AOF: sync data on disk every 32MB when rewriting.

This prevents the kernel from putting too much stuff in the output
buffers, doing too heavy I/O all at once. So the goal of this commit is
to split the disk pressure due to the AOF rewrite process into smaller
spikes.

Please see issue #1019 for more information.

src/aof.c

@@ -855,6 +855,7 @@ int rewriteAppendOnlyFile(char *filename) {
     }
 
     rioInitWithFile(&aof,fp);
+    rioSetAutoSync(&aof,REDIS_AOF_AUTOSYNC_BYTES);
     for (j = 0; j < server.dbnum; j++) {
         char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
         redisDb *db = server.db+j;

src/redis.h

@@ -106,6 +106,7 @@
 #define REDIS_INLINE_MAX_SIZE   (1024*64) /* Max size of inline reads */
 #define REDIS_MBULK_BIG_ARG     (1024*32)
 #define REDIS_LONGSTR_SIZE      21          /* Bytes needed for long -> str */
+#define REDIS_AOF_AUTOSYNC_BYTES (1024*1024*32) /* fdatasync every 32MB */
 
 /* Hash table parameters */
 #define REDIS_HT_MINFILL        10      /* Minimal hash table fill 10% */