_OK ||
1244 port < 0 || port > (65535-REDIS_CLUSTER_PORT_INCR))
1245 {
1246 addReplyError(c,"Invalid TCP port specified");
1247 return;
1248 }
1249
1250 /* Finally add the node to the cluster with a random name, this
1251 * will get fixed in the first handshake (ping/pong). */
1252 n = createClusterNode(NULL,REDIS_NODE_HANDSHAKE|REDIS_NODE_MEET);
1253 strncpy(n->ip,inet_ntoa(sa.sin_addr),sizeof(n->ip));
1254 n->port = port;
1255 clusterAddNode(n);
1256 addReply(c,shared.ok);
1257 } else if (!strcasecmp(c->argv[1]->ptr,"nodes") && c->argc == 2) {
1258 robj *o;
1259 sds ci = clusterGenNodesDescription();
1260
1261 o = createObject(REDIS_STRING,ci);
1262 addReplyBulk(c,o);
1263 decrRefCount(o);
1264 } else if ((!strcasecmp(c->argv[1]->ptr,"addslots") ||
1265 !strcasecmp(c->argv[1]->ptr,"delslots")) && c->argc >= 3)
1266 {
1267 /* CLUSTER ADDSLOTS <slot> [slot] ... */
1268 /* CLUSTER DELSLOTS <slot> [slot] ... */
1269 int j, slot;
1270 unsigned char *slots = zmalloc(REDIS_CLUSTER_SLOTS);
1271 int del = !strcasecmp(c->argv[1]->ptr,"delslots");
1272
1273 memset(slots,0,REDIS_CLUSTER_SLOTS);
1274 /* Check that all the arguments are parsable and that all the
1275 * slots are not already busy. */
1276 for (j = 2; j < c->argc; j++) {
1277 if ((slot = getSlotOrReply(c,c->argv[j])) == -1) {
1278 zfree(slots);
1279 return;
1280 }
1281 if (del && server.cluster.slots[slot] == NULL) {
1282 addReplyErrorFormat(c,"Slot %d is already unassigned", slot);
1283 zfree(slots);
1284 return;
1285 } else if (!del && server.cluster.slots[slot]) {
1286 addReplyErrorFormat(c,"Slot %d is already busy", slot);
1287 zfree(slots);
1288 return;
1289 }
1290 if (slots[slot]++ == 1) {
1291 addReplyErrorFormat(c,"Slot %d specified multiple times",
1292 (int)slot);
1293 zfree(slots);
1294 return;
1295 }
1296 }
1297 for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) {
1298 if (slots[j]) {
1299 int retval;
1300
1301 /* If this slot was set as importing we can clear this
1302 * state as now we are the real owner of the slot. */
1303 if (server.cluster.importing_slots_from[j])
1304 server.cluster.importing_slots_from[j] = NULL;
1305
1306 retval = del ? clusterDelSlot(j) :
1307 clusterAddSlot(server.cluster.myself,j);
1308 redisAssertWithInfo(c,NULL,retval == REDIS_OK);
1309 }
1310 }
1311 zfree(slots);
1312 clusterUpdateState();
1313 clusterSaveConfigOrDie();
1314 addReply(c,shared.ok);
1315 } else if (!strcasecmp(c->argv[1]->ptr,"setslot") && c->argc >= 4) {
1316 /* SETSLOT 10 MIGRATING <node ID> */
1317 /* SETSLOT 10 IMPORTING <node ID> */
1318 /* SETSLOT 10 STABLE */
1319 /* SETSLOT 10 NODE <node ID> */
1320 int slot;
1321 clusterNode *n;
1322
1323 if ((slot = getSlotOrReply(c,c->argv[2])) == -1) return;
1324
1325 if (!strcasecmp(c->argv[3]->ptr,"migrating") && c->argc == 5) {
1326 if (server.cluster.slots[slot] != server.cluster.myself) {
1327 addReplyErrorFormat(c,"I'm not the owner of hash slot %u",slot);
1328 return;
1329 }
1330 if ((n = clusterLookupNode(c->argv[4]->ptr)) == NULL) {
1331 addReplyErrorFormat(c,"I don't know about node %s",
1332 (char*)c->argv[4]->ptr);
1333 return;
1334 }
1335 server.cluster.migrating_slots_to[slot] = n;
1336 } else if (!strcasecmp(c->argv[3]->ptr,"importing") && c->argc == 5) {
1337 if (server.cluster.slots[slot] == server.cluster.myself) {
1338 addReplyErrorFormat(c,
1339 "I'm already the owner of hash slot %u",slot);
1340 return;
1341 }
1342 if ((n = clusterLookupNode(c->argv[4]->ptr)) == NULL) {
1343 addReplyErrorFormat(c,"I don't know about node %s",
1344 (char*)c->argv[3]->ptr);
1345 return;
1346 }
1347 server.cluster.importing_slots_from[slot] = n;
1348 } else if (!strcasecmp(c->argv[3]->ptr,"stable") && c->argc == 4) {
1349 /* CLUSTER SETSLOT <SLOT> STABLE */
1350 server.cluster.importing_slots_from[slot] = NULL;
1351 server.cluster.migrating_slots_to[slot] = NULL;
1352 } else if (!strcasecmp(c->argv[3]->ptr,"node") && c->argc == 5) {
1353 /* CLUSTER SETSLOT <SLOT> NODE <NODE ID> */
1354 clusterNode *n = clusterLookupNode(c->argv[4]->ptr);
1355
1356 if (!n) addReplyErrorFormat(c,"Unknown node %s",
1357 (char*)c->argv[4]->ptr);
1358 /* If this hash slot was served by 'myself' before to switch
1359 * make sure there are no longer local keys for this hash slot. */
1360 if (server.cluster.slots[slot] == server.cluster.myself &&
1361 n != server.cluster.myself)
1362 {
1363 int numkeys;
1364 robj **keys;
1365
1366 keys = zmalloc(sizeof(robj*)*1);
1367 numkeys = GetKeysInSlot(slot, keys, 1);
1368 zfree(keys);
1369 if (numkeys != 0) {
1370 addReplyErrorFormat(c, "Can't assign hashslot %d to a different node while I still hold keys for this hash slot.", slot);
1371 return;
1372 }
1373 }
1374 /* If this node was the slot owner and the slot was marked as
1375 * migrating, assigning the slot to another node will clear
1376 * the migratig status. */
1377 if (server.cluster.slots[slot] == server.cluster.myself &&
1378 server.cluster.migrating_slots_to[slot])
1379 server.cluster.migrating_slots_to[slot] = NULL;
1380
1381 /* If this node was importing this slot, assigning the slot to
1382 * itself also clears the importing status. */
1383 if (n == server.cluster.myself && server.cluster.importing_slots_from[slot])
1384 server.cluster.importing_slots_from[slot] = NULL;
1385
1386 clusterDelSlot(slot);
1387 clusterAddSlot(n,slot);
1388 } else {
1389 addReplyError(c,"Invalid CLUSTER SETSLOT action or number of arguments");
1390 return;
1391 }
1392 clusterSaveConfigOrDie();
1393 addReply(c,shared.ok);
1394 } else if (!strcasecmp(c->argv[1]->ptr,"info") && c->argc == 2) {
1395 char *statestr[] = {"ok","fail","needhelp"};
1396 int slots_assigned = 0, slots_ok = 0, slots_pfail = 0, slots_fail = 0;
1397 int j;
1398
1399 for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) {
1400 clusterNode *n = server.cluster.slots[j];
1401
1402 if (n == NULL) continue;
1403 slots_assigned++;
1404 if (n->flags & REDIS_NODE_FAIL) {
1405 slots_fail++;
1406 } else if (n->flags & REDIS_NODE_PFAIL) {
1407 slots_pfail++;
1408 } else {
1409 slots_ok++;
1410 }
1411 }
1412
1413 sds info = sdscatprintf(sdsempty(),
1414 "cluster_state:%s\r\n"
1415 "cluster_slots_assigned:%d\r\n"
1416 "cluster_slots_ok:%d\r\n"
1417 "cluster_slots_pfail:%d\r\n"
1418 "cluster_slots_fail:%d\r\n"
1419 "cluster_known_nodes:%lu\r\n"
1420 , statestr[server.cluster.state],
1421 slots_assigned,
1422 slots_ok,
1423 slots_pfail,
1424 slots_fail,
1425 dictSize(server.cluster.nodes)
1426 );
1427 addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",
1428 (unsigned long)sdslen(info)));
1429 addReplySds(c,info);
1430 addReply(c,shared.crlf);
1431 } else if (!strcasecmp(c->argv[1]->ptr,"keyslot") && c->argc == 3) {
1432 sds key = c->argv[2]->ptr;
1433
1434 addReplyLongLong(c,keyHashSlot(key,sdslen(key)));
1435 } else if (!strcasecmp(c->argv[1]->ptr,"getkeysinslot") && c->argc == 4) {
1436 long long maxkeys, slot;
1437 unsigned int numkeys, j;
1438 robj **keys;
1439
1440 if (getLongLongFromObjectOrReply(c,c->argv[2],&slot,NULL) != REDIS_OK)
1441 return;
1442 if (getLongLongFromObjectOrReply(c,c->argv[3],&maxkeys,NULL) != REDIS_OK)
1443 return;
1444 if (slot < 0 || slot >= REDIS_CLUSTER_SLOTS || maxkeys < 0 ||
1445 maxkeys > 1024*1024) {
1446 addReplyError(c,"Invalid slot or number of keys");
1447 return;
1448 }
1449
1450 keys = zmalloc(sizeof(robj*)*maxkeys);
1451 numkeys = GetKeysInSlot(slot, keys, maxkeys);
1452 addReplyMultiBulkLen(c,numkeys);
1453 for (j = 0; j < numkeys; j++) addReplyBulk(c,keys[j]);
1454 zfree(keys);
1455 } else {
1456 addReplyError(c,"Wrong CLUSTER subcommand or number of arguments");
1457 }
1458 }
1459
1460 /* -----------------------------------------------------------------------------
1461 * DUMP, RESTORE and MIGRATE commands
1462 * -------------------------------------------------------------------------- */
1463
1464 /* Generates a DUMP-format representation of the object 'o', adding it to the
1465 * io stream pointed by 'rio'. This function can't fail. */
1466 void createDumpPayload(rio *payload, robj *o) {
1467 unsigned char buf[2];
1468 uint64_t crc;
1469
1470 /* Serialize the object in a RDB-like format. It consist of an object type
1471 * byte followed by the serialized object. This is understood by RESTORE. */
1472 rioInitWithBuffer(payload,sdsempty());
1473 redisAssert(rdbSaveObjectType(payload,o));
1474 redisAssert(rdbSaveObject(payload,o));
1475
1476 /* Write the footer, this is how it looks like:
1477 * ----------------+---------------------+---------------+
1478 * ... RDB payload | 2 bytes RDB version | 8 bytes CRC64 |
1479 * ----------------+---------------------+---------------+
1480 * RDB version and CRC are both in little endian.
1481 */
1482
1483 /* RDB version */
1484 buf[0] = REDIS_RDB_VERSION & 0xff;
1485 buf[1] = (REDIS_RDB_VERSION >> 8) & 0xff;
1486 payload->io.buffer.ptr = sdscatlen(payload->io.buffer.ptr,buf,2);
1487
1488 /* CRC64 */
1489 crc = crc64(0,(unsigned char*)payload->io.buffer.ptr,
1490 sdslen(payload->io.buffer.ptr));
1491 memrev64ifbe(&crc);
1492 payload->io.buffer.ptr = sdscatlen(payload->io.buffer.ptr,&crc,8);
1493 }
1494
1495 /* Verify that the RDB version of the dump payload matches the one of this Redis
1496 * instance and that the checksum is ok.
1497 * If the DUMP payload looks valid REDIS_OK is returned, otherwise REDIS_ERR
1498 * is returned. */
1499 int verifyDumpPayload(unsigned char *p, size_t len) {
1500 unsigned char *footer;
1501 uint16_t rdbver;
1502 uint64_t crc;
1503
1504 /* At least 2 bytes of RDB version and 8 of CRC64 should be present. */
1505 if (len < 10) return REDIS_ERR;
1506 footer = p+(len-10);
1507
1508 /* Verify RDB version */
1509 rdbver = (footer[1] << 8) | footer[0];
1510 if (rdbver != REDIS_RDB_VERSION) return REDIS_ERR;
1511
1512 /* Verify CRC64 */
1513 crc = crc64(0,p,len-8);
1514 memrev64ifbe(&crc);
1515 return (memcmp(&crc,footer+2,8) == 0) ? REDIS_OK : REDIS_ERR;
1516 }
1517
1518 /* DUMP keyname
1519 * DUMP is actually not used by Redis Cluster but it is the obvious
1520 * complement of RESTORE and can be useful for different applications. */
1521 void dumpCommand(redisClient *c) {
1522 robj *o, *dumpobj;
1523 rio payload;
1524
1525 /* Check if the key is here. */
1526 if ((o = lookupKeyRead(c->db,c->argv[1])) == NULL) {
1527 addReply(c,shared.nullbulk);
1528 return;
1529 }
1530
1531 /* Create the DUMP encoded representation. */
1532 createDumpPayload(&payload,o);
1533
1534 /* Transfer to the client */
1535 dumpobj = createObject(REDIS_STRING,payload.io.buffer.ptr);
1536 addReplyBulk(c,dumpobj);
1537 decrRefCount(dumpobj);
1538 return;
1539 }
1540
1541 /* RESTORE key ttl serialized-value */
1542 void restoreCommand(redisClient *c) {
1543 long ttl;
1544 rio payload;
1545 int type;
1546 robj *obj;
1547
1548 /* Make sure this key does not already exist here... */
1549 if (lookupKeyWrite(c->db,c->argv[1]) != NULL) {
1550 addReplyError(c,"Target key name is busy.");
1551 return;
1552 }
1553
1554 /* Check if the TTL value makes sense */
1555 if (getLongFromObjectOrReply(c,c->argv[2],&ttl,NULL) != REDIS_OK) {
1556 return;
1557 } else if (ttl < 0) {
1558 addReplyError(c,"Invalid TTL value, must be >= 0");
1559 return;
1560 }
1561
1562 /* Verify RDB version and data checksum. */
1563 if (verifyDumpPayload(c->argv[3]->ptr,sdslen(c->argv[3]->ptr)) == REDIS_ERR) {
1564 addReplyError(c,"DUMP payload version or checksum are wrong");
1565 return;
1566 }
1567
1568 rioInitWithBuffer(&payload,c->argv[3]->ptr);
1569 if (((type = rdbLoadObjectType(&payload)) == -1) ||
1570 ((obj = rdbLoadObject(type,&payload)) == NULL))
1571 {
1572 addReplyError(c,"Bad data format");
1573 return;
1574 }
1575
1576 /* Create the key and set the TTL if any */
1577 dbAdd(c->db,c->argv[1],obj);
1578 if (ttl) setExpire(c->db,c->argv[1],mstime()+ttl);
1579 signalModifiedKey(c->db,c->argv[1]);
1580 addReply(c,shared.ok);
1581 server.dirty++;
1582 }
1583
1584 /* MIGRATE host port key dbid timeout */
1585 void migrateCommand(redisClient *c) {
1586 int fd;
1587 long timeout;
1588 long dbid;
1589 long long ttl = 0, expireat;
1590 robj *o;
1591 rio cmd, payload;
1592
1593 /* Sanity check */
1594 if (getLongFromObjectOrReply(c,c->argv[5],&timeout,NULL) != REDIS_OK)
1595 return;
1596 if (getLongFromObjectOrReply(c,c->argv[4],&dbid,NULL) != REDIS_OK)
1597 return;
1598 if (timeout <= 0) timeout = 1;
1599
1600 /* Check if the key is here. If not we reply with success as there is
1601 * nothing to migrate (for instance the key expired in the meantime), but
1602 * we include such information in the reply string. */
1603 if ((o = lookupKeyRead(c->db,c->argv[3])) == NULL) {
1604 addReplySds(c,sdsnew("+NOKEY\r\n"));
1605 return;
1606 }
1607
1608 /* Connect */
1609 fd = anetTcpNonBlockConnect(server.neterr,c->argv[1]->ptr,
1610 atoi(c->argv[2]->ptr));
1611 if (fd == -1) {
1612 addReplyErrorFormat(c,"Can't connect to target node: %s",
1613 server.neterr);
1614 return;
1615 }
1616 if ((aeWait(fd,AE_WRITABLE,timeout*1000) & AE_WRITABLE) == 0) {
1617 addReplySds(c,sdsnew("-IOERR error or timeout connecting to the client\r\n"));
1618 return;
1619 }
1620
1621 /* Create RESTORE payload and generate the protocol to call the command. */
1622 rioInitWithBuffer(&cmd,sdsempty());
1623 redisAssertWithInfo(c,NULL,rioWriteBulkCount(&cmd,'*',2));
1624 redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,"SELECT",6));
1625 redisAssertWithInfo(c,NULL,rioWriteBulkLongLong(&cmd,dbid));
1626
1627 expireat = getExpire(c->db,c->argv[3]);
1628 if (expireat != -1) {
1629 ttl = expireat-mstime();
1630 if (ttl < 1) ttl = 1;
1631 }
1632 redisAssertWithInfo(c,NULL,rioWriteBulkCount(&cmd,'*',4));
1633 redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,"RESTORE",7));
1634 redisAssertWithInfo(c,NULL,c->argv[3]->encoding == REDIS_ENCODING_RAW);
1635 redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,c->argv[3]->ptr,sdslen(c->argv[3]->ptr)));
1636 redisAssertWithInfo(c,NULL,rioWriteBulkLongLong(&cmd,ttl));
1637
1638 /* Finally the last argument that is the serailized object payload
1639 * in the DUMP format. */
1640 createDumpPayload(&payload,o);
1641 redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,payload.io.buffer.ptr,
1642 sdslen(payload.io.buffer.ptr)));
1643 sdsfree(payload.io.buffer.ptr);
1644
1645 /* Tranfer the query to the other node in 64K chunks. */
1646 {
1647 sds buf = cmd.io.buffer.ptr;
1648 size_t pos = 0, towrite;
1649 int nwritten = 0;
1650
1651 while ((towrite = sdslen(buf)-pos) > 0) {
1652 towrite = (towrite > (64*1024) ? (64*1024) : towrite);
1653 nwritten = syncWrite(fd,buf+pos,towrite,timeout);
1654 if (nwritten != (signed)towrite) goto socket_wr_err;
1655 pos += nwritten;
1656 }
1657 }
1658
1659 /* Read back the reply. */
1660 {
1661 char buf1[1024];
1662 char buf2[1024];
1663
1664 /* Read the two replies */
1665 if (syncReadLine(fd, buf1, sizeof(buf1), timeout) <= 0)
1666 goto socket_rd_err;
1667 if (syncReadLine(fd, buf2, sizeof(buf2), timeout) <= 0)
1668 goto socket_rd_err;
1669 if (buf1[0] == '-' || buf2[0] == '-') {
1670 addReplyErrorFormat(c,"Target instance replied with error: %s",
1671 (buf1[0] == '-') ? buf1+1 : buf2+1);
1672 } else {
1673 robj *aux;
1674
1675 dbDelete(c->db,c->argv[3]);
1676 signalModifiedKey(c->db,c->argv[3]);
1677 addReply(c,shared.ok);
1678 server.dirty++;
1679
1680 /* Translate MIGRATE as DEL for replication/AOF. */
1681 aux = createStringObject("DEL",3);
1682 rewriteClientCommandVector(c,2,aux,c->argv[3]);
1683 decrRefCount(aux);
1684 }
1685 }
1686
1687 sdsfree(cmd.io.buffer.ptr);
1688 close(fd);
1689 return;
1690
1691 socket_wr_err:
1692 addReplySds(c,sdsnew("-IOERR error or timeout writing to target instance\r\n"));
1693 sdsfree(cmd.io.buffer.ptr);
1694 close(fd);
1695 return;
1696
1697 socket_rd_err:
1698 addReplySds(c,sdsnew("-IOERR error or timeout reading from target node\r\n"));
1699 sdsfree(cmd.io.buffer.ptr);
1700 close(fd);
1701 return;
1702 }
1703
1704 /* The ASKING command is required after a -ASK redirection.
1705 * The client should issue ASKING before to actualy send the command to
1706 * the target instance. See the Redis Cluster specification for more
1707 * information. */
1708 void askingCommand(redisClient *c) {
1709 if (server.cluster_enabled == 0) {
1710 addReplyError(c,"This instance has cluster support disabled");
1711 return;
1712 }
1713 c->flags |= REDIS_ASKING;
1714 addReply(c,shared.ok);
1715 }
1716
1717 /* -----------------------------------------------------------------------------
1718 * Cluster functions related to serving / redirecting clients
1719 * -------------------------------------------------------------------------- */
1720
1721 /* Return the pointer to the cluster node that is able to serve the query
1722 * as all the keys belong to hash slots for which the node is in charge.
1723 *
1724 * If the returned node should be used only for this request, the *ask
1725 * integer is set to '1', otherwise to '0'. This is used in order to
1726 * let the caller know if we should reply with -MOVED or with -ASK.
1727 *
1728 * If the request contains more than a single key NULL is returned,
1729 * however a request with more then a key argument where the key is always
1730 * the same is valid, like in: RPOPLPUSH mylist mylist.*/
1731 clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **argv, int argc, int *hashslot, int *ask) {
1732 clusterNode *n = NULL;
1733 robj *firstkey = NULL;
1734 multiState *ms, _ms;
1735 multiCmd mc;
1736 int i, slot = 0;
1737
1738 /* We handle all the cases as if they were EXEC commands, so we have
1739 * a common code path for everything */
1740 if (cmd->proc == execCommand) {
1741 /* If REDIS_MULTI flag is not set EXEC is just going to return an
1742 * error. */
1743 if (!(c->flags & REDIS_MULTI)) return server.cluster.myself;
1744 ms = &c->mstate;
1745 } else {
1746 /* In order to have a single codepath create a fake Multi State
1747 * structure if the client is not in MULTI/EXEC state, this way
1748 * we have a single codepath below. */
1749 ms = &_ms;
1750 _ms.commands = &mc;
1751 _ms.count = 1;
1752 mc.argv = argv;
1753 mc.argc = argc;
1754 mc.cmd = cmd;
1755 }
1756
1757 /* Check that all the keys are the same key, and get the slot and
1758 * node for this key. */
1759 for (i = 0; i < ms->count; i++) {
1760 struct redisCommand *mcmd;
1761 robj **margv;
1762 int margc, *keyindex, numkeys, j;
1763
1764 mcmd = ms->commands[i].cmd;
1765 margc = ms->commands[i].argc;
1766 margv = ms->commands[i].argv;
1767
1768 keyindex = getKeysFromCommand(mcmd,margv,margc,&numkeys,
1769 REDIS_GETKEYS_ALL);
1770 for (j = 0; j < numkeys; j++) {
1771 if (firstkey == NULL) {
1772 /* This is the first key we see. Check what is the slot
1773 * and node. */
1774 firstkey = margv[keyindex[j]];
1775
1776 slot = keyHashSlot((char*)firstkey->ptr, sdslen(firstkey->ptr));
1777 n = server.cluster.slots[slot];
1778 redisAssertWithInfo(c,firstkey,n != NULL);
1779 } else {
1780 /* If it is not the first key, make sure it is exactly
1781 * the same key as the first we saw. */
1782 if (!equalStringObjects(firstkey,margv[keyindex[j]])) {
1783 decrRefCount(firstkey);
1784 getKeysFreeResult(keyindex);
1785 return NULL;
1786 }
1787 }
1788 }
1789 getKeysFreeResult(keyindex);
1790 }
1791 if (ask) *ask = 0; /* This is the default. Set to 1 if needed later. */
1792 /* No key at all in command? then we can serve the request
1793 * without redirections. */
1794 if (n == NULL) return server.cluster.myself;
1795 if (hashslot) *hashslot = slot;
1796 /* This request is about a slot we are migrating into another instance?
1797 * Then we need to check if we have the key. If we have it we can reply.
1798 * If instead is a new key, we pass the request to the node that is
1799 * receiving the slot. */
1800 if (n == server.cluster.myself &&
1801 server.cluster.migrating_slots_to[slot] != NULL)
1802 {
1803 if (lookupKeyRead(&server.db[0],firstkey) == NULL) {
1804 if (ask) *ask = 1;
1805 return server.cluster.migrating_slots_to[slot];
1806 }
1807 }
1808 /* Handle the case in which we are receiving this hash slot from
1809 * another instance, so we'll accept the query even if in the table
1810 * it is assigned to a different node, but only if the client
1811 * issued an ASKING command before. */
1812 if (server.cluster.importing_slots_from[slot] != NULL &&
1813 c->flags & REDIS_ASKING) {
1814 return server.cluster.myself;
1815 }
1816 /* It's not a -ASK case. Base case: just return the right node. */
1817 return n;
1818 }