Data in S3, avoiding write latency reduction by retaining raft, but what about reads? Is there still S3 latency?

Translated
1

Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.

Original topic: 数据在 S3 ,通过保留 raft 避免降低写延迟，那么读呢？是不是还是有 s3 的延迟？

| username: TiDBer_jYQINSnf

If the data is in S3, retaining raft avoids reducing write latency, but what about reads? Is there still S3 latency?

2
| username: sunxiaoguang | Original post link

There is a cache on the local disk for reading.

3
| username: TiDBer_jYQINSnf | Original post link

It’s equivalent to using the local disk as a level-1 cache. If there’s a cache miss, there will still be the latency of accessing S3, right?

4
| username: TiDBer_jYQINSnf | Original post link

I have another question: multiple regions do not share the LSM tree. Has this changed RocksDB or just the way RocksDB is used? Are there still 4 column families? Is it multiple LSM trees under one column family or directly using RocksDB’s column families, creating region count * 3 column families directly?

5
| username: sunxiaoguang | Original post link

Yes, but if you are concerned about performance, you can choose a 100% caching strategy.

6
| username: TiDBer_jYQINSnf | Original post link

Great! If the cache percentage can be configured, the flexibility is excellent! The experts at PingCAP truly have all sorts of ingenious ideas!

7
| username: sunxiaoguang | Original post link

Serverless does not use RocksDB; it is an independent storage engine designed for the cloud. Additionally, the main branch version of TiKV has also implemented the ability to have multiple regions without sharing the LSM Tree on RocksDB. Each region has its own independent instance of RocksDB’s LSM Tree.

8
| username: system | Original post link

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.