Fix homea mirror priority and record upgrade verification

ansible/README.md

@@ -169,4 +169,12 @@ master 逐台升级，worker 保持并行（默认 20 forks）。已有 server
 
 在安装 Ansible 的 Python 环境运行就绪检查回归：`python -m unittest discover -s tests -v`。
 
+### 2026-10-16 执行记录
 
+16 个节点已升级至 `v1.35.9+k3s1`，3 个 server 逐台执行，13 个 worker 并行执行。升级前的不压缩快照以 `pre-upgrade-v1.35.9+k3s1-<node>-<timestamp>` 保存在 snapshots 目录，恢复用 token 为 server 目录下的 `pre-upgrade-v1.35.9+k3s1.token`。
+
+homea 的 Velero 镜像拉取曾停滞：确认两个现有镜像源能下载、磁盘充足且没有进程代理后，通过 `ansible homea -m ansible.builtin.systemd -a 'name=k3s-agent state=restarted' -e ansible_port=2103` 释放旧下载请求，DaemonSet 随后完成更新。Velero 的 Helm 等待因此超时，通过相同时间戳的 `reconcile.fluxcd.io/requestedAt` 和 `reconcile.fluxcd.io/resetAt` 注解重新协调成功。
+
+Grafana 的大镜像层再次在该节点的 1ms 链路停滞，分段下载验证 Daocloud 可用后，通过 homea 的 `registry_mirror_overrides` 调整 Docker 镜像优先级；其他 registry 继承全局配置。镜像配置单独变化时，role 会重启对应 K3s 服务并等待 Ready；版本升级时由安装流程加载新配置。调整后 Grafana 镜像约 27 秒拉取完成；Prometheus 栈因旧下载触发的部署超时也通过上述注解重新协调成功。
+
+CloudNativePG/Barman 升级重建了两套单实例 PostgreSQL Pod，恢复后均为 2/2 Ready；PostgreSQL 数据镜像保持 17.4。各执行一次 `SELECT pg_switch_wal();` 验证新插件，两个实例的归档计数均增长，归档失败计数未增加。未进行数据恢复演练。

ansible/inventory/hosts.yml

@@ -108,6 +108,11 @@ all:
           node_hostname: homea
           node_region: cn-sh
           use_mirror: true
+          # 此线路访问 1ms 的 Docker 镜像层会停滞，优先使用现有备用源。
+          registry_mirror_overrides:
+            docker.io:
+              - docker.m.daocloud.io
+              - docker.1ms.run
 
     # 节点分组
     k3s_cluster:

ansible/roles/k3s/tasks/main.yml

@@ -51,6 +51,7 @@
     dest: /etc/rancher/k3s/registries.yaml
     mode: "0644"
   when: use_mirror | default(false)
+  register: k3s_registries_config
 
 # 判断是否需要安装/重启
 # 使用 -e force_reinstall=true 可强制重新安装
@@ -136,6 +137,15 @@
     path: /tmp/k3s-install.sh
     state: absent
 
+- name: Restart K3s after updating registry mirrors
+  ansible.builtin.systemd:
+    name: "{{ 'k3s' if inventory_hostname in groups['masters'] else 'k3s-agent' }}"
+    state: restarted
+  when:
+    - k3s_registries_config.changed | default(false)
+    - k3s_binary.stat.exists
+    - k3s_version in k3s_current_version.stdout
+
 # /readyz includes the local API server's etcd readiness check.
 - name: Wait for the local API server and etcd to be ready
   ansible.builtin.command:

ansible/roles/k3s/templates/registries.yaml.j2

@@ -1,7 +1,7 @@
 # 镜像加速配置
 ---
 mirrors:
-{% for registry, endpoints in registry_mirrors.items() %}
+{% for registry, endpoints in (registry_mirrors | combine(registry_mirror_overrides | default({}))).items() %}
   "{{ registry }}":
     endpoint:
 {% for endpoint in endpoints %}

ansible/tests/test_readiness.py

@@ -3,6 +3,7 @@ import unittest
 from pathlib import Path
 
 import yaml
+from ansible.plugins.filter.core import combine
 from jinja2 import Environment, StrictUndefined
 
 
@@ -14,6 +15,9 @@ class ReadinessTest(unittest.TestCase):
         )
         env = Environment(undefined=StrictUndefined)
         env.filters["from_json"] = json.loads
+        env.filters["combine"] = combine
+        cls.env = env
+        cls.tasks = tasks
         cls.conditions = {
             task["register"]: env.compile_expression(task["until"])
             for task in tasks
@@ -47,6 +51,37 @@ class ReadinessTest(unittest.TestCase):
             with self.subTest(rc=rc, stdout=stdout):
                 self.assertEqual(condition(k3s_api_ready={"rc": rc, "stdout": stdout}), expected)
 
+    def test_mirror_changes_restart_existing_nodes_without_a_version_upgrade(self):
+        task = next(t for t in self.tasks if t["name"] == "Restart K3s after updating registry mirrors")
+        condition = self.env.compile_expression(" and ".join(f"({c})" for c in task["when"]))
+        for changed, exists, current, expected in [
+            (True, True, "v1.35.9+k3s1", True),
+            (False, True, "v1.35.9+k3s1", False),
+            (True, True, "v1.35.6+k3s1", False),
+            (True, False, "", False),
+        ]:
+            with self.subTest(changed=changed, exists=exists, current=current):
+                self.assertEqual(condition(
+                    k3s_registries_config={"changed": changed},
+                    k3s_binary={"stat": {"exists": exists}},
+                    k3s_current_version={"stdout": current} if exists else {},
+                    k3s_version="v1.35.9+k3s1",
+                ), expected)
+
+    def test_host_mirror_override_preserves_other_registries(self):
+        template = self.env.from_string(
+            (Path(__file__).parents[1] / "roles/k3s/templates/registries.yaml.j2").read_text()
+        )
+        mirrors = {"docker.io": ["docker.1ms.run", "docker.m.daocloud.io"], "ghcr.io": ["ghcr.m.daocloud.io"]}
+        for override in [{}, {"docker.io": list(reversed(mirrors["docker.io"]))}]:
+            result = yaml.safe_load(template.render(
+                registry_mirrors=mirrors, registry_mirror_overrides=override,
+            ))["mirrors"]
+            self.assertEqual(result["ghcr.io"]["endpoint"], ["https://ghcr.m.daocloud.io"])
+            self.assertEqual(result["docker.io"]["endpoint"], [
+                "https://" + host for host in override.get("docker.io", mirrors["docker.io"])
+            ])
+
 
 if __name__ == "__main__":
     unittest.main()