[Dylan16807]

> Set overcommit policy and malloc would just fail, which is handled in mysql and would abort the query instead of invoking oom killer.

So many things depend on so many gigabytes of overcommit, that seems like a pretty bad way to go about it in the general case.

> And a crashing service should get restarted, involving journal file recovery which is consistent state.

I think you're agreeing with me with this sentence?