Boosting an event on a federated platform results in Beam being killed by the OOM manager in the kernel
Description
Crash when an event which appears on a federated partner service is boosted.
Steps to Reproduce
- On the Mobilizòn server open the log in follow mode (tail -f, journalctl -f ..., etc)
- In a federated Mastodon or Miskey instance find an event you haven't already shared
- Click Boost, share, or whatever your instance calls it.
- Watch as a number of GET requests equal to your followers arrive in the logs
- Watch as the Beam is killed by the OOM killer after a certain number of requests
Expected behaviour:
Mobilizòn serves all of these requests, and goes on about its business, with the beam
Actual behaviour:
mobilizon.service: A process of this unit has been killed by the OOM killer
Reproduces how often:
100% under the conditions described
Versions
5.1.2
Additional Information
I'm going to attempt a workaround for this by getting Nginx to cache the events pages. But my primary goal is to find whatever is causing the high memory use, and fix it.