Add multi-lock support for atomic multi-lock.

[bhallionOhbibi]

Main changes

The purpose of thoses changes is to be able to obtain multiple lock in a single atomic operation. This is useful in many situation were you need to lock resources access by multiple process.

For exemple, if we have:

• Process A accessing resource 1
• Process B accessing resources 1 & 2

We want Process A to be able to lock resource 1 and therefore preventing Process B to access it until Process A is done.

I'v been siting on this update since August 2022 and have been using it on production without issues.

Other changes

• Update test to use unique lockKeys each run.

[bhallionOhbibi]

Is there any chance someone takes a look a this ? :smile:

[dim]

@bhallionOhbibi apologies for the delay, but we have been exceptionally busy lately. We will take a look in the next few days

[mxmCherry]

Note: I'm all in for the Lua scripts instead of using Redis Pipelines here. We don't need any retvals from loop-ed batch operations (done within loops here as there's no / I have found no multi-key-EXPIRE command etc), so let's not waste "traffic" if we can do all the checks we want right in Lua.

[bhallionOhbibi]

Thanks for the review and for considering merging this. I suppose no changes are needed on my side since you plan on polish this to your taste (which I agree with btw). I wont touch the code anymore, unless you need me to.

[mxmCherry]

This one is now up to @dim (ping-ping). I can take over then, unless dim pushes for Pipelines (IMO, I personally prefer PR-ed Lua approach).

[bhallionOhbibi]

Ping

[tomaswarynyca]

@bhallionOhbibi do you have enough time to fix the problems?

[bhallionOhbibi]

I'll try to adresses thoses. I may have some questions

[tomaswarynyca]

@bhallionOhbibi I am very pleased to hear that you will be able to continue with the work! :+1: @dim @mxmCherry I'm sure they will help you with the questions you have!

[bhallionOhbibi]

I made a major update to this whole PR. I think there is no breaking changes and the changes are up to date with the latest version of main.

This is my second dive into lua scripting, bear with me if this is not very idiomatic. Feel free to comment on anything of course. I hope this only needs few changes to be of your liking before merging.

[dim]

Alternatively, I would not be opposed to having an ObtainMany which (atomically) obtains a slice of []*Lock. For example:

type MultiLock []*Lock

func (l MultiLock) MinTTL(ctx context.Context) (time.Duration, error)
func (l MultiLock) Release(ctx context.Context) error

func (c *Client) ObtainMany(ctx context.Context, keys []string, ttl time.Duration, opt *Options) (MultiLock, error)

[bhallionOhbibi]

About multi-lock and hypothetical partially held locks:

Theorically, you cannot hold one lock and not the other. msetnx make sure that you either set all the locks or none. Also, all locks held together share the same value and ttl. Therefore once you know that you and only you own all the locks, any of the commands (Refresh, TTL, Release) will come together nicely.

• Refresh: you own them all, you refresh them all.
• TTL: they all have the same ttl (minTTL check is really just for the show)
• Release: Same as refresh.

And for each of theses, if somehow you dont hold one lock, an error is returned and no refresh or release is done on any of the key among the multilock.

[tomaswarynyca]

@dim ping, what do you think of the changes made?

[dim]

I will take a look next week. I think most of my concerns have been addressed through the new release.lua, but need to double-check.

[dim]

It's a relatively big change, I think v0.10.0 is justified

[tomaswarynyca]

ping @bhallionOhbibi

[bhallionOhbibi]

Sorry ! I was on a very long vacation during october. I'll check your feedback soon. Thanks

[bhallionOhbibi]

I updated the code according to your feedback. I also updated the changelog. Please let me know if you're ok with the change on release.lua.

[bhallionOhbibi]

Ping :)

[dim]

Thanks for the epic effort :)

[bhallionOhbibi]

Thank you for all your feedbacks and time dedicated to this