Add warning about legacy snapshot

[alexbozhenko]

Follow-up to #7479
There we fixed the default, but there is no test that we do not revert to the old behavior.
It is possible to imagine a scenario where some regression/race condition is introduced, and we fall back to legacy JSON again.

Add a warning that we should never see. And if we do see it, we would immediately know there is a problem.

Signed-off-by: Alex Bozhenko alexbozhenko@gmail.com

[wallyqs]

LGTM, a warning here just in case would narrow down investigating this (or deleting the code for legacy snapshot)

[neilalexander]

I'm not really convinced that there's much point in logging this. It isn't meaningful or actionable to users, plus it will just result in an awful lot of log spam during a one-step rolling upgrade from a very old version to a current one.

[ripienaar]

Agree with Neil this kind of pointless log spam is already at a quite annoying level and this ads no value to end users.

[alexbozhenko]

@neilalexander @ripienaar
We do not really test that binary snapshots are used and legacy are not, do we?
That was quite an expensive way to discover this edge case.
So I was thinking if there is a regression later, it would potentially get unnoticed until there is another OOM.
Do you have suggestions for other ways to prevent legacy snapshots from showing up again?

[neilalexander]

I would prefer that we add an Antithesis assertion instead, that way it is reported by tooling designed to uncover these edge-cases instead of just creating noise in the logs.

[MauriceVanVeen]

I would prefer that we add an Antithesis assertion instead, that way it is reported by tooling designed to uncover these edge-cases instead of just creating noise in the logs.

#7486