A contradiction in 13.2.1

Dane Foster <studdugie@xxxxxxxxx> · Wed, 27 Jan 2016 17:23:09 -0500

Hello,

I'm trying to understand concurrency in PostgreSQL so I'm slowly reading through chapter 13 of the fine manual and I believe I've found a contradiction in section 13.2.1.

My understanding of the second sentence of the first paragraph is that read committed mode never sees "changes
    committed during query execution by concurrent transactions". For example let's assume two transactions, A & B, and the following:
A started before B
B starts before A commits
My understanding of the second sentence means that if A commits before B then any updates made by A will continue to be invisible to B because B's snapshot was before A committed. Now if I'm wrong about this then there is no contradiction forthcoming.
The final sentence of the first paragraph is where I find the contradiction. It says: "Also note that two
    successive SELECT commands can see
    different data, even though they are within a single
    transaction, if other transactions commit changes after the
    first SELECT starts and before the
    second SELECT starts"
.

So the mental model I've built based on the first four sentences of the first paragraph is that when a transaction starts in read committed mode a snapshot is taken of the (database) universe as it exists at the moment of its creation and that it's only updated by changes made by the transaction that created the snapshot. So for successive SELECTs to see different data because of updates outside of the transaction that created the snapshot is a contradiction.

Now my guess is that I'm thinking about it all wrong so if someone in the know could shed some light on where/how my mental model breaks down I would appreciate it.

Regards,

Dane