upgrades: 🚘 migrations should be idempotent

[cratelyn]

Is your feature request related to a problem? Please describe.

it would be nice if migrations were idempotent, to help make operator's lives easier while performing chain upgrades.

Describe the solution you'd like

it would be nice if we could examine a version number (q: which?), and only apply a migration if the version number was below some threshold, and returning early if not, rather than writing migrations in such a way as to be idempotent.

[conorsch]

we could examine a version number (q: which?)

Ahoy: https://github.com/penumbra-zone/penumbra/blob/v0.73.2/crates/core/app/src/lib.rs#L24 We also reference this value in the repo's COMPATIBILITY.md, but not yet in local state, which would be supremely useful for migrations.

[cratelyn]

is #4373 is a duplicate of this, @erwanor?

[erwanor]

I don't think it's a duplicate? This is about adding migration-scoped check against specific protocol versions, the other ticket is about changing the halt signaling mechanism and adding a broad "is chain halted" check to pd migrate.

[cratelyn]

agreed, i marked it as resolved because i was able to figure that out. thank you!

[hdevalence]

I think that #4373 would close this, since the migrate command would bail out if the halt-bit is unset, thus making all migrations idempotent at the top-level.

[cronokirby]

Closed in #4373.

[conorsch]

Our migrations are not currently idempotent; see reports in #4432. At the very least, we should be exercising rerunning a migration and confirming no error in automated testing #4323 to verify it doesn't error.

[hdevalence]

I don’t see any evidence in that thread that there is an idenpotency issue.