some more comment about bio.c design

[redis.git] / src / rdb.c

diff --git a/src/rdb.c b/src/rdb.c
index 83fe81e503146c231083aa3a388b62c807a9a6aa..cfbec3e8672350b0d106d4839587a80b2b332c01 100644 (file)
--- a/src/rdb.c
+++ b/src/rdb.c
@@ -139,7 +139,7 @@ writeerr:
 }
 
 /* Save a string objet as [len][data] on disk. If the object is a string
 }
 
 /* Save a string objet as [len][data] on disk. If the object is a string

- * representation of an integer value we try to safe it in a special form */
+ * representation of an integer value we try to save it in a special form */

 int rdbSaveRawString(FILE *fp, unsigned char *s, size_t len) {
     int enclen;
     int n, nwritten = 0;
 int rdbSaveRawString(FILE *fp, unsigned char *s, size_t len) {
     int enclen;
     int n, nwritten = 0;


@@ -245,7 +245,7 @@ int rdbSaveDoubleValue(FILE *fp, double val) {
     return rdbWriteRaw(fp,buf,len);
 }
 
     return rdbWriteRaw(fp,buf,len);
 }
 

-/* Save a Redis object. */
+/* Save a Redis object. Returns -1 on error, 0 on success. */

 int rdbSaveObject(FILE *fp, robj *o) {
     int n, nwritten = 0;
 
 int rdbSaveObject(FILE *fp, robj *o) {
     int n, nwritten = 0;
 


@@ -256,27 +256,10 @@ int rdbSaveObject(FILE *fp, robj *o) {
     } else if (o->type == REDIS_LIST) {
         /* Save a list value */
         if (o->encoding == REDIS_ENCODING_ZIPLIST) {
     } else if (o->type == REDIS_LIST) {
         /* Save a list value */
         if (o->encoding == REDIS_ENCODING_ZIPLIST) {

-            unsigned char *p;
-            unsigned char *vstr;
-            unsigned int vlen;
-            long long vlong;
+            size_t l = ziplistBlobLen((unsigned char*)o->ptr);

 
 

-            if ((n = rdbSaveLen(fp,ziplistLen(o->ptr))) == -1) return -1;
+            if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;

             nwritten += n;
             nwritten += n;

-
-            p = ziplistIndex(o->ptr,0);
-            while(ziplistGet(p,&vstr,&vlen,&vlong)) {
-                if (vstr) {
-                    if ((n = rdbSaveRawString(fp,vstr,vlen)) == -1)
-                        return -1;
-                    nwritten += n;
-                } else {
-                    if ((n = rdbSaveLongLongAsStringObject(fp,vlong)) == -1)
-                        return -1;
-                    nwritten += n;
-                }
-                p = ziplistNext(o->ptr,p);
-            }

         } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
             list *list = o->ptr;
             listIter li;
         } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
             list *list = o->ptr;
             listIter li;


@@ -311,56 +294,48 @@ int rdbSaveObject(FILE *fp, robj *o) {
             }
             dictReleaseIterator(di);
         } else if (o->encoding == REDIS_ENCODING_INTSET) {
             }
             dictReleaseIterator(di);
         } else if (o->encoding == REDIS_ENCODING_INTSET) {

-            intset *is = o->ptr;
-            int64_t llval;
-            int i = 0;
+            size_t l = intsetBlobLen((intset*)o->ptr);

 
 

-            if ((n = rdbSaveLen(fp,intsetLen(is))) == -1) return -1;
+            if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;

             nwritten += n;
             nwritten += n;

-
-            while(intsetGet(is,i++,&llval)) {
-                if ((n = rdbSaveLongLongAsStringObject(fp,llval)) == -1) return -1;
-                nwritten += n;
-            }

         } else {
             redisPanic("Unknown set encoding");
         }
     } else if (o->type == REDIS_ZSET) {
         } else {
             redisPanic("Unknown set encoding");
         }
     } else if (o->type == REDIS_ZSET) {

-        /* Save a set value */
-        zset *zs = o->ptr;
-        dictIterator *di = dictGetIterator(zs->dict);
-        dictEntry *de;
-
-        if ((n = rdbSaveLen(fp,dictSize(zs->dict))) == -1) return -1;
-        nwritten += n;
-
-        while((de = dictNext(di)) != NULL) {
-            robj *eleobj = dictGetEntryKey(de);
-            double *score = dictGetEntryVal(de);
+        /* Save a sorted set value */
+        if (o->encoding == REDIS_ENCODING_ZIPLIST) {
+            size_t l = ziplistBlobLen((unsigned char*)o->ptr);

 
 

-            if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
+            if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;

             nwritten += n;
             nwritten += n;

-            if ((n = rdbSaveDoubleValue(fp,*score)) == -1) return -1;
+        } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
+            zset *zs = o->ptr;
+            dictIterator *di = dictGetIterator(zs->dict);
+            dictEntry *de;
+
+            if ((n = rdbSaveLen(fp,dictSize(zs->dict))) == -1) return -1;

             nwritten += n;
             nwritten += n;

+
+            while((de = dictNext(di)) != NULL) {
+                robj *eleobj = dictGetEntryKey(de);
+                double *score = dictGetEntryVal(de);
+
+                if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
+                nwritten += n;
+                if ((n = rdbSaveDoubleValue(fp,*score)) == -1) return -1;
+                nwritten += n;
+            }
+            dictReleaseIterator(di);
+        } else {
+            redisPanic("Unknown sorted set encoding");

         }
         }

-        dictReleaseIterator(di);

     } else if (o->type == REDIS_HASH) {
         /* Save a hash value */
         if (o->encoding == REDIS_ENCODING_ZIPMAP) {
     } else if (o->type == REDIS_HASH) {
         /* Save a hash value */
         if (o->encoding == REDIS_ENCODING_ZIPMAP) {

-            unsigned char *p = zipmapRewind(o->ptr);
-            unsigned int count = zipmapLen(o->ptr);
-            unsigned char *key, *val;
-            unsigned int klen, vlen;
+            size_t l = zipmapBlobLen((unsigned char*)o->ptr);

 
 

-            if ((n = rdbSaveLen(fp,count)) == -1) return -1;
+            if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;

             nwritten += n;
             nwritten += n;

-
-            while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) {
-                if ((n = rdbSaveRawString(fp,key,klen)) == -1) return -1;
-                nwritten += n;
-                if ((n = rdbSaveRawString(fp,val,vlen)) == -1) return -1;
-                nwritten += n;
-            }

         } else {
             dictIterator *di = dictGetIterator(o->ptr);
             dictEntry *de;
         } else {
             dictIterator *di = dictGetIterator(o->ptr);
             dictEntry *de;


@@ -399,12 +374,10 @@ off_t rdbSavedObjectLen(robj *o) {
  * On error -1 is returned.
  * On success if the key was actaully saved 1 is returned, otherwise 0
  * is returned (the key was already expired). */
  * On error -1 is returned.
  * On success if the key was actaully saved 1 is returned, otherwise 0
  * is returned (the key was already expired). */

-int rdbSaveKeyValuePair(FILE *fp, redisDb *db, robj *key, robj *val,
-                        time_t now)
+int rdbSaveKeyValuePair(FILE *fp, robj *key, robj *val,
+                        time_t expiretime, time_t now)

 {
 {

-    time_t expiretime;
-    
-    expiretime = getExpire(db,key);
+    int vtype;

 
     /* Save the expire time */
     if (expiretime != -1) {
 
     /* Save the expire time */
     if (expiretime != -1) {


@@ -413,8 +386,19 @@ int rdbSaveKeyValuePair(FILE *fp, redisDb *db, robj *key, robj *val,
         if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) return -1;
         if (rdbSaveTime(fp,expiretime) == -1) return -1;
     }
         if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) return -1;
         if (rdbSaveTime(fp,expiretime) == -1) return -1;
     }

+    /* Fix the object type if needed, to support saving zipmaps, ziplists,
+     * and intsets, directly as blobs of bytes: they are already serialized. */
+    vtype = val->type;
+    if (vtype == REDIS_HASH && val->encoding == REDIS_ENCODING_ZIPMAP)
+        vtype = REDIS_HASH_ZIPMAP;
+    else if (vtype == REDIS_LIST && val->encoding == REDIS_ENCODING_ZIPLIST)
+        vtype = REDIS_LIST_ZIPLIST;
+    else if (vtype == REDIS_SET && val->encoding == REDIS_ENCODING_INTSET)
+        vtype = REDIS_SET_INTSET;
+    else if (vtype == REDIS_ZSET && val->encoding == REDIS_ENCODING_ZIPLIST)
+        vtype = REDIS_ZSET_ZIPLIST;

     /* Save type, key, value */
     /* Save type, key, value */

-    if (rdbSaveType(fp,val->type) == -1) return -1;
+    if (rdbSaveType(fp,vtype) == -1) return -1;

     if (rdbSaveStringObject(fp,key) == -1) return -1;
     if (rdbSaveObject(fp,val) == -1) return -1;
     return 1;
     if (rdbSaveStringObject(fp,key) == -1) return -1;
     if (rdbSaveObject(fp,val) == -1) return -1;
     return 1;


@@ -429,11 +413,6 @@ int rdbSave(char *filename) {
     int j;
     time_t now = time(NULL);
 
     int j;
     time_t now = time(NULL);
 

-    if (server.ds_enabled) {
-        cacheForcePointInTime();
-        return dsRdbSave(filename);
-    }
-

     snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
     fp = fopen(tmpfile,"w");
     if (!fp) {
     snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
     fp = fopen(tmpfile,"w");
     if (!fp) {


@@ -441,12 +420,12 @@ int rdbSave(char *filename) {
             strerror(errno));
         return REDIS_ERR;
     }
             strerror(errno));
         return REDIS_ERR;
     }

-    if (fwrite("REDIS0001",9,1,fp) == 0) goto werr;
+    if (fwrite("REDIS0002",9,1,fp) == 0) goto werr;

     for (j = 0; j < server.dbnum; j++) {
         redisDb *db = server.db+j;
         dict *d = db->dict;
         if (dictSize(d) == 0) continue;
     for (j = 0; j < server.dbnum; j++) {
         redisDb *db = server.db+j;
         dict *d = db->dict;
         if (dictSize(d) == 0) continue;

-        di = dictGetIterator(d);
+        di = dictGetSafeIterator(d);

         if (!di) {
             fclose(fp);
             return REDIS_ERR;
         if (!di) {
             fclose(fp);
             return REDIS_ERR;


@@ -460,9 +439,11 @@ int rdbSave(char *filename) {
         while((de = dictNext(di)) != NULL) {
             sds keystr = dictGetEntryKey(de);
             robj key, *o = dictGetEntryVal(de);
         while((de = dictNext(di)) != NULL) {
             sds keystr = dictGetEntryKey(de);
             robj key, *o = dictGetEntryVal(de);

+            time_t expire;

             
             initStaticStringObject(key,keystr);
             
             initStaticStringObject(key,keystr);

-            if (rdbSaveKeyValuePair(fp,db,&key,o,now) == -1) goto werr;
+            expire = getExpire(db,&key);
+            if (rdbSaveKeyValuePair(fp,&key,o,expire,now) == -1) goto werr;

         }
         dictReleaseIterator(di);
     }
         }
         dictReleaseIterator(di);
     }


@@ -496,17 +477,13 @@ werr:
 
 int rdbSaveBackground(char *filename) {
     pid_t childpid;
 
 int rdbSaveBackground(char *filename) {
     pid_t childpid;

+    long long start;

 
 

-    if (server.bgsavechildpid != -1 ||
-        server.bgsavethread != (pthread_t) -1) return REDIS_ERR;
+    if (server.bgsavechildpid != -1) return REDIS_ERR;

 
     server.dirty_before_bgsave = server.dirty;
 
 
     server.dirty_before_bgsave = server.dirty;
 

-    if (server.ds_enabled) {
-        cacheForcePointInTime();
-        return dsRdbSaveBackground(filename);
-    }
-
+    start = ustime();

     if ((childpid = fork()) == 0) {
         int retval;
 
     if ((childpid = fork()) == 0) {
         int retval;
 


@@ -517,6 +494,7 @@ int rdbSaveBackground(char *filename) {
         _exit((retval == REDIS_OK) ? 0 : 1);
     } else {
         /* Parent */
         _exit((retval == REDIS_OK) ? 0 : 1);
     } else {
         /* Parent */

+        server.stat_fork_time = ustime()-start;

         if (childpid == -1) {
             redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
                 strerror(errno));
         if (childpid == -1) {
             redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
                 strerror(errno));


@@ -770,11 +748,13 @@ robj *rdbLoadObject(int type, FILE *fp) {
     } else if (type == REDIS_ZSET) {
         /* Read list/set value */
         size_t zsetlen;
     } else if (type == REDIS_ZSET) {
         /* Read list/set value */
         size_t zsetlen;

+        size_t maxelelen = 0;

         zset *zs;
 
         if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
         o = createZsetObject();
         zs = o->ptr;
         zset *zs;
 
         if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
         o = createZsetObject();
         zs = o->ptr;

+

         /* Load every single element of the list/set */
         while(zsetlen--) {
             robj *ele;
         /* Load every single element of the list/set */
         while(zsetlen--) {
             robj *ele;


@@ -784,10 +764,21 @@ robj *rdbLoadObject(int type, FILE *fp) {
             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
             ele = tryObjectEncoding(ele);
             if (rdbLoadDoubleValue(fp,&score) == -1) return NULL;
             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
             ele = tryObjectEncoding(ele);
             if (rdbLoadDoubleValue(fp,&score) == -1) return NULL;

+
+            /* Don't care about integer-encoded strings. */
+            if (ele->encoding == REDIS_ENCODING_RAW &&
+                sdslen(ele->ptr) > maxelelen)
+                    maxelelen = sdslen(ele->ptr);
+

             znode = zslInsert(zs->zsl,score,ele);
             dictAdd(zs->dict,ele,&znode->score);
             incrRefCount(ele); /* added to skiplist */
         }
             znode = zslInsert(zs->zsl,score,ele);
             dictAdd(zs->dict,ele,&znode->score);
             incrRefCount(ele); /* added to skiplist */
         }

+
+        /* Convert *after* loading, since sorted sets are not stored ordered. */
+        if (zsetLength(o) <= server.zset_max_ziplist_entries &&
+            maxelelen <= server.zset_max_ziplist_value)
+                zsetConvert(o,REDIS_ENCODING_ZIPLIST);

     } else if (type == REDIS_HASH) {
         size_t hashlen;
 
     } else if (type == REDIS_HASH) {
         size_t hashlen;
 


@@ -834,6 +825,54 @@ robj *rdbLoadObject(int type, FILE *fp) {
                 dictAdd((dict*)o->ptr,key,val);
             }
         }
                 dictAdd((dict*)o->ptr,key,val);
             }
         }

+    } else if (type == REDIS_HASH_ZIPMAP ||
+               type == REDIS_LIST_ZIPLIST ||
+               type == REDIS_SET_INTSET ||
+               type == REDIS_ZSET_ZIPLIST)
+    {
+        robj *aux = rdbLoadStringObject(fp);
+
+        if (aux == NULL) return NULL;
+        o = createObject(REDIS_STRING,NULL); /* string is just placeholder */
+        o->ptr = zmalloc(sdslen(aux->ptr));
+        memcpy(o->ptr,aux->ptr,sdslen(aux->ptr));
+        decrRefCount(aux);
+
+        /* Fix the object encoding, and make sure to convert the encoded
+         * data type into the base type if accordingly to the current
+         * configuration there are too many elements in the encoded data
+         * type. Note that we only check the length and not max element
+         * size as this is an O(N) scan. Eventually everything will get
+         * converted. */
+        switch(type) {
+            case REDIS_HASH_ZIPMAP:
+                o->type = REDIS_HASH;
+                o->encoding = REDIS_ENCODING_ZIPMAP;
+                if (zipmapLen(o->ptr) > server.hash_max_zipmap_entries)
+                    convertToRealHash(o);
+                break;
+            case REDIS_LIST_ZIPLIST:
+                o->type = REDIS_LIST;
+                o->encoding = REDIS_ENCODING_ZIPLIST;
+                if (ziplistLen(o->ptr) > server.list_max_ziplist_entries)
+                    listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
+                break;
+            case REDIS_SET_INTSET:
+                o->type = REDIS_SET;
+                o->encoding = REDIS_ENCODING_INTSET;
+                if (intsetLen(o->ptr) > server.set_max_intset_entries)
+                    setTypeConvert(o,REDIS_ENCODING_HT);
+                break;
+            case REDIS_ZSET_ZIPLIST:
+                o->type = REDIS_ZSET;
+                o->encoding = REDIS_ENCODING_ZIPLIST;
+                if (zsetLength(o) > server.zset_max_ziplist_entries)
+                    zsetConvert(o,REDIS_ENCODING_SKIPLIST);
+                break;
+            default:
+                redisPanic("Unknown encoding");
+                break;
+        }

     } else {
         redisPanic("Unknown object type");
     }
     } else {
         redisPanic("Unknown object type");
     }


@@ -868,7 +907,7 @@ void stopLoading(void) {
 int rdbLoad(char *filename) {
     FILE *fp;
     uint32_t dbid;
 int rdbLoad(char *filename) {
     FILE *fp;
     uint32_t dbid;

-    int type, retval, rdbver;
+    int type, rdbver;

     redisDb *db = server.db+0;
     char buf[1024];
     time_t expiretime, now = time(NULL);
     redisDb *db = server.db+0;
     char buf[1024];
     time_t expiretime, now = time(NULL);


@@ -884,7 +923,7 @@ int rdbLoad(char *filename) {
         return REDIS_ERR;
     }
     rdbver = atoi(buf+5);
         return REDIS_ERR;
     }
     rdbver = atoi(buf+5);

-    if (rdbver != 1) {
+    if (rdbver < 1 || rdbver > 2) {

         fclose(fp);
         redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
         return REDIS_ERR;
         fclose(fp);
         redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
         return REDIS_ERR;


@@ -931,11 +970,8 @@ int rdbLoad(char *filename) {
             continue;
         }
         /* Add the new object in the hash table */
             continue;
         }
         /* Add the new object in the hash table */

-        retval = dbAdd(db,key,val);
-        if (retval == REDIS_ERR) {
-            redisLog(REDIS_WARNING,"Loading DB, duplicated key (%s) found! Unrecoverable error, exiting now.", key->ptr);
-            exit(1);
-        }
+        dbAdd(db,key,val);
+

         /* Set the expire time if needed */
         if (expiretime != -1) setExpire(db,key,expiretime);
 
         /* Set the expire time if needed */
         if (expiretime != -1) setExpire(db,key,expiretime);
 


@@ -966,15 +1002,13 @@ void backgroundSaveDoneHandler(int exitcode, int bysignal) {
         rdbRemoveTempFile(server.bgsavechildpid);
     }
     server.bgsavechildpid = -1;
         rdbRemoveTempFile(server.bgsavechildpid);
     }
     server.bgsavechildpid = -1;

-    server.bgsavethread = (pthread_t) -1;
-    server.bgsavethread_state = REDIS_BGSAVE_THREAD_UNACTIVE;

     /* Possibly there are slaves waiting for a BGSAVE in order to be served
      * (the first stage of SYNC is a bulk transfer of dump.rdb) */
     updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
 }
 
 void saveCommand(redisClient *c) {
     /* Possibly there are slaves waiting for a BGSAVE in order to be served
      * (the first stage of SYNC is a bulk transfer of dump.rdb) */
     updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
 }
 
 void saveCommand(redisClient *c) {

-    if (server.bgsavechildpid != -1 || server.bgsavethread != (pthread_t)-1) {
+    if (server.bgsavechildpid != -1) {

         addReplyError(c,"Background save already in progress");
         return;
     }
         addReplyError(c,"Background save already in progress");
         return;
     }


@@ -986,11 +1020,11 @@ void saveCommand(redisClient *c) {
 }
 
 void bgsaveCommand(redisClient *c) {
 }
 
 void bgsaveCommand(redisClient *c) {

-    if (server.bgsavechildpid != -1 || server.bgsavethread != (pthread_t)-1) {
+    if (server.bgsavechildpid != -1) {

         addReplyError(c,"Background save already in progress");
         addReplyError(c,"Background save already in progress");

-        return;
-    }
-    if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
+    } else if (server.bgrewritechildpid != -1) {
+        addReplyError(c,"Can't BGSAVE while AOF log rewriting is in progress");
+    } else if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {

         addReplyStatus(c,"Background saving started");
     } else {
         addReply(c,shared.err);
         addReplyStatus(c,"Background saving started");
     } else {
         addReply(c,shared.err);