Bridge crashes after significant QUIT buildup

[Half-Shot]

The bridge has recently started to crash more after it sees enough QUITs that it gets horribly backed up in requests, and is quickly killed. The logging doesn't suggest much other than a huge amount of quits happening in a short space of time.

We have the QuitDebouncer which runs inside the IRCHandler, which would hopefully take care of the increase however it doesn't appear to be having much effect.

[Half-Shot]

Having thoughts on this, we might be debouncing too late and too inefficiently. The solution may be to deliberately drop QUITs to the floor as soon as we see them and as soon as we are above a threshold. Perhaps we could even store a set of channels we have seen a QUIT from, and then mark them as requiring a names refresh once the netsplit is over.

[Half-Shot]

The proposed solution is #1091

[Half-Shot]

We've not seen issues since.