I was reading the documentation for mongdb, and I'm trying to understand what is necessary for "read your own writes" consistency before Mongo 3.6's causal consistency sessions were introduced.
According to this page: https://docs.mongodb.com/manual/reference/read-concern-majority/#read-your-own-writes
Prior to MongoDB 3.6, you must have issued your write operation with { w: "majority" } write concern and then use either "majority" or "linearizable" read concern for the read operations to ensure that a single thread can read its own writes.
However, in the example that they provide on that same page above (https://docs.mongodb.com/manual/reference/read-concern-majority/#example), there seems to be information that does not agree with this. Specifically, they have a chart (that i've copied below):
In this diagram, at time t3, the write is acknowledged. After receiving acknowledgement of the write, a majority read concern at t4 to either of the secondarys would return an old value, in contradiction with the first claim that "a single thread could read its own writes" as long as you use majority write + read concerns.
If the statement was amended to be "read from the primary with majority or linearizable concern", I think it would become true... But, I wonder if maybe I am misunderstanding something.
Could someone provide insight on:
a) What this documentation means, and
b) How to get read-your-own-writes semantics before 3.6's causal sessions?
Thank you!
I found your ticket when I was about to raise the exactly same question as yours. I think this statement from MongoDB Documentation is incorrect. Double "majority" without the causal consistency session cannot guarantee "read your own writes" semantics. As I understand it, before 3.6, read concern linearizable can provide this semantics with write concern majority.