Backend down is not Braid's status: a dot for the server, and backend cells that outlive the run #391
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
What happens
Backend downin red. A run's backend isn't Braid's status, and it says so on every page./api/backendslists only live runs, so the cells go when the run ends, whatever the backend is doing.Fix
Backend down. This reverses #387, whose reason goes once the word does.vLLM qwen3.8-27b, for any run, live or finished, with the same dot and an info hover.Backend Planning Repair.Done when
Reported by the owner from the console, on run 172.
.eyebrowand five others #392.eyebrowand five others #392Shipped in PR #394, production
382c998.Backend down. This reverses #387.vLLM qwen3.8-27b, for any run, live or finished, with the same dot and an info hover. Planning and Repair share its caption, or have a cell of their own when they differ.BRAIDkeeps its capitals, as the wordmark reads.The cause of the missing cells:
/api/backendslists only live runs, so a finished run lost them whatever vLLM was doing.Evidence, on production just after the deploy, with vLLM really down: run 172's bar reads
Backend Planning Repair,vLLM qwen3.8-27b, statedown, whyIt is not answering (ConnectError), and the Braid dot isok. That is this issue's done-when, on the real thing.Also
tests/api/test_run_backends.py,ui/src/components/chrome.test.ts, and new journeys inui/e2e/backends.spec.ts. One of those tests caught a Planning backend on another server taking the wrong profile's name, now fixed.Not proved: that the journeys fail on the old build, as they were not run against it. No real run on a separate Planning backend has been seen in the bar; that layout was read with a faked route, and its grouping is held by backend tests.