add scan command #199
add scan command #199
Conversation
|
@HDT3213 貌似是raft这块的逻辑 |
| @@ -412,6 +412,91 @@ func execCopy(mdb *Server, conn redis.Connection, args [][]byte) redis.Reply { | |||
| return protocol.MakeIntReply(1) | |||
| } | |||
|
|
|||
| // execScan iteratively output all keys in the current db | |||
| func execScan(db *DB, args [][]byte) redis.Reply { | |||
There was a problem hiding this comment.
scan 的参数有 MATCH，COUNT, TYPE 3个，实际执行 scan 的函数签名应该是 scan0(cursor, pattern, count, typ)。
现在的解析命令行逻辑过于晦涩， 可以参考一下 execSet 的实现
datastruct/dict/concurrent.go
Outdated
| if count >= size { | ||
| return dict.Keys(), nextCursor | ||
| } | ||
| remainingKeys := dict.table[cursor:] |
There was a problem hiding this comment.
table 中取出来的是 shard, 为啥起名叫 remainingKeys?
datastruct/dict/concurrent.go
Outdated
| @@ -435,3 +436,56 @@ func (dict *ConcurrentDict) RWUnLocks(writeKeys []string, readKeys []string) { | |||
| } | |||
| } | |||
| } | |||
|
|
|||
| // ScanKeys iteratively output all keys in the current db | |||
| func (dict *ConcurrentDict) ScanKeys(cursor, count int, matchKey string) ([]string, int) { | |||
datastruct/dict/concurrent.go
Outdated
| for key, _ := range s.m { | ||
| if key != "" { | ||
| if matchKey != "*" { | ||
| pattern, err := wildcard.CompilePattern(matchKey) |
datastruct/dict/concurrent.go
Outdated
| nextCursor := 0 | ||
| size := dict.Len() | ||
| result := make([]string, count) | ||
| r := make(map[string]struct{}) |
datastruct/dict/concurrent.go
Outdated
| i++ | ||
| } | ||
| } | ||
| if count <= i { |
There was a problem hiding this comment.
scan 命令不保证返回的元素数一定等于 count 参数，但是保证从遍历开始直到完整遍历结束期间， 一直存在于数据集内的所有元素都会被返回。如果到了 count 就中止的话，当前shard 中可能有一些 key 在遍历期间一直存在但因为遍历中止未被返回。
所以应完整遍历shard, 当 len(result) >= count 后返回即可。
for _, shard := range shards {
for k, _ := range shard.m {
result.add(key)
}
if len(result) >= count {
break
}
}
|
两个大问题
所以应完整遍历shard, 当 len(result) >= count 后返回即可。 |
| return dict.Keys(), nextCursor | ||
| } | ||
| remainingKeys := dict.table[cursor:] | ||
| i := 0 |
|
按照您的建议做了一些修改，貌似test failed。略有些不解🤔。 |
新加的 TestScan 挂了 |
|
我本地跑是正常的
|
|
知道了， 每次scan迭代返回的key是无序的，所以在test比对的时候会偶发不一致。 |
database/keys.go
Outdated
| return protocol.MakeNullBulkReply() | ||
| } | ||
|
|
||
| func scanWithArg(db *DB, args [][]byte, match bool) redis.Reply { |
There was a problem hiding this comment.
函数签名改成： execScan0(cursor, pattern, count, typ)
| } | ||
|
|
||
| // execScanWithArg execute scan command based on cli args | ||
| func execScanWithArg(db *DB, args [][]byte, argType int) redis.Reply { |
There was a problem hiding this comment.
函数签名改成 execScan0(cursor string, pattern string, count int)
type 参数可以先不加
There was a problem hiding this comment.
上次说的 type 不是参数类型， scan 命令可以根据 string/list/hash/set/sortedset 类型进行过滤.
|
|
||
| // execScan iteratively output all keys in the current db | ||
| func execScan(db *DB, args [][]byte) redis.Reply { | ||
| const ( |
There was a problem hiding this comment.
var count int
var pattern string
for i := 1; i < argsNum; i++ {
arg := strings.ToLower(string(args[i]))
if arg == "count" {
count = itoa(args[i+1])
i++
}
....
}
execScan0(cursor, pattern, count)
这样不就完事了，为啥写这么麻烦
| i++ | ||
| } | ||
| } | ||
| if len(m) >= count { |
There was a problem hiding this comment.
这里有个问题没想明白，如果每次都需要把所有的shard遍历完再返回的话
1、性能上是不是存在而外的开销？
2、每次遍历以count参数的个数作为一个取值范围，直至遍历完成，不也可以保证保证数据的全量被遍历到么？（不保证已经被遍历过的shard发生数据变化）
add scan command