[jschmidt-icinga]

First preview version with only GraphiteWriter as a basis for discussion (not even that for now).

Currently this works by Cancelling the PerfdataWriterConnection when a configurable queue size is reached. Then inside the queues the entries cancel themselves until we're under the threshold again (within a certain amount of hysteresis).

This is a bit inelegant and awkward, but the advantage is that when the backend reappears the beginning of the gap will line up with the time it went out and the maximum amount of new perfdata will be sent right away.

I've also set the default value of the threshold to 1000000 entries for now, which is 1/10 the maximum queue size and on my testing setup results in ~1.2GiB of memory usage, though this is obviously subject to vary based on the checks being run and other factors and will vary even more between the individual writers.

I do think that we absolutely have to make this value configurable for the user because there is no reliable way to use any memory pressure heuristic for this, due to the diverse environments Icinga 2 will run in. I think the default can be pretty generous (or there might even be an argument for keeping it off by default), because we can just rely on the documentation and community to refer users that run into this issue to the solution.

TODO

• Get approval on and finish the design
• Implement the remaining writers
• Fine-tune default values
• Documentation

Fixes #10630 (eventually).