| Setting aside the fact that the document storage model is rarely a good fit for most applications (and the primary reason I had once been a fan of MongoDB, Meteor, has gone the way of the dodo), the Jepsen distributed system audits of MongoDB's behavior under load were damning for much of the past decade. They've improved in recent years as shown below, but this remains a company that allowed these types of bugs to be considered acceptable, and it's still easy to use MongoDB in an unsafe way. 2013: https://aphyr.com/posts/284-jepsen-mongodb > To recap: MongoDB is neither AP nor CP. The defaults can cause significant loss of acknowledged writes. The strongest consistency offered has bugs which cause false acknowledgements, and even if they’re fixed, doesn’t prevent false failures. [N.B. I thought this was just an edge case until it happened to me in production. Luckily, I had separate storage of ground truth, but I very easily could have been in real trouble.] 2015: https://aphyr.com/posts/322-jepsen-mongodb-stale-reads > In this post, we’ll see that Mongo’s consistency model is broken by design: not only can “strictly consistent” reads see stale versions of documents, but they can also return garbage data from writes that never should have occurred. The former is (as far as I know) a new result which runs contrary to all of Mongo’s consistency documentation. The latter has been a documented issue in Mongo for some time. We’ll also touch on a result from the previous Jepsen post: almost all write concern levels allow data loss. 2017: https://jepsen.io/analyses/mongodb-3-4-0-rc3 > MongoDB has devoted significant resources to improved safety in the past two years, and much of that ground-work is paying off in 3.2 and 3.4. Dirty reads, which we covered in the last Jepsen post, can now be avoided by using the WiredTiger storage engine and selecting majority read concern. Because dirty reads can be written back to the database in read-modify-update cycles, potentially causing the loss of committed writes, users of ODMs and other data mappers should take particular care to use majority reads unless making careful use of findAndModify. |