redis
redis的压缩是针对key的压缩
只针对string和list的value
所有的压缩最终都会调用lzf_compress/lzf_decompress
需要配置文件配置rdb_compression rdb压缩才会生效
lzf压缩限制长度要大于20,即使是aaaaaaaaaaaaaaaaaaaa也压不了,大于20才能压。原因没有深究
rdb内部的压缩
- 如何确认这个record是被压缩/解压的?
rdb解析每条数据,都有标识字段,压缩的record自然是单独的类型
- ssize_t rdbSaveLzfStringObject(rio *rdb, unsigned char *s, size_t len) {
- ...
- comprlen = lzf_compress(s, len, out, outlen);
- if (comprlen == 0) {
- zfree(out);
- return 0;
- }
- ssize_t nwritten = rdbSaveLzfBlob(rdb, out, comprlen, len);
- ...
- }
-
- ssize_t rdbSaveLzfBlob(rio *rdb, void *data, size_t compress_len,
- size_t original_len) {
- ...
-
- byte = (RDB_ENCVAL<<6)|RDB_ENC_LZF;
- if ((n = rdbWriteRaw(rdb,&byte,1)) == -1) goto writeerr;
- nwritten += n;
- ...
- }
解压缩
- void *rdbGenericLoadStringObject(rio *rdb, int flags, size_t *lenptr) {
- ...
- if (isencoded) {
- switch(len) {
- case RDB_ENC_INT8:
- case RDB_ENC_INT16:
- case RDB_ENC_INT32:
- return rdbLoadIntegerObject(rdb,len,flags,lenptr);
- case RDB_ENC_LZF:
- return rdbLoadLzfStringObject(rdb,flags,lenptr);
- default:
- rdbReportCorruptRDB("Unknown RDB string encoding type %llu",len);
- return NULL;
- }
- }
- ...
-
- void *rdbLoadLzfStringObject(rio *rdb, int flags, size_t *lenptr) {
- ...
-
-
- if (rioRead(rdb,c,clen) == 0) goto err;
- if (lzf_decompress(c,clen,val,len) != len) {
- rdbReportCorruptRDB("Invalid LZF compressed string");
- ...
- }
接口简单容易定位
所有的类型string/hash具体到底层,都是string,就会走这个压缩的过程rdbSaveRawString,内部来调用rdbSaveLzfStringObject
- ssize_t rdbSaveObject(rio *rdb, robj *o, robj *key, int dbid) {
- ssize_t n = 0, nwritten = 0;
-
- if (o->type == OBJ_STRING) {
-
- if ((n = rdbSaveStringObject(rdb,o)) == -1) return -1;
- nwritten += n;
- } else if (o->type == OBJ_LIST) {
-
- if (quicklistNodeIsCompressed(node)) {
- void *data;
- size_t compress_len = quicklistGetLzf(node, &data);
- if ((n = rdbSaveLzfBlob(rdb,data,compress_len,node->sz)) == -1) return -1;
- nwritten += n;
- } else {
- if ((n = rdbSaveRawString(rdb,node->zl,node->sz)) == -1) return -1;
- nwritten += n;
- }
- node = node->next;
- }
- } else {
- serverPanic("Unknown list encoding");
- }
- 。。。
- }
quicklist的压缩
链表压缩可以选择深度,quicklist是redis list的底层数据结构
什么时候做压缩?
-
- REDIS_STATIC void __quicklistInsertNode(quicklist *quicklist,
- quicklistNode *old_node,
- quicklistNode *new_node, int after) {
- if (after) {
- new_node->prev = old_node;
- if (old_node) {
- new_node->next = old_node->next;
- if (old_node->next)
- old_node->next->prev = new_node;
- old_node->next = new_node;
- }
- if (quicklist->tail == old_node)
- quicklist->tail = new_node;
- } else {
- new_node->next = old_node;
- if (old_node) {
- new_node->prev = old_node->prev;
- if (old_node->prev)
- old_node->prev->next = new_node;
- old_node->prev = new_node;
- }
- if (quicklist->head == old_node)
- quicklist->head = new_node;
- }
-
- if (quicklist->len == 0) {
- quicklist->head = quicklist->tail = new_node;
- }
-
-
- quicklist->len++;
-
- if (old_node)
- quicklistCompress(quicklist, old_node);
- }
也就是说,头尾不会压缩,其他的节点会压缩,在修改的时候同事把旧的节点给压缩了
这里有个问题,这里的节点压缩了,rdb存储的时候还要特别处理一下,判定已经压缩过,走rdbSaveLzfBlob
需要有个record头来记录一个compression的标记
rocksdb
类似redis,还是很好找的,UncompressData/CompressData
针对sst的压缩
调用关系
UncompressBlockContentsForCompressionType -> UncompressData
WriteBlock/BGWorkCompression -> CompressAndVerifyBlock -> CompressBlock -> CompressData
block本身有信息标记是否是压缩
写入的时候才压缩
blobdb
CompressBlobIfNeeded -> CompressData
GetCompressedSlice -> CompressData
总结
- 需要文件本身知道自己是压缩的,有元信息记录
- 在内存中是否压缩要考虑业务场景,比如redis这个quicklist 压缩,因为list最近访问的就是头尾,其他不重要