Conflicts getting ahead of queued pushes?

[michielbdejong]

this starter-kit bug looks like a problem with the library. It is easy to replicate, @silverbucket and I have both seen it happen.

It looks like maybe there are conflicts coming in while outgoing pushes are still being queued. Maybe they're queued in https://github.com/remotestorage/remotestorage.js/blob/master/src/cachinglayer.js#L306-L315, maybe in the commit-cache, maybe just in sync tasks. Does make me think that maybe we have too many queues! ;) Should nonetheless be easy to fix, if you just reproduce it and trace what happens.

[silverbucket]

Sounds complicated :) I don't know enoug of this area of the code (and don't have a lot of expereince with syncing concepts in general) to be able to help out at the moment... but I whole-heartedly agree that this area of the code is overly complex / too many code paths.

[michielbdejong]

@silverbucket You added the ready label but then @skddc removed it... This is quite serious though, if it even makes the starter-kit hello world example fail. We should probably find time to fix this. Do you want to pair on this?

[raucao]

I removed it, because we don't use that label. It was automatically added by a tool, that only @silverbucket uses personally.

[silverbucket]

yeah, sorry about that - i've since disabled that feature (was doing it automatically)

[karlb]

Any updates on this? I think I'm still experiencing this problem with v2.0.0-beta.6 (but also earlier versions).

[raucao]

Wow, this is one is ancient! 😳

@karlb Could you provide an example to reproduce it perhaps? The starter kit that the bug has initially been created for has been deprecated for a very long time.

[karlb]

I don't have a minimal example, but when I type quickly at https://static.karl.berlin/doagain/ while being signed in to my 5apps remote storage, I get a conflict despite my browser being the only device sending writes. This sounded similar to this issue, but without further research it is just a guess.

[raucao]

Oof, that app is doing immediate writes to the same document on every keystroke with zero debouncing. That's definitely never a good idea, aside from potential race conditions in the library!

However, I think the bug's explanation is that outgoing pushes/PUTs to the server are queued up for successive sync immediately. And when there's a push still outgoing that hasn't updated the current revision (ETAG) yet, then the already queued next push would use the outdated ETAG and thus fail. (Just recapping the issue to be solved).

[karlb]

Oof, that app is doing immediate writes to the same document on every keystroke with zero debouncing. That's definitely never a good idea, aside from potential race conditions in the library!

You are totally right of course and I also noticed while looking at the problem. I just wanted to understand the conflict problem before, so that I don't just hide the problem. Now that I improved my conflict handling and the bug is at least tracked properly, I'll go and add the debouncing.

Thanks for looking into it and your (and everyone else's) work remotestorage!