MongoDB - WiredTiger Snapshots vs. Locking

问题:

I am not completely understanding how these two features relate to one another in a (WiredTiger) MongoDB program:

1) WiredTiger Snapshots

2) Data Locking

If each read operation using the WiredTiger engine is, at read-time, provided with a database level 'snapshot' (so as to create consistency (the C in ACID), why then, do we also need locking? Let's use an example.

I perform a query at the Document level (a read operation). Okay, so I know I get the database level snapshot, so that my data is consistent EVEN IF another user is concurrently writing to that same Document, updating it.

So at this point, what is the use for having a Shared-Lock on that document, which is blocking all write (exclusive) operations on that document until the Shared-Lock is released? What could possibly go wrong in writing to that Document concurrently while I'm reading it, if I am, in fact, using a snapshot of the Document that was provided to me at read-time? Why would I care if the Document is locked during my read-operation period or not? I already have my (consistent) data from that point-in-time, no?

I'm obviously missing a key concept here... Any help?

Thanks.

回答1:

You are right that the read operation will acquire a snapshot. When using the WiredTiger storage engine, MongoDB does not lock individual documents for either reads or writes. Instead WiredTiger uses Multi-Version Concurrency Control, MVCC. When performing an update of a document, that update will succeed as long as the document still has the same version as it had when acquiring the snapshot. If not, WiredTiger will return an error (WT_ROLLBACK) indicating that the update had write conflicts. In this case, the update will abort and all pending changes are undone. MongoDB will then transparently retry the operation.