seq-kv reads never return final state for most recent write/cas about maelstrom HOT 5 OPEN

Not exactly: there's both a probabilistic and a deterministic way to use seq-kv here. My thinking in designing this challenge was that folks would have this exact kind of "aha!" moment realizing that sequentially consistent stores didn't guarantee recency, and that they'd have an easy-but-inelegant way (doing lots of retries, as shown here) to work around that staleness. Because it's probabilistic and feels a bit inelegant, I was hoping it might prompt people to ask "wait, is there a more elegant, deterministic way I could do this?"--and that might lead them to discover the trick, and in the process gain (or show off) a deeper intuition about sequential consistency.

from maelstrom.

This may actually be a bug (er, sort of, but it's certainly not consistent with lww-kv) in seq-kv! I was sprinting to write these challenges and didn't have time to fully go through and test this one. Give me a little bit and I'll see if I can get that fixed for you.

from maelstrom.

HA! There was a bug of sorts. seq-kv was doing something legal but v frustrating where reads and other re-orderable operations would converge on the next to latest state, but would never consider the latest state unless forced to. Version 0.2.3 should converge if you just spam it with requests. This affects #39 too.

from maelstrom.

@aphyr, is it correct to say that if we want all the nodes to converge to the same value, using a sequential store gets there probabilistically, and a linear store deterministically? I'm wondering a bit what the pedagogy in challenge 4 is - certainly I've realised that sequential consistency is much looser than you'd think at first!

from maelstrom.

@aphyr Is the "trick" in question the one that you mention in this comment #39 (comment)?

Because later in the same issue thread, someone asked if this "trick" was a property of sequential consistency (#39 (comment)), and your response seemed to indicate that this is not a property of sequential consistency, but a property of your particular implementation of data-store.

If this is the case, I definitely think you should clarify the challenge description to make it clear that the data-store has some additional properties beyond sequential consistency.

from maelstrom.