Fix statement store deadlock

[gui1117]

We are doing 2 reads in one time. But this is not good.

Per parking lot doc:

/// This lock uses a task-fair locking policy which avoids both reader and
/// writer starvation. This means that readers trying to acquire the lock will
/// block even if the lock is unlocked when there are writers waiting to acquire
/// the lock. Because of this, attempts to recursively acquire a read lock
/// within a single thread may result in a deadlock.

[acatangiu]

why do you need to keep the write lock for the whole duration of this function?

[gui1117]

somehow I thought unnamed variable would still be dropped at the end of their "scope" but indeed it is dropped instantly.

Everywhere in the code it seems to lock the index until the operation of the DB is done. But not here.

[gui1117]

I just removed this change then, it is not necessary.

[gui1117]

We can have this situation:

• index and db have statement s1 in expired.
• maintain starts: index has no expired statements, db still have s1 in expired
• other operation in between: s1 is submitted then s2 is submitted in the same channel as s1: index has s2 as statement, s1 as expired statement, db has s2 as statement and s1 as expired statement
• maintain finishes: remove the statement s1 from expired in db: now index has s2 as statement and s1 as expired, but db only has s2 as statement. They are inconsistent, but it is not harmful AFAICT.

[AndreiEres]

Suggested change

	let mut index = self.index.write();
	let deleted = index.maintain(self.timestamp());
	let deleted: Vec<_> = {
	let deleted = self.index.write().maintain(self.timestamp());
	deleted.into_iter().map(|hash| (col::EXPIRED, hash.to_vec(), None)).collect()
	};
	let (purged, active, expired) = {
	let index = self.index.read();
	(deleted.len(), index.total_size, index.expired.len())
	};

[gui1117]

I just removed this change, indeed it is not needed the write lock is dropped instantly so there is no deadlock.

[AndreiEres]

#9868

[AndreiEres]

I would prefer to continue work in #9868 where I can add the other stuff related to statement-store performance

[gui1117]

closed in favor of #9868