47 Commits
Author SHA1 Message Date
薛佳帆 0afaa17112 release: bump chart version to v1.0.98 2026-11-11 16:40:05 +08:00
薛佳帆 bf0170aa26 fix(dotnet): 全部 .NET 主容器加 appArmorProfile Unconfined——AppArmor 拒线程定向信号致 GC abort
根因(dotnet/runtime#136603,已由同事在 sciveda 现场实测验证):Ubuntu 26.04 新内核
配套 AppArmor 行为禁止容器内进程向自身线程发信号(pthread_kill/tgkill → EACCES),
CoreCLR GC 挂起线程失败 → 运行时 abort → dmesg 表现为 'general protection fault
in libc.so.6[+0x289a2]'(.NET BGC/Server GC/TP Worker 线程),容器 exit 139
CrashLoop,负载越高崩得越频,且 journalctl 无 apparmor=DENIED 记录。
同机制亦解释此前 postgres 'could not signal for checkpoint: Permission denied'
(privileged 之所以有效 = privileged 容器不受 AppArmor 约束)。
修复:agent/sso/ultra/edge/gateway/platform 六处 .NET 主容器
securityContext.appArmorProfile.type: Unconfined。
2026-11-11 16:38:24 +08:00
薛佳帆 2df3373304 release: bump chart version to v1.0.95 2026-11-10 16:23:26 +08:00
薛佳帆 9b002c6491 fix(postgres): 全部 postgres StatefulSet 三件套修复——绕过镜像内 gosu 直跑 + 特权容器
根因:官方 postgres:17 镜像(2026-08 构建)内 gosu 在新内核(Ubuntu 26.04 kernel 7.0.0-39+)上
用户态死循环,容器假活(Running 但 5432 无服务、PGDATA 空、零日志);绕过 gosu 后 PG17 恢复期
signal checkpoint 再遇权限死循环,故三件缺一不可:
- pod securityContext: runAsUser/runAsGroup/fsGroup=999(以 postgres 用户直跑,跳过 gosu)
- container command: docker-entrypoint.sh postgres(非 root 下 entrypoint 自动跳过 gosu)
- container securityContext: privileged(否则恢复期 'could not signal for checkpoint: Permission denied')

涉及 5 个模板:sso/ultra/audit(audit-db)/platform(audit)/agent-unit 各自内嵌的 postgres。
同内容已提交 develop 分支 407aef8;实测:xjf1641(Xeon+kernel43)与 wytest(i5+kernel42)
原始定义必死、打补丁后稳定;kernel 7.0.0-38 老机暂未触发但不免疫,升级内核前必须先吃本修复。
2026-11-10 16:21:57 +08:00
薛佳帆 e031d656a0 release: bump chart version to v1.0.88 2026-10-30 03:45:53 +00:00
薛佳帆 347180d77b release: bump chart version to v1.0.87 2026-10-30 03:42:15 +00:00
薛佳帆 321b602905 release: bump chart version to v1.0.84 2026-10-27 06:24:59 +00:00
薛佳帆 8733595824 release: bump chart version to v1.0.83 2026-10-26 05:29:55 +00:00
薛佳帆 ff5ecf710e release: bump chart version to v1.0.82 2026-10-23 01:18:36 +00:00
薛佳帆 c2b78f8ff0 release: bump chart version to v1.0.81 2026-10-23 01:17:47 +00:00
薛佳帆 b0e1dcd4f0 release: bump chart version to v1.0.80 2026-10-23 00:57:04 +00:00
薛佳帆 e7f4b71cd0 release: bump chart version to v1.0.79 2026-10-22 05:10:33 +00:00
薛佳帆 4bcf4fe0a3 release: bump chart version to v1.0.78 2026-10-22 02:44:25 +00:00
薛佳帆 33de837556 release: bump chart version to v1.0.77 2026-10-22 01:36:57 +00:00
薛佳帆 b1d65fab30 release: bump chart version to v1.0.76 2026-10-22 01:35:03 +00:00
薛佳帆 133c032144 release: bump chart version to v1.0.68 2026-10-13 05:46:43 +00:00
薛佳帆 7bd4d58a6d release: bump chart version to v1.0.50 2026-09-29 11:19:27 +00:00
薛佳帆 60c8b4d90e release: bump chart version to v1.0.48 2026-09-29 05:37:56 +00:00
薛佳帆 22fd52f20f release: bump chart version to v1.0.47 2026-09-29 03:08:53 +00:00
薛佳帆 7d1bc2d90c release: bump chart version to v1.0.46 2026-09-29 02:33:26 +00:00
薛佳帆 ada89158fa release: bump chart version to v1.0.45 2026-09-29 02:32:30 +00:00
薛佳帆 c3f27ee62b Revert "feat(agent-unit): 特征值预警 Redis 近端窗口开关 env(FeatureValueWarningCache__*,AgentUnit 默认取单元名)"
This reverts commit b8659270e6.
2026-09-29 09:24:43 +08:00
薛佳帆 b8659270e6 feat(agent-unit): 特征值预警 Redis 近端窗口开关 env(FeatureValueWarningCache__*,AgentUnit 默认取单元名) 2026-09-29 09:06:59 +08:00
薛佳帆 c698fe14cc release: bump chart version to v1.0.38 2026-09-23 07:15:32 +00:00
薛佳帆 ee378168c0 release: bump chart version to v1.0.37 2026-09-23 01:56:11 +00:00
薛佳帆 41ccb587a9 release: bump chart version to v1.0.36 2026-09-23 01:55:39 +00:00
薛佳帆 bfec34fc70 release: bump chart version to v1.0.33 2026-09-21 11:32:03 +00:00
薛佳帆 b42271b704 release: bump chart version to v1.0.32 2026-09-21 11:30:49 +00:00
薛佳帆 51123401a1 release: bump chart version to v1.0.28 2026-09-20 08:28:10 +00:00
薛佳帆 e5bf7090b4 release: bump chart version to v1.0.24 2026-09-17 09:00:20 +00:00
薛佳帆 e988974374 release: bump chart version to v1.0.22 2026-09-15 09:54:26 +00:00
薛佳帆 4613388d0c chore: 删除 provision/rollback-host 迁移脚本——主机已全面由 NM 托管且 networkd 已 mask,迁移工具不再需要;rollback 依赖 networkd 接管原配置,mask 状态下执行反而会导致接口失联 2026-09-15 10:09:17 +08:00
薛佳帆 b6c35f28e3 feat(network-agent): headless Service 打 mixingx.io/role=node-id 标签 —— 纳入 Gateway K8s 服务发现,生成 /api/network/** 透传路由;注释说明前端弹窗改经 Gateway 同源转发 2026-09-14 14:37:08 +08:00
薛佳帆 05a476d932 feat: provision 写入 NM keep-configuration=no——禁止开机收编 initramfs(dracut 遗留 DHCP)抢配的连接,强制按 autoconnect-priority 选路激活静态档案;initrd 的 DHCP 租约以副地址共存(双 IP:静态主+DHCP 副兜底);rollback 对称移除 2026-09-10 16:21:42 +08:00
薛佳帆 9035f9eb96 release: bump chart version to v1.0.17 2026-09-10 08:02:26 +00:00
薛佳帆 869754606e release: bump chart version to v1.0.16 2026-09-10 08:01:59 +00:00
薛佳帆 405eef034a feat(network-agent): 挂载宿主机状态目录 /var/lib/mixingx/network-agent(DirectoryOrCreate)——网卡别名 aliases.json 持久化,节点重启/Pod 重建均保留;chart 1.0.16 2026-09-10 09:47:34 +08:00
薛佳帆 4395e900e9 refactor(network-agent): 移除 NET_ADMIN capability 与 nm-run 挂载——agent 已无 ip addr del 等网络写操作(切静态清残留动态地址逻辑删除,networkd 将在宿主机 mask),特权动作全部经 D-Bus 由主机 NM 执行;/run/NetworkManager 源码未使用一并移除;chart 1.0.7 2026-09-08 16:20:53 +08:00
薛佳帆 352aa78a8f Merge branch 'develop/v.1.0.1' 2026-09-07 15:48:19 +08:00
薛佳帆 0c105e5224 fix(network-agent): 容器加 NET_ADMIN 能力——agent 清除切换静态后残留的动态地址需要 ip addr del,默认能力集无此权限被静默拒绝 2026-09-07 15:48:16 +08:00
薛佳帆 e5f1ba44c8 Merge branch 'develop/v.1.0.1' 2026-09-07 09:40:13 +08:00
薛佳帆 f0f67afbcb fix(scripts): provision-host 修复——DNS 仅收集 IPv4(原版塞入 IPv6 地址被 nmcli 拒绝导致 set-e 中止);保留原 netplan 配置仅切全局 renderer(原版删配置导致 WiFi 机器断网失联,netplan 自动渲染等值 NM profile 含 SSID/PSK);WiFi 主口跳过 rescue-dhcp 2026-09-07 09:40:10 +08:00
薛佳帆 90d8838e15 Merge branch 'develop/v.1.0.1' 2026-09-04 17:24:13 +08:00
薛佳帆 bcc5e0d4d0 fix(network-agent): 容器加 apparmor=unconfined 注解——Ubuntu dbus-daemon 的 AppArmor 仲裁拦截容器内 nmcli 的 D-Bus 调用(AccessDenied) 2026-09-04 17:24:10 +08:00
薛佳帆 2e7d079f65 Merge branch 'develop/v.1.0.1' 2026-09-04 15:39:35 +08:00
薛佳帆 da7762b0cf feat(network-agent): 新增 mixingx-network-agent chart——DaemonSet(hostNetwork:9750,root 仅挂载 dbus socket,非 privileged)+ Headless 诊断 Service 2026-09-04 14:56:09 +08:00
薛佳帆 61edf87116 feat(scripts): network-agent 主机初始化脚本——装 NM 并等值静态连接切换 netplan renderer(不断网),rescue-dhcp 低优先级兜底,rfkill 解锁无线与 polkit 放行,附 rollback 反向脚本 2026-09-04 14:56:09 +08:00