rancher/dashboard PR 19764 (codyrancher) at head ad9b8b26, tested 2026-11-04. It fixes your issue 19761: outside Cluster Explorer, every side nav group collapsed whenever the nav was rebuilt.
Preview: rancherpr19764r.mf-created-dashboard.ui.rancher.space, built from ad9b8b26, backend mf Rancher. Baseline: the shared dashboard-master preview. I confirmed the PR's currentNavState is on the running PR build and absent on master.
RANCHER_PASSWORD). The login page defaults to OpenLDAP, so click "Log in as a local user" first.default in the namespace filter. Master collapses all four.The namespace and workspace choices save to the shared admin's prefs when you do this by hand. My runs answered every write in the browser, so nothing changed.
| Case | master | PR 19764 |
|---|---|---|
Harvester registered in Rancher (harvester-118), the real PR build with no hand patch: all 4 groups open, header namespace dropdown to default | All 4 collapse | All 4 stay open |
Language switch to zh-hans, Cluster Management, Providers + Advanced open | Both collapse | Both stay open, labels translate |
A schema arrives (the real allSchemasIds watcher): Cluster Management Advanced, Fleet Resources | Both collapse | Both stay open |
| Group expanded inside the 500ms debounce window, before the rebuild runs | Collapses | Stays open |
| Direct load of a Cluster Management page under Advanced, then a rebuild | Active group collapses | Stays open |
| Direct load of a nested Explorer page (More Resources > API), then a rebuild | API collapses, parent stays | Both stay open |
| Fleet: jump-to "Bundle" with Resources closed, then a rebuild | Jump opens it, rebuild collapses it | Stays open |
| Fleet: collapse-all button after a rebuild | Button gone, nothing is expanded | Still there. Click closes all, focus moves to the first header, stays closed through another rebuild |
| Keyboard: Enter on Advanced, Space on Providers, focus back, rebuild | Focus stays, aria-expanded goes false on both | Focus stays, aria-expanded stays true on both |
Explorer with localStorage full (setItem throws) | Rebuild collapses the group | Stays open, a small bonus of the fix |
| Standard User: Cluster Management, Providers + Advanced open, rebuild | Both collapse | Both stay open |
| User Base: same walk | Both collapse | Both stay open |
| Case | Both builds |
|---|---|
Cluster switch with a rebuild still queued: local → created-dashboard-2 in the same tick | Nothing from local reaches the new cluster, and local restores on return |
local Explorer → Fleet → created-dashboard-2 Explorer | Nothing crosses, no Fleet names in the saved state |
| Harvester → Explorer → Harvester on one real cluster | No shared names, no Harvester names saved in Explorer. Harvester comes back collapsed on both, nothing is saved for it |
| Active group collapsed by hand, then a rebuild, then a route change | Stays closed through the rebuild, the route change reopens it |
Starred through the real addFavorite / removeFavorite | Starts open, stays collapsed once collapsed, other groups untouched |
| Explorer nested group opened by hand, rebuild, reload | Kept |
Saved state corrupted: "oops", [1,2], 5, {bad, null | Page loads, groups still work |
Explorer subgroup whose schemas go away and come back (inUse_K3s) | Returns open from localStorage |
| Fleet group whose schemas go away and come back (Resources) | Returns closed: the tree lost it and nothing is saved (card 4) |
Preferences and back, which remounts SideNav | Cluster Management Advanced closed, Explorer restored (card 4) |
Standard User as a cluster member on created-dashboard-2 Explorer: 3 groups open, rebuild, reload | All kept |
| Fleet for Standard User and User Base | No collapsible groups, so nothing to carry. User Base has no Explorer here, it lands on Home |
| Mutation of the fix, in a fresh clone | SideNav.test.ts (28 tests) |
|---|---|
stampNavState back to load() (master's code) | 3 fail |
| Precedence swapped, storage outranks the tree | 1 fails |
| Storage ignored in the merge | 5 fail |
| Only top-level groups carried, nested ones dropped | 4 fail |
Only true carried | 3 fail |
Unmodified, the head passes all 28. The wider set (shell/components/nav, SideNav, shell/composables) passes 47 suites, 918 tests. The file was restored after each mutation.
From round 1, already in your pending review. The rule that the tree on screen beats localStorage is guarded by a single test (the precedence-swap mutation above fails only that one), so the one-line comment is still worth asking for.
Still open from round 1, and rechecked today. The gate says master expects v2.16.0, and there is still no release-2.16 branch. The regression came from PR 18195, which is on v2.16.0. So if this stays on 2.17, 2.16.0 ships with the collapse in Fleet, Cluster Management and Harvester unless it is backported. Moving it to v2.16.0 would turn the gate green. It is the only check holding the PR apart from the review itself.
Tab A opens Storage. Tab B then rebuilds its nav. On master, B opens Storage too, because the rebuild read localStorage. On the PR, B keeps what it had, and its next save writes Storage back to closed, so A's choice is lost on reload.
This is narrower than it sounds. Without a rebuild, master overwrites A's choice in exactly the same way (I ran that too), so tabs were already last-write-wins. It follows directly from the on-screen tree outranking storage, and a tab not changing under your hands is arguably better. I would not raise it.
The fix carries state from the tree on screen, so it can't help once that tree is gone. Going to Preferences and back remounts SideNav, and Cluster Management's Advanced comes back closed. A Fleet group whose schemas leave and return also comes back closed. Both match master, and the PR description says persistence outside Explorer is deliberately left out. Your issue only asks for rebuilds, so this is not a gap in the PR. If someone reports the Preferences round trip, it needs its own issue.
e2e-test (admin, @virtualizationMgmt, @fleet), which round 1 saw failing, passes on the same commit after the auto-retry. It failed on master runs that day too, so this PR isn't the cause.SideNav.vue, Group.vue or the storage composable on current master is older than the baseline, so the comparison still holds.manager (providers, advanced), fleet (resources), harvester (networks, backupAndSnapshot, rbac, advanced) and Explorer, only advanced repeats, and that move changes cluster, which clears the tree first (round 1 checked it live). auth, settings, uiplugins and harvesterManager have no collapsible groups.zz19764std and zz19764base on mf for the role cases, plus a cluster-member binding on created-dashboard-2, and deleted them afterwards. No user, global role binding or cluster binding is left.kube-node-lease, so I forced that case by removing schemas in the browser.getGroups(). The real header dropdown was used for Harvester today and for Fleet in round 1.