Possible race condition in plugin migrations
The following discussion from !1209 (merged) should be addressed:
- @gmoshkin started a discussion:
// FIXME: currently it possible that migrations will be initiated from
// 2 different instances simultaneously which can break some invariants.
// What we should do is to introduce a global lock in _pico_property
// such that the client first checks if the lock is acquired, then does
// a CaS request to acquire the lock and only after that starts doing
// the migrations, and releases the lock at the end.