tests(pd): automated migration testing

[conorsch]

Describe your changes

Adds intra-CI standalone testing of migration behavior. This can be run locally on developer workstations, and also in CI. The job is currently taking >20m, and requires a lot of disk space, but it's worth it for the assurance.

Building on the smoke-test rewrite to use process-compose, let's script the migration process, so that we can test
current HEAD of the monorepo against a prior tagged version, and validate that necessary migrations are in place.

One possible approach is to fetch prebuilt binaries from uploaded artifacts on Github. That's fine for pd, but doesn't work for running the smoke tests, due to client/server incompatibility. Therefore we'll clone the entire repo in a git-ignored subdir, and build the old binaries there. Heavy, but reliable.

Adds a new rust crate, strictly for running the migration-test suite of integration tests, which is very similar in nature to the already-existing network-integration tests AKA smoke tests. Copy/pastes a lot of code from the smokes, which we can always factor out into reusable utils, but not bothering with that right now.

Issue ticket number and link

Refs #4323.

Checklist before requesting a review

• If this code contains consensus-breaking changes, I have added the "consensus-breaking" label. Otherwise, I declare my belief that there are not consensus-breaking changes, for the following reason:

  Testing-only, no changes to application logic.

Review, running locally

Check out this branch, and run just migration-test. It'll take a while to build! The test as written duplicates the local repo, so plan on ~100GB of disk space utilization.

[hdevalence]

Confoundingly, I'm getting a proto incompatible error, about missing AuctionParams.

I think this was an actual bug #4338. We shouldn't assume the migrations actually work until we do the testing. See also #4340

[conorsch]

Marking as ready for review, so I can stop rebasing it. 😅 The immediate consequence of merging this PR is that CI runs will take >20m again, but I'm happy to dial that back and rely on the cron schedule. For now, though, having it run on every PR will give us confidence about the state of migrations as we push toward #4497.

[cratelyn]

i'm going to somewhat cautiously hit the "approve" button here. thanks so much for your work to give us automated coverage of migrations. that's tremendously valuable work.

we should have automated migration testing, and i want that sooner than later, but.. i'm a little nervous about how these are shaped. they feel somewhat heavy on boilerplate yaml, and there's a lot of shell glue surrounding the actual test.

i think a lot of the logic here could also be (and is, in the case of delegating) covered by mock consensus tests, but the broader vision of having integration tests that run a real CometBFT node feel topically relevant for migrations.

i've left a smattering of comments below. i'm happy to see this merge before those things are addressed, being acutely aware that this life easier while managing upgrades ❤️ ...but maybe file a tracking ticket(s) for following up on these points.

[cratelyn]

nice 👍

[cratelyn]

Suggested change

	fn main() {
	println!("Hello, world!");
	}
	//! This crate is used to group migration tests, this binary does not do anything.
	fn main() {
	println!("Hello, world!");
	}

[cratelyn]

this empty file was a little confusing for a moment, so a comment might help signal that this main.rs is only here as a stub.

[cratelyn]

Suggested change

	name = "migration-test"
	name = "penumbra-migration-test"

a nit, but this should probably have the same penumbra- prefix as other crates in the workspace.

[cratelyn]

i appreciate these thoughtful breadcrumbs for future tinkering. nice 🪙

[cratelyn]

it's a bit difficult to understand what this is testing. my understanding is that this is running the smoke tests on the v0.76.0 release? i don't think i am connecting the dots to how this is exercising migration logic.

it'd be helpful to (a) name these files more commucatively (possibly via putting them in a folder together), and (b) putting a comment at the top of the file, in place of the boilerplate there right now. this isn't a configuration for running migration tests, it is a migration test, aiui.

[cratelyn]

...as i've read this more, i see these are phases. i'll mark this as resolved, but i did find coming to this understanding to be difficult. 😓

[cratelyn]

Suggested change

	// CLI interfacts for the pcli binaries.
	// CLI interfacts for the pcli binaries.

i think a sentence got garbled here!

[cratelyn]

Suggested change

	#[cfg_attr(not(feature = "migration-test"), ignore)]
	#[cfg(feature = "migration-test")]

ditto, another conditional compilation attribute we can simplify

[cratelyn]

Suggested change

	[dependencies]
	anyhow = {workspace = true}
	directories = {workspace = true}
	once_cell = {workspace = true}
	penumbra-keys = {workspace = true, default-features = false}
	serde = {workspace = true, features = ["derive"]}
	serde_json = {workspace = true}
	serde_with = {workspace = true, features = ["hex"]}
	toml = {workspace = true, features = ["preserve_order"]}
	tracing = {workspace = true}
	tracing-subscriber = {workspace = true, features = ["env-filter", "ansi"]}
	[dev-dependencies]
	assert_cmd = {workspace = true}
	predicates = "2.1"
	regex = {workspace = true}
	tempfile = {workspace = true}
	[dev-dependencies]
	anyhow = { workspace = true }
	assert_cmd = { workspace = true }
	directories = { workspace = true }
	once_cell = { workspace = true }
	penumbra-keys = { workspace = true, default-features = false }
	predicates = "2.1"
	regex = { workspace = true }
	serde = { workspace = true, features = ["derive"] }
	serde_json = { workspace = true }
	serde_with = { workspace = true, features = ["hex"] }
	tempfile = { workspace = true }
	toml = { workspace = true, features = ["preserve_order"] }
	tracing = { workspace = true }
	tracing-subscriber = { workspace = true, features = ["env-filter", "ansi"] }

these should all be dev-dependencies, since this crate only contains tests.

otherwise, cargo build --package migration-test will mean building all of these crates, despite the main.rs being empty and the --features migration-test not being provided

[conorsch]

Downgrading to draft until I address the review comments. Overall, the shape of this approach is rather heavy. Will continue to discuss testing options with @cratelyn, so that we maximize what's possible with mock-consensus, but still retain integration tests for migration behavior. Notably, this PR does not catch bugs in restarting a multi-validator setup post-upgrade, e.g. #4508.