Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
25 commits
Select commit Hold shift + click to select a range
0fc90bf
docs(solutions): add 9 and update 0 troubleshooting articles
jing2uo Nov 10, 2026
faafd9d
docs(solutions): add 14 and update 0 troubleshooting articles
jing2uo Nov 10, 2026
2709935
docs(solutions): add 9 and update 2 troubleshooting articles
jing2uo Nov 10, 2026
3f41079
docs(solutions): review fixes for the 20261110 batch
jing2uo Nov 12, 2026
e067a9b
docs(solutions): add 24 and update 2 troubleshooting articles
jing2uo Nov 12, 2026
d9af042
docs(solutions): add 23 and update 2 troubleshooting articles
jing2uo Nov 12, 2026
c16f3fd
docs(solutions): add 21 and update 1 troubleshooting articles
jing2uo Nov 12, 2026
6a0ee19
docs(solutions): add 24 and update 1 troubleshooting articles
jing2uo Nov 12, 2026
4dd3406
docs(solutions): add 20 and update 0 troubleshooting articles
jing2uo Nov 12, 2026
fb550bc
docs(solutions): add 28 and update 1 troubleshooting articles
jing2uo Nov 12, 2026
9f8c8d2
docs(solutions): add 15 and update 1 troubleshooting articles
jing2uo Nov 12, 2026
5c6b1b0
docs(solutions): add 20 and update 0 troubleshooting articles
jing2uo Nov 12, 2026
f74b5a7
docs(solutions): add 17 and update 0 troubleshooting articles
jing2uo Nov 12, 2026
a495c2c
docs(solutions): add 14 and update 0 troubleshooting articles
jing2uo Nov 18, 2026
88dbd1f
docs(solutions): add 27 and update 0 troubleshooting articles
jing2uo Nov 18, 2026
f62daf8
docs(solutions): add 6 and update 2 troubleshooting articles
jing2uo Nov 18, 2026
f5e8bf7
docs(solutions): add 16 and update 1 troubleshooting articles
jing2uo Nov 18, 2026
fc20490
docs(solutions): add 11 and update 2 troubleshooting articles
jing2uo Nov 18, 2026
b6fd65c
docs(solutions): add 16 and update 0 troubleshooting articles
jing2uo Nov 18, 2026
499d4d1
docs(solutions): add 6 and update 2 troubleshooting articles
jing2uo Nov 18, 2026
7245c3a
docs(solutions): add 8 and update 0 troubleshooting articles
jing2uo Nov 18, 2026
4b15be3
docs(solutions): add 12 and update 0 troubleshooting articles
jing2uo Nov 18, 2026
4dc175f
docs(solutions): add 3 and update 1 troubleshooting articles
jing2uo Nov 18, 2026
9d81df3
docs(solutions): add 6 and update 1 troubleshooting articles
jing2uo Nov 18, 2026
beae376
docs(solutions): add 1 and update 0 troubleshooting articles
jing2uo Nov 18, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
Original file line number Diff line number Diff line change
@@ -0,0 +1,94 @@
---
kind:
- Troubleshooting
products:
- Alauda Container Platform
ProductsVersion:
- "3.18"
- "4.2"
---

# ACP console session survives cookie deletion or a browser close, or the login page shows the previous account

## Issue

The platform web console keeps, loses, or reuses a login in a way the user does not expect. One of these is seen:

- After logging in and deleting the site's cookies, a page refresh keeps the account logged in; the user is not logged out. This is often found during a security assessment.
- With "end the session when all product tabs are closed" enabled, closing the browser and opening the platform again goes straight in without a login.
- Users can log in only after clearing the browser cache and restarting the browser.
- The login page says the license code has expired, although the license is still valid.
- After a direct local Dex login URL is opened, the form accepts account B but the session shown is still account A.
- The Service Mesh page says that no authorization token was provided, while the console login itself works.

Related articles: "Forgotten local admin password on ACP: reset the password hash, set a chosen password, and password policy limits" covers local passwords. "kubectl returns the server has asked for the client to provide credentials after ac login, or ac login fails, on ACP" covers the CLI.

## Root Cause

### Cause 1: login session data is also kept in the browser's local storage

The console stores data in the browser's Local Storage (keys such as `meepo-project` and `theme-mode`) as well as in cookies, and part of the login state is kept in Local Storage. Deleting only the cookies leaves that state, and the next page load restores the login. Clearing both cookies and Local Storage for the platform site logs the user out.

### Cause 2: the session ends about 30 seconds after the last product tab closes

On 4.2, Access Control offers "end the session when all product tabs are closed". The browser does not clear the session at the instant it closes; the front end calls the logout API through its multi-tab mechanism, and that call is delayed. The session therefore stays valid for about 30 seconds after the last product tab closes. Reopening within that window does not ask for a login; after it, a login is required. This is the designed behavior.

### Cause 3: the login page sat idle until its session cookie expired

The login page was left open with no activity until its session cookie expired, so submitting the stale form fails. Reopening the platform access URL, or clearing the cache, loads a fresh form. This is expected.

### Cause 4: the license-expired message is stale page cache

A license-expired message that does not appear in an incognito window comes from cached page content, not from the license. Retyping the platform address does not clear it.

### Cause 5: a direct Dex login URL reuses the previous account's cached token

On 4.2, account A logs in and the browser is closed without logging out. A direct local Dex URL (path `/console-dex/login/local` with `client_id=alauda-auth` and `redirect_uri` pointing at `/console-portal`) is then opened and account B is entered. The browser still holds A's token, and that URL uses the cached token instead of B's credentials, so the session is A's. Whether it reproduces depends on the browser.

### Cause 6: the token held by a long-open page has expired

On 3.18 the Service Mesh page reports that no authorization token was provided after the browser session has been open long enough for its token to expire. The page has nothing valid to send. A fresh login gets a new token.

## Resolution

### Cause 1: login session data is also kept in the browser's local storage

1. Open the browser developer tools (F12) on the platform page, go to **Application**, and check both **Cookies** and **Local Storage** for the platform address.
2. Clear both for the platform site, or use **Clear site data**, then refresh. The user is logged out.
3. For a normal logout, use the console's logout action, which ends the session on the server instead of only removing browser data.

### Cause 2: the session ends about 30 seconds after the last product tab closes

No change is needed. Wait more than about 30 seconds after the last product tab closes before expecting a new login. This is not an idle-timeout setting.

### Cause 3: the login page sat idle until its session cookie expired

Open the platform access URL again, or clear the site cache, and log in. Do not treat this as a wrong password or a Dex fault when a fresh page accepts the same user.

### Cause 4: the license-expired message is stale page cache

Check the license validity date in the platform first. If it is still valid, open an incognito window, or clear the site cache, and log in. If a wrong-password error follows, it is a password problem, not the license.

### Cause 5: a direct Dex login URL reuses the previous account's cached token

Log out, or clear the site's cookies and Local Storage, before using `/console-dex/login/local` to switch accounts. No platform object needs to change.

### Cause 6: the token held by a long-open page has expired

Log out and log in again, or open the platform in an incognito window. The page then loads. No mesh configuration change is needed.

## Diagnostic Steps

1. On the platform page, open the developer tools (F12):
- **Application** > **Cookies** and **Local Storage**: note which entries remain for the platform address.
- **Network**: refresh, select a platform API request, and check whether it still carries `Cookie` or `Authorization`.
2. Open the same address in an incognito window and compare the behavior.

| What you see | Cause |
| --- | --- |
| Cookies deleted, refresh keeps the login, Local Storage still has entries; clearing it logs out | 1 |
| Session survives a browser close for less than about 30 seconds with tab-close logout enabled | 2 |
| Login works only after a cache clear, the form was left idle | 3 |
| License-expired text, license still valid, incognito window is clean | 4 |
| Direct Dex URL logs in as the previous user | 5 |
| Service Mesh page says no authorization token, a fresh login fixes it | 6 |
Original file line number Diff line number Diff line change
@@ -0,0 +1,127 @@
---
kind:
- How To
products:
- Alauda Container Platform
ProductsVersion:
- "3.12"
- "3.18"
- "4.0"
- "4.1"
- "4.2"
- "4.3"
---

# ACP control-plane limits: apiserver address change, snapshot.db scope, total master loss, x86 to ARM, and planned shutdown

## Issue

- A business-cluster apiserver VIP or URL has changed and the ACP console still uses the old address.
- A namespace was deleted and the only copy left is an etcd `snapshot.db` or `snapshot-etcd` file.
- All masters are lost and the only copy is the platform etcd backup; a version-specific command list for a full restore is requested.
- The `global` cluster runs on x86 and has to move to ARM.
- The whole cluster has a scheduled shutdown.

These are procedure and product-limit questions, not a single-member etcd repair and not a database-full alarm.

Related articles: "Replacing or rejoining an ACP master node: same-IP replacement, etcd-join, extra masters, and control-plane requests".

## Root Cause

### Cause 1: A business cluster apiserver address change has no standard platform procedure

On platform versions 3.12, 3.18, and 4.2 there is no standard procedure to point the console or the `global` cluster at a new business-cluster apiserver VIP or address.

### Cause 2: A deleted namespace cannot be restored on its own from snapshot.db

A single namespace cannot be extracted from `snapshot.db` (running `strings` on the file does not produce the objects). The supported path is a restore of the whole cluster from an etcd backup, which returns every namespace to that backup's point in time. There is no reference procedure for starting a separate etcd just to read one namespace.

### Cause 3: All masters lost with only an etcd backup has no supported restore

There is no supported procedure to rebuild a cluster from the etcd backup alone after all masters are lost, on any version including 4.3, with or without bare-metal masters. An etcd snapshot stores cluster data but not the certificates and keys the cluster needs. The backup job can also save node files (`/etc/kubernetes/pki`, manifests, kubelet configuration, container runtime units), but those are version-specific; a cluster rebuilt from them is not a standard deployment and is likely to fail a later upgrade. A `snapshot-etcd` file is the platform backup, not a namespace extract and not a rebuild kit.

### Cause 4: An x86 global cluster cannot be moved to ARM by replacing servers or importing etcd

Replacing the servers does not change the cluster's CPU architecture, and redeploying an ARM `global` and importing the existing etcd data does not let it take over the existing business clusters. An ARM `global` is a new deployment plus a business migration.

### Cause 5: A scheduled shutdown of the whole cluster

A planned outage needs an etcd backup first and a normal `shutdown` rather than a power cut. Power the nodes off together when possible; otherwise shut down workers first and masters after. After power-on, platform components recover on their own.

## Resolution

### Cause 1: A business cluster apiserver address change has no standard platform procedure

Do not create a procedure by rewriting kube-apiserver manifests or restoring etcd. Collect the old and new URLs and contact Alauda support for a version-specific procedure. The following prints the server URL a kubeconfig uses; it does not update the platform:

```bash
kubectl config view --minify -o jsonpath='{.clusters[0].cluster.server}{"\n"}'
```

### Cause 2: A deleted namespace cannot be restored on its own from snapshot.db

Restoring the namespace means restoring the whole cluster from the etcd backup, which is a planned-downtime decision for every namespace, not a partial extract. `file` only identifies the blob:

```bash
file snapshot.db
```

### Cause 3: All masters lost with only an etcd backup has no supported restore

Plan for this before it happens:

1. Take full server or VM backups on a schedule, in addition to the etcd backup.
2. Protect master disks (for example with RAID) and repair failed disks promptly.
3. Run critical workloads on more than one independent cluster.

Check how many etcd members are still running:

```bash
kubectl -n kube-system get pods -l component=etcd -o wide
```

Zero running members with only `snapshot-etcd` or `snapshot.db` left is this limit. With one or more healthy members, the remaining members are repaired or replaced one at a time instead.

### Cause 4: An x86 global cluster cannot be moved to ARM by replacing servers or importing etcd

Plan a new ARM deployment and a migration of the business workloads. `uname -m` only confirms the current architecture:

```bash
uname -m
kubectl get nodes -o wide
```

### Cause 5: A scheduled shutdown of the whole cluster

Take the etcd backup before any power-off: run the backup on the platform's etcd backup task and confirm that a current backup record exists, or take a snapshot with `etcdctl snapshot save` from a healthy member.

```bash
kubectl get etcdbackupconfiguration -A
```

> **Warning:** `shutdown -h now` powers this node off. The etcd backup must already exist. Run it on workers before masters when the nodes cannot go down together. Rollback is powering the node on again. A node that does not return after power-on is a new fault, not a step of this shutdown.

```bash
sudo shutdown -h now
```

After power-on, wait for the platform components to recover and check:

```bash
kubectl get nodes
kubectl get pods -A | grep -v -E 'Running|Completed'
```

## Diagnostic Steps

```bash
kubectl -n kube-system get pods -l component=etcd -o wide
kubectl get nodes -l node-role.kubernetes.io/control-plane= -o wide
kubectl get nodes -o wide
```

- Address change of an existing business cluster on 3.12, 3.18, or 4.2: Cause 1.
- `snapshot.db` in hand and one missing namespace: Cause 2.
- Every master lost and the only copy is an etcd backup: Cause 3.
- x86 `global` to ARM by server swap or etcd import: Cause 4.
- Planned power-off of the whole cluster: Cause 5.
Loading
Loading