From 23edd2cf4fa68eae7bd9c7f01fedf7ce8965bf9f Mon Sep 17 00:00:00 2001
From: Houzhong Xu <houzhongxu@seekkey.tech>
Date: Thu, 9 Oct 2025 10:17:34 +0000
Subject: [PATCH] CRITICAL FIX: Repair cluster stability

- Fix semaphore Nomad config (was incorrectly set to influxdb1)
- Fix ash1d and ash2e bind_addr from 0.0.0.0 to proper Tailscale addresses
- Restore cluster to expected 3+ server nodes
- Emergency cluster repair
---
 .../ansible/playbooks/deploy-nomad-config.yml | 31 ++++++++++++++++---
 nomad-configs/servers/ash1d.hcl               |  2 +-
 nomad-configs/servers/ash2e.hcl               |  2 +-
 3 files changed, 29 insertions(+), 6 deletions(-)

diff --git a/deployment/ansible/playbooks/deploy-nomad-config.yml b/deployment/ansible/playbooks/deploy-nomad-config.yml
index 56bca99..ebfeab1 100644
--- a/deployment/ansible/playbooks/deploy-nomad-config.yml
+++ b/deployment/ansible/playbooks/deploy-nomad-config.yml
@@ -1,10 +1,14 @@
 ---
-- name: 部署Nomad服务器配置模板
-  hosts: nomad_servers
+- name: 部署Nomad配置到所有节点
+  hosts: nomad_cluster
   become: yes
   
   tasks:
-    - name: 部署Nomad配置文件
+    - name: 检查节点类型
+      set_fact:
+        node_type: "{{ 'server' if inventory_hostname in groups['nomad_servers'] else 'client' }}"
+    
+    - name: 部署Nomad服务器配置文件
       template:
         src: nomad-server.hcl.j2
         dest: /etc/nomad.d/nomad.hcl
@@ -12,6 +16,17 @@
         owner: root
         group: root
         mode: '0644'
+      when: node_type == 'server'
+    
+    - name: 部署Nomad客户端配置文件
+      get_url:
+        url: "https://gitea.tailnet-68f9.ts.net/ben/mgmt/raw/branch/main/nomad-configs/nodes/{{ inventory_hostname }}.hcl"
+        dest: /etc/nomad.d/nomad.hcl
+        backup: yes
+        owner: root
+        group: root
+        mode: '0644'
+      when: node_type == 'client'
     
     - name: 重启Nomad服务
       systemd:
@@ -24,6 +39,14 @@
         port: 4646
         host: "{{ ansible_host }}"
         timeout: 30
+      when: node_type == 'server'
+    
+    - name: 等待Nomad客户端服务启动
+      wait_for:
+        port: 4646
+        host: "{{ ansible_host }}"
+        timeout: 30
+      when: node_type == 'client'
     
     - name: 显示Nomad服务状态
       systemd:
@@ -32,7 +55,7 @@
     
     - name: 显示服务状态
       debug:
-        msg: "{{ inventory_hostname }} Nomad服务状态: {{ nomad_status.status.ActiveState }}"
+        msg: "{{ inventory_hostname }} ({{ node_type }}) Nomad服务状态: {{ nomad_status.status.ActiveState }}"
 
 
 
diff --git a/nomad-configs/servers/ash1d.hcl b/nomad-configs/servers/ash1d.hcl
index e3f3520..bfad2c9 100644
--- a/nomad-configs/servers/ash1d.hcl
+++ b/nomad-configs/servers/ash1d.hcl
@@ -4,7 +4,7 @@ plugin_dir = "/opt/nomad/plugins"
 log_level = "INFO"
 name = "ash1d"
 
-bind_addr = "0.0.0.0"
+bind_addr = "ash1d.tailnet-68f9.ts.net"
 
 addresses {
   http = "ash1d.tailnet-68f9.ts.net"
diff --git a/nomad-configs/servers/ash2e.hcl b/nomad-configs/servers/ash2e.hcl
index 324f06d..37352f5 100644
--- a/nomad-configs/servers/ash2e.hcl
+++ b/nomad-configs/servers/ash2e.hcl
@@ -4,7 +4,7 @@ plugin_dir = "/opt/nomad/plugins"
 log_level = "INFO"
 name = "ash2e"
 
-bind_addr = "0.0.0.0"
+bind_addr = "ash2e.tailnet-68f9.ts.net"
 
 addresses {
   http = "ash2e.tailnet-68f9.ts.net"