MongoDB Implementation?

[mhsimkin]

Hi.

Do you know if anyone has built a MongoDB implementation?

Thanks

[madelson]

Currently there is no MongoDB implementation included in this library.

Looking at NuGet, there is at least one package available that does distributed locking with Mongo (https://www.nuget.org/packages/DistributedLock.Mongo/). That package has no relationship with this library and I have no experience with it so I can't speak to its features or quality but I would definitely recommend giving it a try if you're looking for a Mongo solution in the near-term.

I would definitely be open to adding a MongoDB implementation to this library in the future and I will use this issue to track that feature request. From some quick googling, there doesn't seem to be a "standard" pattern that people follow for locking in Mongo. https://github.com/square/mongo-lock built by Square (written in Go) offers a straightforward-looking approach and could be a good model.

If you would be interested in contributing an implementation to this library, let me know and we can discuss what the steps would look like. It is typically a good amount of work to build out a new provider from scratch.

[mhsimkin]

@madelson Thank you for the links. What I found interesting about this specific library was support for Azure, MySql, SQL Server, Local (File, Wait Handle), and Oracle. My employer uses all all of those and Mongo, and a few other NoSql stores.

I've spent the last two days, trying to get my head around how the library works, so I could implement a Mongo provider. Conceptually it all makes sense. Understanding the actual implementation is not going well. All the interfaces are a bit overwhelming.

When I dig into implementation specifics, I get total confused with all the interfaces and the use of internal classes. Having a reference implementation, even if it a duplicate of an existing one, that just uses the public interfaces/classes would make it easier to build a Mongo implementation.

Talking Mongo specific, depending upon the version of the server there are different ways to achieve the same result. Using a BSON document with a two phase approach works on all versions.

The collection for the locks would have a unique index on the lock name field. The document that represents the lock would have several fields including lock name, id, lock state, etc.

When a lock is requested I plan on inserting a document with the lock name, unique id and an initial state of "initializing". If successful, I can then finalize the lock by setting lock state to "acquired", duration to expiration, last renewal time and a few other fields.

If there is already a document for the lock, a duplicate key error will be thrown. Then I could figure out if the existing lock is valid or expired and handle appropriately.

I haven't come to a decision on whether to use a separate sweep task to manage lock expiration.

As for contributing, that might be an issue. I have certain constraints from my employer. We are not prevented from using Open Source, At the same time, code I develop as part of my job is not mine to contribute.

End result, I'm really interested in seeing a reference implementation that doesn't rely on internals.

Cheers.