docs: validation-substrate design (supersedes sei-k8s-controller#143)#96
Merged
Conversation
Captures the pivot away from the merged ValidationRun CRD LLD toward a CLI-substrate model: seictl primitives (chain/rpc/load/harness/rules) composed by sugar verbs (bench/qa/shadow), with the same workload runtime contract from platform#235 already implemented in `bench up`. Coral round dispatched platform-engineer (substrate), product-manager (scope discipline), and product-engineer (cross-surface ergonomics) specialists. Synthesis: v1 ships effectively zero new code — today's `bench up` covers the seiload-nightly use case. Primitives land on demand with named triggers. Distribution: single binary, two install paths — standalone `seictl` plus kubectl plugin via `kubectl-sei` symlink (one symlink, no code change). Anti-features explicitly called out: unified validation.sei.io YAML schema, generic harness substrate, symmetric verb sets, observability- as-test-oracle in the CLI. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced Apr 30, 2026
…nt emitter
Per maintainer pushback: harness is a category error. seictl provisions
ephemeral chain infrastructure + in-cluster load generation; out-of-
cluster test binaries (qa-testing's TS suite, fuzzers, integration
tests) run wherever the engineer/CI invokes them and consume seictl-
emitted endpoints via env vars/flags. Exit code + logs are the verdict.
This is the autobake pattern, and the kubectl pattern.
Changes:
- Drop `harness` from the primitive surface (4 primitives now: chain,
rpc, load, rules).
- Drop `qa up` composite. Only `bench up` (shipped) and `shadow up`
(deferred, replayer is typed on SeiNode today).
- Reframe goals around endpoint emission as the contract for downstream
tooling.
- Add a new Design subsection ("Endpoint emission") with concrete shell
+ GHA examples showing how external test binaries consume endpoints.
- Replace the "generic harness substrate" anti-feature bullet with
"seictl as a test runner" — explicit-rejected, with the rationale.
- Update v1 ship cut table: harness/qa up moves from "Defer" to
"Out of scope (not deferred — rejected)".
- Update MCP graduation, label contract, wait semantics, references.
- New Open Question on endpoint field shape (lock the contract early).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3 tasks
bdchatham
added a commit
that referenced
this pull request
Apr 30, 2026
## Summary Follow-up to #96 locking the endpoint emission contract. Brandon directed split-by-RPC-type on the merge thread; this PR captures the shape concretely and removes the resolved Open Question. ## Contract ```json { "endpoints": { "tendermintRpc": ["http://...:26657", ...], "evmJsonRpc": ["http://...:8545", ...], "cosmosGrpc": ["...:9090", ...] } } ``` Three rules behind it: - **Split by RPC type, not flat-with-discriminator.** Most consumers know which port they want — split eliminates filter logic on the consumer side. - **Per-pod, not aggregate.** Consumers shard, round-robin, or pick. Aggregate Service URLs only allow kube-proxy round-robin. - **Full URLs (with scheme + port) for HTTP types; \`host:port\` for gRPC.** gRPC clients don't take \`http://\` prefixes — Sei's gRPC is h2c. Adding new types (\`evmWebSocket\`, \`cosmosLcd\`, etc.) is backwards-compatible. Renaming \`endpoints.<type>\` is not — locked. ## Test plan - [ ] Skim the diff - [ ] Confirm shell + GHA examples reflect the typed shape - [ ] No other harness/qa references re-emerged 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Captures the pivot away from the merged ValidationRun CRD LLD (sei-protocol/sei-k8s-controller#143) toward a CLI-substrate model: seictl primitives (chain/rpc/load/harness/rules) composed by sugar verbs (bench/qa/shadow). The runtime workload contract from sei-protocol/platform#235 is kept verbatim and already implemented in `bench up`.
Key decisions captured
Anti-features (deliberate)
The doc explicitly enumerates what the LLD's gravitational pull would tempt us to build:
Process
Coral round dispatched three specialists in parallel — platform-engineer (substrate), product-manager (scope discipline), product-engineer (cross-surface ergonomics). Outputs synthesized inline. The PM's "v1 ships nothing new" stance won on scope; the platform-engineer's label contract + peer-discovery mechanism won on substrate; the product-engineer's MCP composite-as-tool / kubectl-sei prefix won on distribution.
Test plan
🤖 Generated with Claude Code