feat(node): seictl node watch --until=caught-up — SDK-backed readiness gate

[bdchatham]

Wires seictl to the SDK's shared chain-readiness primitive so the nightly's bespoke bash catching_up probe can be deleted — the readiness logic lives once, in reusable Go.

What

• seictl node watch --until=caught-up: waits for Running, then gates on the node's published endpoints via sei.WaitCaughtUp (TM /status: height>1 && catching_up==false) and sei.WaitEVMServing (EVM eth_blockNumber bound, when the node serves EVM).
• Inputs come from .status.endpoint (read verbatim, never reconstructed). endpointsFrom unit-tested across fullNode / non-EVM / pre-publish shapes.

Why

The nightly orchestration stays seictl-CLI-driven; the win is the readiness/wait complexity moving out of bespoke curl /status | jq '(.catching_up==false) and height>1' bash into one reusable Go implementation (sdk/sei) shared by the SDK, the seitask Task steps, and the CLI.

Interim dependency note

Imports github.com/sei-protocol/sei-k8s-controller/sdk/sei (core) — a temporary mutual module require (the controller already imports seictl/sidecar/client). The seictl→controller monorepo consolidation (designed + xreviewed) later dissolves this into a clean in-repo cli→seiclient import via go.work. Verified the import adds nothing to seictl's build beyond what it already carries (it's already a k8s CLI with client-go + controller-runtime): go mod tidy made only benign indirect bumps; seictl stays at its own k8s v0.36, and sdk/sei core is stdlib-only (0 k8s/controller-runtime deps) so pruning keeps the controller's k8s graph out.

Test

go build/golangci-lint (0 issues)/go test ./seinode/... green; gofmt clean.

🤖 Generated with Claude Code

[cursor]

PR Summary

Medium Risk
New cross-module dependency and live HTTP readiness probes against published node endpoints; behavior change for automation replacing bash catching_up checks.

Overview
Adds seictl node watch --until=caught-up, a readiness sentinel distinct from phase-based --until. After the existing watch reaches Running, the CLI reads Tendermint and optional EVM RPC URLs from .status.endpoint (verbatim) and gates on sei.WaitCaughtUp and, when EVM is published, sei.WaitEVMServing from github.com/sei-protocol/sei-k8s-controller/sdk/sei.

go.mod gains a direct require on sei-k8s-controller (plus routine indirect bumps from go mod tidy). Help text and validation now accept caught-up alongside the SeiNode phase enum. TestEndpointsFrom locks the .status.endpoint field paths used by the gate.

^{Reviewed by Cursor Bugbot for commit 5e70706. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 5e70706. Configure here.}

[cursor]

Caught-up wait timeout wrong reason

Medium Severity

For --until=caught-up, after the Running watch succeeds, failures from sei.WaitCaughtUp or sei.WaitEVMServing (including when the shared --timeout elapses during those polls) are emitted via EmitStatus without the same WatchExitError mapping RunWatch uses. stderr then shows InternalError instead of Timeout, which breaks the command’s documented metav1.Status.reason contract for nightly orchestration.

 

^{Reviewed by Cursor Bugbot for commit 5e70706. Configure here.}