xref: /f-stack/app/redis-5.0.5/src/multi.c (revision 572c4311)
1 /*
2  * Copyright (c) 2009-2012, Salvatore Sanfilippo <antirez at gmail dot com>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions are met:
7  *
8  *   * Redistributions of source code must retain the above copyright notice,
9  *     this list of conditions and the following disclaimer.
10  *   * Redistributions in binary form must reproduce the above copyright
11  *     notice, this list of conditions and the following disclaimer in the
12  *     documentation and/or other materials provided with the distribution.
13  *   * Neither the name of Redis nor the names of its contributors may be used
14  *     to endorse or promote products derived from this software without
15  *     specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
21  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27  * POSSIBILITY OF SUCH DAMAGE.
28  */
29 
30 #include "server.h"
31 
32 /* ================================ MULTI/EXEC ============================== */
33 
34 /* Client state initialization for MULTI/EXEC */
initClientMultiState(client * c)35 void initClientMultiState(client *c) {
36     c->mstate.commands = NULL;
37     c->mstate.count = 0;
38     c->mstate.cmd_flags = 0;
39 }
40 
41 /* Release all the resources associated with MULTI/EXEC state */
freeClientMultiState(client * c)42 void freeClientMultiState(client *c) {
43     int j;
44 
45     for (j = 0; j < c->mstate.count; j++) {
46         int i;
47         multiCmd *mc = c->mstate.commands+j;
48 
49         for (i = 0; i < mc->argc; i++)
50             decrRefCount(mc->argv[i]);
51         zfree(mc->argv);
52     }
53     zfree(c->mstate.commands);
54 }
55 
56 /* Add a new command into the MULTI commands queue */
queueMultiCommand(client * c)57 void queueMultiCommand(client *c) {
58     multiCmd *mc;
59     int j;
60 
61     c->mstate.commands = zrealloc(c->mstate.commands,
62             sizeof(multiCmd)*(c->mstate.count+1));
63     mc = c->mstate.commands+c->mstate.count;
64     mc->cmd = c->cmd;
65     mc->argc = c->argc;
66     mc->argv = zmalloc(sizeof(robj*)*c->argc);
67     memcpy(mc->argv,c->argv,sizeof(robj*)*c->argc);
68     for (j = 0; j < c->argc; j++)
69         incrRefCount(mc->argv[j]);
70     c->mstate.count++;
71     c->mstate.cmd_flags |= c->cmd->flags;
72 }
73 
discardTransaction(client * c)74 void discardTransaction(client *c) {
75     freeClientMultiState(c);
76     initClientMultiState(c);
77     c->flags &= ~(CLIENT_MULTI|CLIENT_DIRTY_CAS|CLIENT_DIRTY_EXEC);
78     unwatchAllKeys(c);
79 }
80 
81 /* Flag the transacation as DIRTY_EXEC so that EXEC will fail.
82  * Should be called every time there is an error while queueing a command. */
flagTransaction(client * c)83 void flagTransaction(client *c) {
84     if (c->flags & CLIENT_MULTI)
85         c->flags |= CLIENT_DIRTY_EXEC;
86 }
87 
multiCommand(client * c)88 void multiCommand(client *c) {
89     if (c->flags & CLIENT_MULTI) {
90         addReplyError(c,"MULTI calls can not be nested");
91         return;
92     }
93     c->flags |= CLIENT_MULTI;
94     addReply(c,shared.ok);
95 }
96 
discardCommand(client * c)97 void discardCommand(client *c) {
98     if (!(c->flags & CLIENT_MULTI)) {
99         addReplyError(c,"DISCARD without MULTI");
100         return;
101     }
102     discardTransaction(c);
103     addReply(c,shared.ok);
104 }
105 
106 /* Send a MULTI command to all the slaves and AOF file. Check the execCommand
107  * implementation for more information. */
execCommandPropagateMulti(client * c)108 void execCommandPropagateMulti(client *c) {
109     robj *multistring = createStringObject("MULTI",5);
110 
111     propagate(server.multiCommand,c->db->id,&multistring,1,
112               PROPAGATE_AOF|PROPAGATE_REPL);
113     decrRefCount(multistring);
114 }
115 
execCommand(client * c)116 void execCommand(client *c) {
117     int j;
118     robj **orig_argv;
119     int orig_argc;
120     struct redisCommand *orig_cmd;
121     int must_propagate = 0; /* Need to propagate MULTI/EXEC to AOF / slaves? */
122     int was_master = server.masterhost == NULL;
123 
124     if (!(c->flags & CLIENT_MULTI)) {
125         addReplyError(c,"EXEC without MULTI");
126         return;
127     }
128 
129     /* Check if we need to abort the EXEC because:
130      * 1) Some WATCHed key was touched.
131      * 2) There was a previous error while queueing commands.
132      * A failed EXEC in the first case returns a multi bulk nil object
133      * (technically it is not an error but a special behavior), while
134      * in the second an EXECABORT error is returned. */
135     if (c->flags & (CLIENT_DIRTY_CAS|CLIENT_DIRTY_EXEC)) {
136         addReply(c, c->flags & CLIENT_DIRTY_EXEC ? shared.execaborterr :
137                                                   shared.nullmultibulk);
138         discardTransaction(c);
139         goto handle_monitor;
140     }
141 
142     /* If there are write commands inside the transaction, and this is a read
143      * only slave, we want to send an error. This happens when the transaction
144      * was initiated when the instance was a master or a writable replica and
145      * then the configuration changed (for example instance was turned into
146      * a replica). */
147     if (!server.loading && server.masterhost && server.repl_slave_ro &&
148         !(c->flags & CLIENT_MASTER) && c->mstate.cmd_flags & CMD_WRITE)
149     {
150         addReplyError(c,
151             "Transaction contains write commands but instance "
152             "is now a read-only slave. EXEC aborted.");
153         discardTransaction(c);
154         goto handle_monitor;
155     }
156 
157     /* Exec all the queued commands */
158     unwatchAllKeys(c); /* Unwatch ASAP otherwise we'll waste CPU cycles */
159     orig_argv = c->argv;
160     orig_argc = c->argc;
161     orig_cmd = c->cmd;
162     addReplyMultiBulkLen(c,c->mstate.count);
163     for (j = 0; j < c->mstate.count; j++) {
164         c->argc = c->mstate.commands[j].argc;
165         c->argv = c->mstate.commands[j].argv;
166         c->cmd = c->mstate.commands[j].cmd;
167 
168         /* Propagate a MULTI request once we encounter the first command which
169          * is not readonly nor an administrative one.
170          * This way we'll deliver the MULTI/..../EXEC block as a whole and
171          * both the AOF and the replication link will have the same consistency
172          * and atomicity guarantees. */
173         if (!must_propagate && !(c->cmd->flags & (CMD_READONLY|CMD_ADMIN))) {
174             execCommandPropagateMulti(c);
175             must_propagate = 1;
176         }
177 
178         call(c,server.loading ? CMD_CALL_NONE : CMD_CALL_FULL);
179 
180         /* Commands may alter argc/argv, restore mstate. */
181         c->mstate.commands[j].argc = c->argc;
182         c->mstate.commands[j].argv = c->argv;
183         c->mstate.commands[j].cmd = c->cmd;
184     }
185     c->argv = orig_argv;
186     c->argc = orig_argc;
187     c->cmd = orig_cmd;
188     discardTransaction(c);
189 
190     /* Make sure the EXEC command will be propagated as well if MULTI
191      * was already propagated. */
192     if (must_propagate) {
193         int is_master = server.masterhost == NULL;
194         server.dirty++;
195         /* If inside the MULTI/EXEC block this instance was suddenly
196          * switched from master to slave (using the SLAVEOF command), the
197          * initial MULTI was propagated into the replication backlog, but the
198          * rest was not. We need to make sure to at least terminate the
199          * backlog with the final EXEC. */
200         if (server.repl_backlog && was_master && !is_master) {
201             char *execcmd = "*1\r\n$4\r\nEXEC\r\n";
202             feedReplicationBacklog(execcmd,strlen(execcmd));
203         }
204     }
205 
206 handle_monitor:
207     /* Send EXEC to clients waiting data from MONITOR. We do it here
208      * since the natural order of commands execution is actually:
209      * MUTLI, EXEC, ... commands inside transaction ...
210      * Instead EXEC is flagged as CMD_SKIP_MONITOR in the command
211      * table, and we do it here with correct ordering. */
212     if (listLength(server.monitors) && !server.loading)
213         replicationFeedMonitors(c,server.monitors,c->db->id,c->argv,c->argc);
214 }
215 
216 /* ===================== WATCH (CAS alike for MULTI/EXEC) ===================
217  *
218  * The implementation uses a per-DB hash table mapping keys to list of clients
219  * WATCHing those keys, so that given a key that is going to be modified
220  * we can mark all the associated clients as dirty.
221  *
222  * Also every client contains a list of WATCHed keys so that's possible to
223  * un-watch such keys when the client is freed or when UNWATCH is called. */
224 
225 /* In the client->watched_keys list we need to use watchedKey structures
226  * as in order to identify a key in Redis we need both the key name and the
227  * DB */
228 typedef struct watchedKey {
229     robj *key;
230     redisDb *db;
231 } watchedKey;
232 
233 /* Watch for the specified key */
watchForKey(client * c,robj * key)234 void watchForKey(client *c, robj *key) {
235     list *clients = NULL;
236     listIter li;
237     listNode *ln;
238     watchedKey *wk;
239 
240     /* Check if we are already watching for this key */
241     listRewind(c->watched_keys,&li);
242     while((ln = listNext(&li))) {
243         wk = listNodeValue(ln);
244         if (wk->db == c->db && equalStringObjects(key,wk->key))
245             return; /* Key already watched */
246     }
247     /* This key is not already watched in this DB. Let's add it */
248     clients = dictFetchValue(c->db->watched_keys,key);
249     if (!clients) {
250         clients = listCreate();
251         dictAdd(c->db->watched_keys,key,clients);
252         incrRefCount(key);
253     }
254     listAddNodeTail(clients,c);
255     /* Add the new key to the list of keys watched by this client */
256     wk = zmalloc(sizeof(*wk));
257     wk->key = key;
258     wk->db = c->db;
259     incrRefCount(key);
260     listAddNodeTail(c->watched_keys,wk);
261 }
262 
263 /* Unwatch all the keys watched by this client. To clean the EXEC dirty
264  * flag is up to the caller. */
unwatchAllKeys(client * c)265 void unwatchAllKeys(client *c) {
266     listIter li;
267     listNode *ln;
268 
269     if (listLength(c->watched_keys) == 0) return;
270     listRewind(c->watched_keys,&li);
271     while((ln = listNext(&li))) {
272         list *clients;
273         watchedKey *wk;
274 
275         /* Lookup the watched key -> clients list and remove the client
276          * from the list */
277         wk = listNodeValue(ln);
278         clients = dictFetchValue(wk->db->watched_keys, wk->key);
279         serverAssertWithInfo(c,NULL,clients != NULL);
280         listDelNode(clients,listSearchKey(clients,c));
281         /* Kill the entry at all if this was the only client */
282         if (listLength(clients) == 0)
283             dictDelete(wk->db->watched_keys, wk->key);
284         /* Remove this watched key from the client->watched list */
285         listDelNode(c->watched_keys,ln);
286         decrRefCount(wk->key);
287         zfree(wk);
288     }
289 }
290 
291 /* "Touch" a key, so that if this key is being WATCHed by some client the
292  * next EXEC will fail. */
touchWatchedKey(redisDb * db,robj * key)293 void touchWatchedKey(redisDb *db, robj *key) {
294     list *clients;
295     listIter li;
296     listNode *ln;
297 
298     if (dictSize(db->watched_keys) == 0) return;
299     clients = dictFetchValue(db->watched_keys, key);
300     if (!clients) return;
301 
302     /* Mark all the clients watching this key as CLIENT_DIRTY_CAS */
303     /* Check if we are already watching for this key */
304     listRewind(clients,&li);
305     while((ln = listNext(&li))) {
306         client *c = listNodeValue(ln);
307 
308         c->flags |= CLIENT_DIRTY_CAS;
309     }
310 }
311 
312 /* On FLUSHDB or FLUSHALL all the watched keys that are present before the
313  * flush but will be deleted as effect of the flushing operation should
314  * be touched. "dbid" is the DB that's getting the flush. -1 if it is
315  * a FLUSHALL operation (all the DBs flushed). */
touchWatchedKeysOnFlush(int dbid)316 void touchWatchedKeysOnFlush(int dbid) {
317     listIter li1, li2;
318     listNode *ln;
319 
320     /* For every client, check all the waited keys */
321     listRewind(server.clients,&li1);
322     while((ln = listNext(&li1))) {
323         client *c = listNodeValue(ln);
324         listRewind(c->watched_keys,&li2);
325         while((ln = listNext(&li2))) {
326             watchedKey *wk = listNodeValue(ln);
327 
328             /* For every watched key matching the specified DB, if the
329              * key exists, mark the client as dirty, as the key will be
330              * removed. */
331             if (dbid == -1 || wk->db->id == dbid) {
332                 if (dictFind(wk->db->dict, wk->key->ptr) != NULL)
333                     c->flags |= CLIENT_DIRTY_CAS;
334             }
335         }
336     }
337 }
338 
watchCommand(client * c)339 void watchCommand(client *c) {
340     int j;
341 
342     if (c->flags & CLIENT_MULTI) {
343         addReplyError(c,"WATCH inside MULTI is not allowed");
344         return;
345     }
346     for (j = 1; j < c->argc; j++)
347         watchForKey(c,c->argv[j]);
348     addReply(c,shared.ok);
349 }
350 
unwatchCommand(client * c)351 void unwatchCommand(client *c) {
352     unwatchAllKeys(c);
353     c->flags &= (~CLIENT_DIRTY_CAS);
354     addReply(c,shared.ok);
355 }
356