Daijiro Fukuda

> > Why do we need this deletion? > > Whey the key for `@tails` is path, old TailWatcher can be dropped just by `@tails[path] = tw`. But when the...

If we use the inodes for the keys of `@tails`, shouldn't we recreate the watchers by the update of the paths for each inode? Otherwise, it's not symmetrical. (I don't...

Sorry, maybe I'm saying something wrong... I will reconsider this asymmetry.

Now I understand the cause of this asymmetry. It is caused by the watcher getting a new inode from the watcher's (old) path to determine the update.

I have still some concerns about changing the key... > Now I understand the cause of this asymmetry. It is caused by the watcher getting a new inode from the...

Thanks, everyone. I am wondering if the key of `@tails` is not a fundamental part of this problem. I'm still not sure, but I commented in https://github.com/fluent/fluentd/issues/3614#issuecomment-1580130292 on what I...

> We understand what's wrong and why this PR can fix the issue. We are talking about whether there are more ideal solution or not. Yes, I too think so....

> Sorry, I couldn't understand what is "asymmetry". Sorry, I just wanted to say that `path` as the key is reasonable for the current structure, regardless of `follow_inodes`. I think...

Thanks to this PR and #4191, I am starting to see problems that I didn't understand before. Thank you very much. I would like to think about it some more.

@masaki-hatada Thanks for considering my concern! >> The important point is that the current TailWatcher is updated based on its path by TimerTrigger or StateWatcher, and we don't want to...