How we scaled the GitHub API with a sharded, replicated rate limiter in Redis

In that case, since the second boundary happened between the call to TTL and Time.now, the resulting timestamp was one second bigger than the previous ones. We could have tried increasing the precision of this operation (eg, Redis PTTL), but there would still have been some wobble, even if it was greatly reduced. Another possibility was to calculate the time using only Redis, instead of mixing Rub

[efcl]

GitHub APIのrate-limitの実装。 RedisをバックエンドにしたRate limiterの実装。 アプリケーションとRedisで微妙なズレが発生するケースはRedis上のLuaで全部計算することで回避、レプリカでのfalse positiveについてなど

[bootJP]

“場合”

[miya-jan]

GitHub の rate limit を memcached から redis に移行した事例