8
2

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 3 years have passed since last update.

なぜ本番環境のRedisにKEYSを使ってはいけないのか検証する

Last updated at Posted at 2019-12-07

Redisでよく使うコマンドにKEYSがあると思う。公式ドキュメントには、以下のように書いてあるんだけど、具体的にどうダメなのか見ていきたい。

Warning: consider KEYS as a command that should only be used in production environments with extreme care. It may ruin performance when it is executed against large databases. This command is intended for debugging and special operations, such as changing your keyspace layout.

警告: KEYSは十分に注意した上で本番環境で使うべき。KEYSは大きなデータセットで実行された場合はパフォーマンスに大きな影響がある。このコマンドはデバッグやキーのレイアウトを返る等の特別な場合のみに使われるべきである。

実験

  • CentOS Linux release 7.4.1708 (Core)
  • Intel(R) Core(TM) i7-7700 CPU @ 3.60GHz
  • DDR4 16GB RAM
  • redis-server 5.0.7

の環境でそれぞれKEYSCANを実行しながらredis-benchmarkのRPS(Request per second)を計測してみる。

準備

redis-serverをrunする

docker run --rm -it --name redis -p6379:6379 redis:5.0.7

redis-benchmarkの起動方法

redis-benchmark -t get -c 1

用意したプログラム(Rust)

Cargo.toml
[package]
name = "dont-use-redis-keys"
version = "0.1.0"
authors = ["yukinarit"]
edition = "2018"

# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html

[dependencies]
redis = "*"
structopt = "*"
tokio = "0.1"
futures = "0.1"
main.rs
use std::time::Instant;

use futures::{
    future::{loop_fn, Loop},
    prelude::*,
    stream,
};
use redis::{Client as Redis, Commands};
use structopt::StructOpt;
use tokio::runtime::Builder;

#[derive(StructOpt)]
enum Opt {
    Setup {
        /// Setup `count` K size of keys/values.
        #[structopt(short = "c", long = "count", default_value = "1")]
        count: usize,
    },
    /// Just continuously run "KEYS".
    Keys,
    /// Just continuously run "SCAN".
    Scan,
    /// Benchmarking "KEYS".
    BenchKeys {
        #[structopt(flatten)]
        opt: BenchOpt,
    },
    /// Benchmarking "SCAN".
    BenchScan {
        #[structopt(flatten)]
        opt: BenchOpt,
    },
}

#[derive(StructOpt)]
struct BenchOpt {
    /// Total number of requests.
    #[structopt(short = "c", long = "count", default_value = "1")]
    count: usize,
    /// Number of parallel connections.
    #[structopt(short = "p", long = "pipeline", default_value = "1")]
    pipeline: usize,
}

fn main() {
    let opt = Opt::from_args();
    let mut redis = Redis::open("redis://localhost:6379").expect("Coundn't connect Redis");
    let mut rt = Builder::new().core_threads(1).build().unwrap();
    let start = Instant::now();

    match opt {
        Opt::Setup { count } => {
            let mut conn = redis.get_connection().unwrap();
            for n in 0..count {
                let data: Vec<_> = (0..1000)
                    .map(|v| {
                        let v = n * 1000 + v;
                        vec![format!("key{}", v), format!("val{}", v)]
                    })
                    .flatten()
                    .collect();
                let _: () = redis::cmd("MSET").arg(data).query(&mut conn).unwrap();
            }
            println!("Elapsed: {} msec", (Instant::now() - start).as_millis());
        }
        Opt::Keys => loop {
            let mut conn = redis.get_connection().unwrap();
            loop {
                let val: Vec<String> = redis::cmd("KEYS").arg("key0").query(&mut conn).unwrap();
                println!("Got {}", val[0]);
            }
        },
        Opt::Scan => {
            let mut conn = redis.get_connection().unwrap();

            loop {
                conn.scan_match("key0").unwrap().for_each(|s: String| {
                    if s == "key0" {
                        println!("Got {}", s);
                    }
                });
            }
        }
        Opt::BenchKeys { opt } => {
            let count = opt.count;
            let f = redis
                .clone()
                .get_shared_async_connection()
                .map_err(|e| eprintln!("{}", e))
                .and_then(move |conn| {
                    stream::iter_ok::<_, ()>(0..opt.count)
                        .map(move |n| {
                            redis::cmd("KEYS")
                                .arg(format!("key{}", n))
                                .query_async(conn.clone())
                                .map(|(_, val): (_, Vec<String>)| {
                                    if !val.is_empty() {
                                        println!("{}", val[0]);
                                    }
                                })
                                .map_err(|e| eprintln!("{}", e))
                        })
                        .buffered(opt.pipeline)
                        .for_each(|_| Ok(()))
                });

            rt.block_on(f).unwrap();
            println!(
                "RPS:  {}",
                count as f64 / (Instant::now() - start).as_secs_f64()
            );
        }
        Opt::BenchScan { opt } => {
            let count = opt.count;
            let f = redis
                .clone()
                .get_shared_async_connection()
                .map_err(|e| eprintln!("{}", e))
                .and_then(move |conn| {
                    stream::iter_ok::<_, ()>(0..opt.count)
                        .map(move |n| {
                            loop_fn((conn.clone(), 0), move |(conn, cursor)| {
                                redis::cmd("SCAN")
                                    .arg(cursor)
                                    .arg("MATCH")
                                    .arg(format!("key{}", n))
                                    .query_async(conn)
                                    .and_then(|(conn, (cursor, val)): (_, (u32, Vec<String>))| {
                                        if !val.is_empty() {
                                            println!("{}", val[0]);
                                        }
                                        if cursor == 0 {
                                            Ok(Loop::Break((conn, cursor)))
                                        } else {
                                            Ok(Loop::Continue((conn, cursor)))
                                        }
                                    })
                                    .map_err(|e| eprintln!("{}", e))
                            })
                        })
                        .buffered(opt.pipeline)
                        .for_each(|_| Ok(()))
                });
            rt.block_on(f).unwrap();
            println!(
                "RPS: {}",
                count as f64 / (Instant::now() - start).as_secs_f64()
            );
        }
    }
}
  • テストデータ作成
cargo run -- setup -c <NUM_KEYS>
  • KEYSを続けて実行し続ける
cargo run -- keys
  • SCANを続けて実行し続ける
cargo run -- scan

結果

redis-benchmark.png

KEYSの場合、Redisのキーを増やていくとredis-benchmarkのRPSが著しく遅くなる。SCANは何もしていない場合と比べて若干遅いが、キーの数が増えてもRPSは安定している。

計測結果の生データは以下。

なにも実行してない場合の結果
keyの数 redis-benchmarkのRPS
1,000 29638.41
10,000 29994.00
100,000 30165.91
500,000 29761.91
1,000,000 29931.16
5,000,000 29205.61
10,000,000 28719.13
`KEYS`を実行し続けた場合の結果
keyの数 redis-benchmarkのRPS
1,000 16353.23
10,000 3630.82
100,000 373.79
500,000 31.03
1,000,000 14.02
5,000,000 2.52
10,000,000 1.08
`SCAN`を実行し続けた場合の結果
keyの数 redis-benchmarkのRPS
1,000 26413.10
10,000 25940.34
100,000 25974.03
500,000 25806.45
1,000,000 25680.53
5,000,000 23900.57
10,000,000 24473.81

結論

  • KEYSはRedis全体のKeyの数に対して**O(N)**
  • Redisのメインの処理は**シングスレッド**で行われる
  • 本番環境ではKEYSではなくSCANを使うこと
  • Redisのスレッドを完全にblockしてもいい場合のみだけKEYSを使おう
  • これらのことは全てドキュメントに書いてある
8
2
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
8
2

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?