]>
Commit | Line | Data |
---|---|---|
1 | #include "redis.h" | |
2 | #include "pqsort.h" /* Partial qsort for SORT+LIMIT */ | |
3 | #include <math.h> /* isnan() */ | |
4 | ||
5 | redisSortOperation *createSortOperation(int type, robj *pattern) { | |
6 | redisSortOperation *so = zmalloc(sizeof(*so)); | |
7 | so->type = type; | |
8 | so->pattern = pattern; | |
9 | return so; | |
10 | } | |
11 | ||
12 | /* Return the value associated to the key with a name obtained using | |
13 | * the following rules: | |
14 | * | |
15 | * 1) The first occurence of '*' in 'pattern' is substituted with 'subst'. | |
16 | * | |
17 | * 2) If 'pattern' matches the "->" string, everything on the left of | |
18 | * the arrow is treated as the name of an hash field, and the part on the | |
19 | * left as the key name containing an hash. The value of the specified | |
20 | * field is returned. | |
21 | * | |
22 | * 3) If 'pattern' equals "#", the function simply returns 'subst' itself so | |
23 | * that the SORT command can be used like: SORT key GET # to retrieve | |
24 | * the Set/List elements directly. | |
25 | * | |
26 | * The returned object will always have its refcount increased by 1 | |
27 | * when it is non-NULL. */ | |
28 | robj *lookupKeyByPattern(redisDb *db, robj *pattern, robj *subst) { | |
29 | char *p, *f, *k; | |
30 | sds spat, ssub; | |
31 | robj *keyobj, *fieldobj = NULL, *o; | |
32 | int prefixlen, sublen, postfixlen, fieldlen; | |
33 | ||
34 | /* If the pattern is "#" return the substitution object itself in order | |
35 | * to implement the "SORT ... GET #" feature. */ | |
36 | spat = pattern->ptr; | |
37 | if (spat[0] == '#' && spat[1] == '\0') { | |
38 | incrRefCount(subst); | |
39 | return subst; | |
40 | } | |
41 | ||
42 | /* The substitution object may be specially encoded. If so we create | |
43 | * a decoded object on the fly. Otherwise getDecodedObject will just | |
44 | * increment the ref count, that we'll decrement later. */ | |
45 | subst = getDecodedObject(subst); | |
46 | ssub = subst->ptr; | |
47 | ||
48 | /* If we can't find '*' in the pattern we return NULL as to GET a | |
49 | * fixed key does not make sense. */ | |
50 | p = strchr(spat,'*'); | |
51 | if (!p) { | |
52 | decrRefCount(subst); | |
53 | return NULL; | |
54 | } | |
55 | ||
56 | /* Find out if we're dealing with a hash dereference. */ | |
57 | if ((f = strstr(p+1, "->")) != NULL && *(f+2) != '\0') { | |
58 | fieldlen = sdslen(spat)-(f-spat)-2; | |
59 | fieldobj = createStringObject(f+2,fieldlen); | |
60 | } else { | |
61 | fieldlen = 0; | |
62 | } | |
63 | ||
64 | /* Perform the '*' substitution. */ | |
65 | prefixlen = p-spat; | |
66 | sublen = sdslen(ssub); | |
67 | postfixlen = sdslen(spat)-(prefixlen+1)-(fieldlen ? fieldlen+2 : 0); | |
68 | keyobj = createStringObject(NULL,prefixlen+sublen+postfixlen); | |
69 | k = keyobj->ptr; | |
70 | memcpy(k,spat,prefixlen); | |
71 | memcpy(k+prefixlen,ssub,sublen); | |
72 | memcpy(k+prefixlen+sublen,p+1,postfixlen); | |
73 | decrRefCount(subst); /* Incremented by decodeObject() */ | |
74 | ||
75 | /* Lookup substituted key */ | |
76 | o = lookupKeyRead(db,keyobj); | |
77 | if (o == NULL) goto noobj; | |
78 | ||
79 | if (fieldobj) { | |
80 | if (o->type != REDIS_HASH) goto noobj; | |
81 | ||
82 | /* Retrieve value from hash by the field name. This operation | |
83 | * already increases the refcount of the returned object. */ | |
84 | o = hashTypeGetObject(o, fieldobj); | |
85 | } else { | |
86 | if (o->type != REDIS_STRING) goto noobj; | |
87 | ||
88 | /* Every object that this function returns needs to have its refcount | |
89 | * increased. sortCommand decreases it again. */ | |
90 | incrRefCount(o); | |
91 | } | |
92 | decrRefCount(keyobj); | |
93 | if (fieldobj) decrRefCount(fieldobj); | |
94 | return o; | |
95 | ||
96 | noobj: | |
97 | decrRefCount(keyobj); | |
98 | if (fieldlen) decrRefCount(fieldobj); | |
99 | return NULL; | |
100 | } | |
101 | ||
102 | /* sortCompare() is used by qsort in sortCommand(). Given that qsort_r with | |
103 | * the additional parameter is not standard but a BSD-specific we have to | |
104 | * pass sorting parameters via the global 'server' structure */ | |
105 | int sortCompare(const void *s1, const void *s2) { | |
106 | const redisSortObject *so1 = s1, *so2 = s2; | |
107 | int cmp; | |
108 | ||
109 | if (!server.sort_alpha) { | |
110 | /* Numeric sorting. Here it's trivial as we precomputed scores */ | |
111 | if (so1->u.score > so2->u.score) { | |
112 | cmp = 1; | |
113 | } else if (so1->u.score < so2->u.score) { | |
114 | cmp = -1; | |
115 | } else { | |
116 | /* Objects have the same score, but we don't want the comparison | |
117 | * to be undefined, so we compare objects lexicographycally. | |
118 | * This way the result of SORT is deterministic. */ | |
119 | cmp = compareStringObjects(so1->obj,so2->obj); | |
120 | } | |
121 | } else { | |
122 | /* Alphanumeric sorting */ | |
123 | if (server.sort_bypattern) { | |
124 | if (!so1->u.cmpobj || !so2->u.cmpobj) { | |
125 | /* At least one compare object is NULL */ | |
126 | if (so1->u.cmpobj == so2->u.cmpobj) | |
127 | cmp = 0; | |
128 | else if (so1->u.cmpobj == NULL) | |
129 | cmp = -1; | |
130 | else | |
131 | cmp = 1; | |
132 | } else { | |
133 | /* We have both the objects, use strcoll */ | |
134 | cmp = strcoll(so1->u.cmpobj->ptr,so2->u.cmpobj->ptr); | |
135 | } | |
136 | } else { | |
137 | /* Compare elements directly. */ | |
138 | cmp = compareStringObjects(so1->obj,so2->obj); | |
139 | } | |
140 | } | |
141 | return server.sort_desc ? -cmp : cmp; | |
142 | } | |
143 | ||
144 | /* The SORT command is the most complex command in Redis. Warning: this code | |
145 | * is optimized for speed and a bit less for readability */ | |
146 | void sortCommand(redisClient *c) { | |
147 | list *operations; | |
148 | unsigned int outputlen = 0; | |
149 | int desc = 0, alpha = 0; | |
150 | long limit_start = 0, limit_count = -1, start, end; | |
151 | int j, dontsort = 0, vectorlen; | |
152 | int getop = 0; /* GET operation counter */ | |
153 | int int_convertion_error = 0; | |
154 | robj *sortval, *sortby = NULL, *storekey = NULL; | |
155 | redisSortObject *vector; /* Resulting vector to sort */ | |
156 | ||
157 | /* Lookup the key to sort. It must be of the right types */ | |
158 | sortval = lookupKeyRead(c->db,c->argv[1]); | |
159 | if (sortval && sortval->type != REDIS_SET && sortval->type != REDIS_LIST && | |
160 | sortval->type != REDIS_ZSET) | |
161 | { | |
162 | addReply(c,shared.wrongtypeerr); | |
163 | return; | |
164 | } | |
165 | ||
166 | /* Create a list of operations to perform for every sorted element. | |
167 | * Operations can be GET/DEL/INCR/DECR */ | |
168 | operations = listCreate(); | |
169 | listSetFreeMethod(operations,zfree); | |
170 | j = 2; | |
171 | ||
172 | /* Now we need to protect sortval incrementing its count, in the future | |
173 | * SORT may have options able to overwrite/delete keys during the sorting | |
174 | * and the sorted key itself may get destroied */ | |
175 | if (sortval) | |
176 | incrRefCount(sortval); | |
177 | else | |
178 | sortval = createListObject(); | |
179 | ||
180 | /* The SORT command has an SQL-alike syntax, parse it */ | |
181 | while(j < c->argc) { | |
182 | int leftargs = c->argc-j-1; | |
183 | if (!strcasecmp(c->argv[j]->ptr,"asc")) { | |
184 | desc = 0; | |
185 | } else if (!strcasecmp(c->argv[j]->ptr,"desc")) { | |
186 | desc = 1; | |
187 | } else if (!strcasecmp(c->argv[j]->ptr,"alpha")) { | |
188 | alpha = 1; | |
189 | } else if (!strcasecmp(c->argv[j]->ptr,"limit") && leftargs >= 2) { | |
190 | if ((getLongFromObjectOrReply(c, c->argv[j+1], &limit_start, NULL) != REDIS_OK) || | |
191 | (getLongFromObjectOrReply(c, c->argv[j+2], &limit_count, NULL) != REDIS_OK)) return; | |
192 | j+=2; | |
193 | } else if (!strcasecmp(c->argv[j]->ptr,"store") && leftargs >= 1) { | |
194 | storekey = c->argv[j+1]; | |
195 | j++; | |
196 | } else if (!strcasecmp(c->argv[j]->ptr,"by") && leftargs >= 1) { | |
197 | sortby = c->argv[j+1]; | |
198 | /* If the BY pattern does not contain '*', i.e. it is constant, | |
199 | * we don't need to sort nor to lookup the weight keys. */ | |
200 | if (strchr(c->argv[j+1]->ptr,'*') == NULL) dontsort = 1; | |
201 | j++; | |
202 | } else if (!strcasecmp(c->argv[j]->ptr,"get") && leftargs >= 1) { | |
203 | listAddNodeTail(operations,createSortOperation( | |
204 | REDIS_SORT_GET,c->argv[j+1])); | |
205 | getop++; | |
206 | j++; | |
207 | } else { | |
208 | decrRefCount(sortval); | |
209 | listRelease(operations); | |
210 | addReply(c,shared.syntaxerr); | |
211 | return; | |
212 | } | |
213 | j++; | |
214 | } | |
215 | ||
216 | /* If we have STORE we need to force sorting for deterministic output | |
217 | * and replication. We use alpha sorting since this is guaranteed to | |
218 | * work with any input. | |
219 | * | |
220 | * We also want determinism when SORT is called from Lua scripts, so | |
221 | * in this case we also force alpha sorting. */ | |
222 | if ((storekey || c->flags & REDIS_LUA_CLIENT) && dontsort) { | |
223 | dontsort = 0; | |
224 | alpha = 1; | |
225 | sortby = NULL; | |
226 | } | |
227 | ||
228 | /* Destructively convert encoded sorted sets for SORT. */ | |
229 | if (sortval->type == REDIS_ZSET) | |
230 | zsetConvert(sortval, REDIS_ENCODING_SKIPLIST); | |
231 | ||
232 | /* Load the sorting vector with all the objects to sort */ | |
233 | switch(sortval->type) { | |
234 | case REDIS_LIST: vectorlen = listTypeLength(sortval); break; | |
235 | case REDIS_SET: vectorlen = setTypeSize(sortval); break; | |
236 | case REDIS_ZSET: vectorlen = dictSize(((zset*)sortval->ptr)->dict); break; | |
237 | default: vectorlen = 0; redisPanic("Bad SORT type"); /* Avoid GCC warning */ | |
238 | } | |
239 | vector = zmalloc(sizeof(redisSortObject)*vectorlen); | |
240 | j = 0; | |
241 | ||
242 | if (sortval->type == REDIS_LIST) { | |
243 | listTypeIterator *li = listTypeInitIterator(sortval,0,REDIS_TAIL); | |
244 | listTypeEntry entry; | |
245 | while(listTypeNext(li,&entry)) { | |
246 | vector[j].obj = listTypeGet(&entry); | |
247 | vector[j].u.score = 0; | |
248 | vector[j].u.cmpobj = NULL; | |
249 | j++; | |
250 | } | |
251 | listTypeReleaseIterator(li); | |
252 | } else if (sortval->type == REDIS_SET) { | |
253 | setTypeIterator *si = setTypeInitIterator(sortval); | |
254 | robj *ele; | |
255 | while((ele = setTypeNextObject(si)) != NULL) { | |
256 | vector[j].obj = ele; | |
257 | vector[j].u.score = 0; | |
258 | vector[j].u.cmpobj = NULL; | |
259 | j++; | |
260 | } | |
261 | setTypeReleaseIterator(si); | |
262 | } else if (sortval->type == REDIS_ZSET) { | |
263 | dict *set = ((zset*)sortval->ptr)->dict; | |
264 | dictIterator *di; | |
265 | dictEntry *setele; | |
266 | di = dictGetIterator(set); | |
267 | while((setele = dictNext(di)) != NULL) { | |
268 | vector[j].obj = dictGetKey(setele); | |
269 | vector[j].u.score = 0; | |
270 | vector[j].u.cmpobj = NULL; | |
271 | j++; | |
272 | } | |
273 | dictReleaseIterator(di); | |
274 | } else { | |
275 | redisPanic("Unknown type"); | |
276 | } | |
277 | redisAssertWithInfo(c,sortval,j == vectorlen); | |
278 | ||
279 | /* Now it's time to load the right scores in the sorting vector */ | |
280 | if (dontsort == 0) { | |
281 | for (j = 0; j < vectorlen; j++) { | |
282 | robj *byval; | |
283 | if (sortby) { | |
284 | /* lookup value to sort by */ | |
285 | byval = lookupKeyByPattern(c->db,sortby,vector[j].obj); | |
286 | if (!byval) continue; | |
287 | } else { | |
288 | /* use object itself to sort by */ | |
289 | byval = vector[j].obj; | |
290 | } | |
291 | ||
292 | if (alpha) { | |
293 | if (sortby) vector[j].u.cmpobj = getDecodedObject(byval); | |
294 | } else { | |
295 | if (byval->encoding == REDIS_ENCODING_RAW) { | |
296 | char *eptr; | |
297 | ||
298 | vector[j].u.score = strtod(byval->ptr,&eptr); | |
299 | if (eptr[0] != '\0' || errno == ERANGE || | |
300 | isnan(vector[j].u.score)) | |
301 | { | |
302 | int_convertion_error = 1; | |
303 | } | |
304 | } else if (byval->encoding == REDIS_ENCODING_INT) { | |
305 | /* Don't need to decode the object if it's | |
306 | * integer-encoded (the only encoding supported) so | |
307 | * far. We can just cast it */ | |
308 | vector[j].u.score = (long)byval->ptr; | |
309 | } else { | |
310 | redisAssertWithInfo(c,sortval,1 != 1); | |
311 | } | |
312 | } | |
313 | ||
314 | /* when the object was retrieved using lookupKeyByPattern, | |
315 | * its refcount needs to be decreased. */ | |
316 | if (sortby) { | |
317 | decrRefCount(byval); | |
318 | } | |
319 | } | |
320 | } | |
321 | ||
322 | /* We are ready to sort the vector... perform a bit of sanity check | |
323 | * on the LIMIT option too. We'll use a partial version of quicksort. */ | |
324 | start = (limit_start < 0) ? 0 : limit_start; | |
325 | end = (limit_count < 0) ? vectorlen-1 : start+limit_count-1; | |
326 | if (start >= vectorlen) { | |
327 | start = vectorlen-1; | |
328 | end = vectorlen-2; | |
329 | } | |
330 | if (end >= vectorlen) end = vectorlen-1; | |
331 | ||
332 | if (dontsort == 0) { | |
333 | server.sort_desc = desc; | |
334 | server.sort_alpha = alpha; | |
335 | server.sort_bypattern = sortby ? 1 : 0; | |
336 | if (sortby && (start != 0 || end != vectorlen-1)) | |
337 | pqsort(vector,vectorlen,sizeof(redisSortObject),sortCompare, start,end); | |
338 | else | |
339 | qsort(vector,vectorlen,sizeof(redisSortObject),sortCompare); | |
340 | } | |
341 | ||
342 | /* Send command output to the output buffer, performing the specified | |
343 | * GET/DEL/INCR/DECR operations if any. */ | |
344 | outputlen = getop ? getop*(end-start+1) : end-start+1; | |
345 | if (int_convertion_error) { | |
346 | addReplyError(c,"One or more scores can't be converted into double"); | |
347 | } else if (storekey == NULL) { | |
348 | /* STORE option not specified, sent the sorting result to client */ | |
349 | addReplyMultiBulkLen(c,outputlen); | |
350 | for (j = start; j <= end; j++) { | |
351 | listNode *ln; | |
352 | listIter li; | |
353 | ||
354 | if (!getop) addReplyBulk(c,vector[j].obj); | |
355 | listRewind(operations,&li); | |
356 | while((ln = listNext(&li))) { | |
357 | redisSortOperation *sop = ln->value; | |
358 | robj *val = lookupKeyByPattern(c->db,sop->pattern, | |
359 | vector[j].obj); | |
360 | ||
361 | if (sop->type == REDIS_SORT_GET) { | |
362 | if (!val) { | |
363 | addReply(c,shared.nullbulk); | |
364 | } else { | |
365 | addReplyBulk(c,val); | |
366 | decrRefCount(val); | |
367 | } | |
368 | } else { | |
369 | /* Always fails */ | |
370 | redisAssertWithInfo(c,sortval,sop->type == REDIS_SORT_GET); | |
371 | } | |
372 | } | |
373 | } | |
374 | } else { | |
375 | robj *sobj = createZiplistObject(); | |
376 | ||
377 | /* STORE option specified, set the sorting result as a List object */ | |
378 | for (j = start; j <= end; j++) { | |
379 | listNode *ln; | |
380 | listIter li; | |
381 | ||
382 | if (!getop) { | |
383 | listTypePush(sobj,vector[j].obj,REDIS_TAIL); | |
384 | } else { | |
385 | listRewind(operations,&li); | |
386 | while((ln = listNext(&li))) { | |
387 | redisSortOperation *sop = ln->value; | |
388 | robj *val = lookupKeyByPattern(c->db,sop->pattern, | |
389 | vector[j].obj); | |
390 | ||
391 | if (sop->type == REDIS_SORT_GET) { | |
392 | if (!val) val = createStringObject("",0); | |
393 | ||
394 | /* listTypePush does an incrRefCount, so we should take care | |
395 | * care of the incremented refcount caused by either | |
396 | * lookupKeyByPattern or createStringObject("",0) */ | |
397 | listTypePush(sobj,val,REDIS_TAIL); | |
398 | decrRefCount(val); | |
399 | } else { | |
400 | /* Always fails */ | |
401 | redisAssertWithInfo(c,sortval,sop->type == REDIS_SORT_GET); | |
402 | } | |
403 | } | |
404 | } | |
405 | } | |
406 | if (outputlen) { | |
407 | setKey(c->db,storekey,sobj); | |
408 | server.dirty += outputlen; | |
409 | } else if (dbDelete(c->db,storekey)) { | |
410 | signalModifiedKey(c->db,storekey); | |
411 | server.dirty++; | |
412 | } | |
413 | decrRefCount(sobj); | |
414 | addReplyLongLong(c,outputlen); | |
415 | } | |
416 | ||
417 | /* Cleanup */ | |
418 | if (sortval->type == REDIS_LIST || sortval->type == REDIS_SET) | |
419 | for (j = 0; j < vectorlen; j++) | |
420 | decrRefCount(vector[j].obj); | |
421 | decrRefCount(sortval); | |
422 | listRelease(operations); | |
423 | for (j = 0; j < vectorlen; j++) { | |
424 | if (alpha && vector[j].u.cmpobj) | |
425 | decrRefCount(vector[j].u.cmpobj); | |
426 | } | |
427 | zfree(vector); | |
428 | } | |
429 | ||
430 |