Why not to use shared memory for in RAM DB

[onorua]

The problem

As soon as we increased memory to around 32GB+, we noticed quite a big startup time due to journals restoration, even we have sent TERM signal properly and waited for normal shutdown. Startup takes couple of minutes (up to 10 actually), which is noticeable on graphs.
That is because prometheus is trying to get all the data are to be in RAM into memory.

shared memory proposal

it is possible to attach shared memory from another process, which will speedup the startup time, because you will not need to load journals from disk to RAM, but rather attach already available memory.
This will require checks for data consistency on startup of course and fallback to the current restoration from journals.

What do you think about this functionality?

[beorn7]

I'd expect this to require a lot of effort to get right, and then additional operational effort to run. And it only covers a relatively rare corner case (i.e. where a SIGHUP is not enough but it's not a complete restart of the machine/container running Prometheus).

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.