Add a slow flag to commands with heavy disk IO

[PragmaTwice]

Search before asking

• I had searched in the issues and found no similar issues.

Motivation

We can add a new command flag e.g. slow and apply it to all commands that can lead to heavy disk IO (rocksdb read/write).

Solution

No response

Are you willing to submit a PR?

• I'm willing to submit a PR!

[PokIsemaine]

Is there any mechanism that will accompany this flag after it is added? And what is the metric for heavy disk IO?

[mapleFU]

Is there any mechanism that will accompany this flag after it is added?

Generally there're easier way and heavier way. Tagging can allow better techniques for scheduling, like independent io-task-pool or throttle for these tasks ( These techniques doesn't be implemented now, lol). Generally for user task "heavy request" might have a lower priority and avoid affect the light operations

And what is the metric for heavy disk IO?

No, lol, I guess this should be guessed by user and a bit hard to analysis, especially on zset or other heavy structure

[PragmaTwice]

And what is the metric for heavy disk IO?

A simple method is to check that if a command only involves a constant number of rocksdb read/writes, e.g. LPUSH, SET, HSET.

Otherwise it can be a time consuming operation, e.g. HGETALL, LINSERT... (the number of rocksdb read/writes is not constant, e.g. O(n), where n is a runtime variable like size of the data structure.)

[PragmaTwice]

Is there any mechanism that will accompany this flag after it is added?

Like mentioned by @mapleFU , it has potential benefits for optimization and scheduling.

Also, we can generate some doc from it to let users know which commands are time-consuming.

[mapleFU]

Besides, the heavy disk "write" IO api might have higher risk on new Context api: #2332

[jonathanc-n]

Can I be assigned this issue?

[git-hulk]

@jonathanc-n Assigned, thank you!