[cstns]

If the schema agent can't be reached, its status keeps coming back as "still getting status", which the broker page showed as "starting" forever, with no end and no explanation.

Now that only holds for a short window. Past it the page stops waiting, drops the broker back to an idle state so the toggle can be used again, and shows a single toast saying the agent couldn't be reached. The timed-out state is UI-only and clears on a refresh, a successful status, or another attempt to start, so nothing is persisted and the backend is untouched.

While here, the schema monitor toggle and the connect toggle both fired the start request without waiting on it, so a failed start went unnoticed. Both now wait on the request and surface a failure the same way.

One thing worth knowing for review: the team broker status badge is hardcoded to connected and its toggle reads on/off from the agent's connected flag, so the visible "stuck on starting" is really on third-party brokers. The fix sits on the shared status poll, so it covers both: the third-party spinner stops, and the team-broker path times out cleanly too.

Closes #8996

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.46%. Comparing base (40b10a5) to head (4f87bfc).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #9001   +/-   ##
=======================================
  Coverage   79.46%   79.46%           
=======================================
  Files         475      475           
  Lines       26018    26018           
  Branches     6934     6934           
=======================================
  Hits        20674    20674           
  Misses       5344     5344           

Flag	Coverage Δ
backend	79.46% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[andypalmi]

Nice fix for the endless spinner, and good call awaiting the start requests. A few questions:

1. After the timeout the page shows the broker as idle, but the backend row is still running. toggleAgent decides from the backend state, so the first click after a timeout goes down the running branch and calls suspendBroker, not startBroker. I checked this with the page's methods and mocked responses: after the timeout, one click makes 0 start calls and 1 suspend call, and the UI goes to suspending. So "retry" needs two clicks. Could toggleAgent start when the page has timed out, since the toggle is showing as off?

async toggleAgent () {
    this.loadingState = true
    try {
        const state = await brokerAPI.getBrokerStatus(this.team.id, this.$route.params.brokerId)
        if (state.state === 'running' && !this.startingTimedOut) {
            await brokerAPI.suspendBroker(this.team.id, this.$route.params.brokerId)
            this.brokerState = 'suspending'
        } else {
            // Fresh attempt to start, so give the agent the full window again
            this.resetStartingTimeout()
            await brokerAPI.startBroker(this.team.id, this.$route.params.brokerId)
            this.brokerState = 'starting'
        }
    } catch (err) {
        ...
    }
}

For a third-party broker whose row is running, the start route sends the agent a start command, so this is a real retry. A test that clicks the toggle after a timeout and asserts startBroker was called (and suspendBroker was not) would lock it in.

2. Is 30s enough for STARTING_TIMEOUT on Kubernetes, where a cold agent pod (image pull, scheduling) can take longer? The user would see "Could not reach the broker agent" and then watch it flip to connected a moment later. A longer window might avoid the false alarm.

3. Nit: the new const sits between two imports, and BrokerStatusBadge is imported below it. Moving the const under all the imports would read better.

[cstns]

1. Yep, that was wrong. After a timeout the toggle shows off but the row is still running, so the click went down the suspend branch. It now starts when the page has timed out, and there's a test that clicks after a timeout and checks for one start call and no suspend, plus one that a healthy running agent still suspends.

2. Fair point. The k8s driver returns as soon as the Deployment and Service exist, it doesn't wait for the pod, so a cold pull sits entirely inside the window. Bumped it to 60s. We'll revise if it turns out to still bite someone.

3. Moved under the imports.