OpenUptime

Server monitoring

A small open-source agent runs on each Linux server and reports its CPU, memory, swap, load, disks, disk I/O and network interfaces. The servers show up under Servers in the dashboard, and one that stops reporting raises an alert.

Requirements

  • Linux on amd64 (x86_64) or arm64 (aarch64), with systemd. Other systems are not supported.
  • Root access, through sudo or a root shell, to install and to uninstall.
  • curl or wget.
  • Outbound HTTPS to raw.githubusercontent.com and github.com while installing, and to your OpenUptime address while running. The agent needs no inbound port.

Install

  1. Open Servers and click Add server. Only workspace owners and admins see the key.
  2. Optionally give the server a name and tags. Leave the name empty to use the server's host name.
  3. Click Copy command. The command carries your project's agent key and this site's address.
  4. Run it on the server. It finishes in a few seconds.
  5. The server appears in the list within about a minute, named after its host name unless you gave it a name. You can rename it on its page.

An AI agent connected through MCP can fetch the same command with the get_server_install tool, then read servers with list_servers and get_server.

The command looks like this:

curl -fsSL https://raw.githubusercontent.com/becoolme/openuptime/main/apps/agent/install.sh | sudo sh -s -- --key oua_… --endpoint https://your-openuptime.example

The script comes straight from the public repository, not from your OpenUptime site, so what runs as root is the published source. To read it before running it, download it first:

curl -fsSL https://raw.githubusercontent.com/becoolme/openuptime/main/apps/agent/install.sh -o install.sh
less install.sh
sudo sh install.sh --key oua_… --endpoint https://your-openuptime.example

What the script does

  1. Checks that the machine runs Linux with systemd, that it runs as root, that the key has the right form, and that the address starts with https://.
  2. Downloads the agent for the machine's CPU from the latest GitHub release, together with the release's SHA256SUMS. If the checksum does not match, or the file does not run, it stops and installs nothing.
  3. Writes the key and the address to /etc/openuptime-agent.env, readable by root only.
  4. Installs the openuptime-agent systemd service, enables it at boot and starts it.

It creates these files and nothing else:

PathWhat it is
/usr/local/bin/openuptime-agentThe agent, a single static binary
/etc/openuptime-agent.envThe key and the address, mode 0600
/etc/systemd/system/openuptime-agent.serviceThe systemd unit

Check that it runs

systemctl status openuptime-agent
journalctl -u openuptime-agent -f
openuptime-agent --version

A running agent logs one line when it starts, and after that only errors.

More servers

One key works for every server in a project: run the same command on each one. A server is recognised by /etc/machine-id, so reinstalling on the same machine keeps its history.

Machines cloned from one image can share a machine id and would show up as a single server. The id is listed under System on each server's page. Give each clone its own id:

sudo rm /etc/machine-id
sudo systemd-machine-id-setup
sudo systemctl restart openuptime-agent

A project accepts up to 30 new servers an hour. Each server is backed by a heartbeat monitor that carries its offline alerts; it does not count toward a plan's monitor limit. Where a plan limits servers, the limit covers every project in the workspace.

Options

The install script accepts:

OptionMeaning
--key oua_…The project's agent key. Required to install.
--endpoint URLWhere the agent reports. The dashboard fills in its own address. It must use https://; plain http:// is accepted only for localhost, 127.0.0.1 and [::1], unless you add --allow-http.
--name NAMEThe server's name in OpenUptime. Without it, the server takes its host name. Quotes, $, backticks and backslashes are refused.
--tags A,BComma-separated tags for the server: up to 20, each up to 32 characters.
--allow-httpAccept a plain http:// endpoint on another machine, for a self-hosted install on a network without TLS. The key and the metrics then cross the network unencrypted, and the script prints a warning. The dashboard adds this flag itself when you open it over plain HTTP.
--binary-url URLDownload the agent from this https:// address instead of GitHub. The environment variable OPENUPTIME_AGENT_URL does the same.
--sha256 HEXThe SHA-256 the agent must have. Without it, the script reads SHA256SUMS from the same folder as the agent.
--uninstallStop the service and remove the agent. See Uninstall.

--name and --tags apply only when the server first reports; after that, OpenUptime keeps whatever is set on the server's page, and running the script again with other values does not change it. They need agent 0.1.6 or later; older agents register under the host name.

The agent itself accepts these flags, or the environment variables in the unit's env file:

FlagEnvironmentDefault
--keyOPENUPTIME_KEYnone, required
--endpointOPENUPTIME_ENDPOINThttps://openuptime.app
--nameOPENUPTIME_NAMEthe host name; used when the server registers
--tagsOPENUPTIME_TAGSnone; comma-separated, used when the server registers
--ignore-interfacesOPENUPTIME_IGNORE_INTERFACESlo,docker,veth,br-,virbr,cni,flannel,cali,vxlan: interface name prefixes to leave out
--oncePrint one report as JSON and exit without sending it
--versionPrint the version and exit

To change a setting after installing, edit /etc/openuptime-agent.env and restart the service:

sudoedit /etc/openuptime-agent.env      # OPENUPTIME_KEY=oua_…
sudo systemctl restart openuptime-agent

To see exactly what the agent would send, run openuptime-agent --once.

Without access to GitHub

Build the agent from the repository (GOOS=linux GOARCH=amd64 CGO_ENABLED=0 go build in apps/agent) or copy it from a release, host it over HTTPS where the server can reach it, and pass its address with --binary-url. Copy the install script to the server along with it.

The script checks the download before running it. Either put a SHA256SUMS file next to the agent, with a line for the agent's file name as sha256sum writes it, or pass the checksum yourself:

sudo sh install.sh --key oua_… --endpoint https://your-openuptime.example \
  --binary-url https://mirror.example/openuptime-agent-linux-amd64 \
  --sha256 <64 hex characters from sha256sum>

What is reported

  • CPU: use across all cores, the model and the core count.
  • Memory and swap: used, cache and total.
  • Load: the 1, 5 and 15 minute load averages.
  • Disks: every mounted local disk and network share (ext4, XFS, Btrfs, ZFS, NFS, CIFS and the like), with used and total space, and the read and write rate of the physical disks. Virtual and container file systems are left out. A device mounted in several places is listed at each mount and counted once in the totals.
  • Network: each interface with its MAC address, all of its IPv4 and IPv6 addresses, and its receive and send rate. Loopback, Docker, bridge and overlay interfaces are left out by default.
  • System: host name, distribution, kernel, architecture, boot time and uptime.

A report lists at most 64 disks and 64 network interfaces, with up to 32 IPv4 and 32 IPv6 addresses on each interface. On a host with more, such as one with hundreds of ZFS datasets or VM tap interfaces, the first ones are listed. The network rates still count every interface; the disk totals count the listed disks.

Click a server for its current state in a side panel, or open its page for charts over the last hour, 6 hours, 24 hours or 7 days. Metrics are kept for 7 days.

The agent reads system statistics from /proc, plus /etc/machine-id and /etc/os-release. It does not read logs, processes, users' files or anything else stored on the server.

Report interval

Servers report every 60 seconds by default. Change it per server under Settings on the server's page: 30 seconds, 1, 2 or 5 minutes. The agent picks up a new interval at its next report. A report that arrives much sooner than the interval is acknowledged but not stored.

On the hosted service, reporting every 30 seconds needs a paid plan, and each plan allows a number of servers per workspace, counted across its projects. Servers do not count toward the monitor limit. A self-hosted install allows every interval and any number of servers.

Offline alerts

A server counts as offline after about three intervals without a report: three minutes at the default interval. Each server has a heartbeat monitor tagged server that opens an incident at that point and resolves it when reports resume, so alerts go to the project's alert channels like any other monitor's.

Set them under Offline alerts on the server's page or in its side panel on the list. Turn them off for a server you do not need to hear about, pick the channels they go to (none picked means all of them), or mute them for an hour before a restart or a move; a maintenance window that covers the server mutes them too.

The list marks an online server as High usage when its CPU, memory or fullest disk reaches 90%. That mark is shown only; it sends no alert. The eye button next to the filters masks every address and hides every tag on the server pages for screen sharing, and is remembered in that browser.

Security

  • Report-only. The agent only sends data out. It opens no port and cannot run commands, scripts or updates sent from the dashboard. The only thing it reads back is the report interval. A test in the repository fails if the agent's code starts running programs or listening on a port.
  • Unprivileged. The service runs as a temporary system user (systemd DynamicUser) with NoNewPrivileges, a read-only view of the system (ProtectSystem=strict) and a read-only /home. Root is only needed to install and uninstall.
  • Checked downloads. The script compares the agent with the release's SHA256SUMS before running it, and accepts only HTTPS for the download and the report address.
  • The key is a secret. Anyone with it can register servers in the project and send reports as any of its servers, including false ones that hide an outage. It cannot read data or change settings. Only workspace owners and admins can see or rotate it. On the server it is stored in a file only root can read, so it does not appear in the process list.
  • Rotate the key under Servers → Add server → Rotate key if it leaks. Every server using the old key stops reporting, and its agent tries again every 5 minutes. Put the new key in /etc/openuptime-agent.env and restart the service, or run the install command again.

Update the agent

Run the install command again. It downloads the latest release, replaces the binary, rewrites the env file with the key and address you pass, and restarts the service. The server keeps its history. Check the version with openuptime-agent --version.

Uninstall

Run this on the server:

curl -fsSL https://raw.githubusercontent.com/becoolme/openuptime/main/apps/agent/install.sh | sudo sh -s -- --uninstall

It stops and disables the service, deletes the three files listed under What the script does, and reloads systemd. Nothing else is left on the server: the service's temporary user goes away with it. No key is needed.

Then delete the server in the dashboard: open its page and click Delete server. This removes its history and its heartbeat monitor. Do it in this order: a server deleted while its agent still runs comes back at the next report. The command is also on each server's page and in the Add server dialog.

To uninstall by hand, without downloading the script:

sudo systemctl disable --now openuptime-agent
sudo rm -f /etc/systemd/system/openuptime-agent.service /usr/local/bin/openuptime-agent /etc/openuptime-agent.env
sudo systemctl daemon-reload

Troubleshooting

What you seeWhat to do
only Linux is supported, systemd is required, unsupported CPU architectureThe machine is not one the agent supports. See Requirements.
run as rootPipe into sudo sh -s -- …, not into sh.
download failedThe server cannot reach GitHub, or the release has no agent yet. Check outbound HTTPS, or use --binary-url.
checksum mismatchThe file downloaded is not the one the release lists. Nothing was installed. Run the command again; if it keeps failing, do not install from that source.
could not download SHA256SUMS, SHA256SUMS has no lineThe mirror has no checksum file for the agent. Add one, or pass --sha256. See Without access to GitHub.
--endpoint must start with https://Use the HTTPS address of your OpenUptime site. Put a TLS proxy in front of a self-hosted install that serves plain HTTP, or add --allow-http if the network between them is trusted.
The service runs, but the server never appearsRead journalctl -u openuptime-agent. Check that the server can reach the address in OPENUPTIME_ENDPOINT: it must be the address you open the dashboard at.
the agent key was rejectedThe key was rotated, or the project was deleted. Put the current key in the env file and restart.
HTTP 402, server_limitThe workspace has as many servers as its plan allows. Delete a server you no longer need, or upgrade. After a downgrade, the servers added first keep reporting.
HTTP 429More than 30 new servers in an hour. The agent keeps trying and gets in later.
Two machines show as one serverThey share a machine id. See More servers.
A disk or interface is missingRun openuptime-agent --once to see what it reports. Interfaces whose names start with an ignored prefix are left out; change OPENUPTIME_IGNORE_INTERFACES.

服务器监控

在每台 Linux 服务器上运行一个开源的小型 agent,它会上报 CPU、内存、交换分区、负载、磁盘、磁盘读写和网卡数据。服务器会出现在后台的服务器页面,停止上报时会发出告警。

环境要求

  • 运行在 amd64(x86_64)或 arm64(aarch64)上、使用 systemd 的 Linux。不支持其他系统。
  • 安装和卸载需要 root 权限,可以用 sudo,也可以直接在 root shell 中运行。
  • 需要 curl 或 wget。
  • 安装时需要能通过 HTTPS 访问 raw.githubusercontent.com 和 github.com,运行时需要能访问你的 OpenUptime 地址。agent 不需要开放任何入站端口。

安装

  1. 打开服务器页面,点击添加服务器。只有工作区所有者和管理员能看到密钥。
  2. 可以为服务器填写名称和标签。名称留空时使用服务器的主机名。
  3. 点击复制命令。命令里已经带上了项目的 agent 密钥和当前站点地址。
  4. 在服务器上运行,几秒钟即可完成。
  5. 大约一分钟内,服务器会出现在列表中,没有填写名称时显示它的主机名,之后可以在详情页修改。

通过 MCP 连接的 AI 助手可以用 get_server_install 工具拿到同样的安装命令,再用 list_servers 和 get_server 查看服务器。

命令大致如下:

curl -fsSL https://raw.githubusercontent.com/becoolme/openuptime/main/apps/agent/install.sh | sudo sh -s -- --key oua_… --endpoint https://your-openuptime.example

安装脚本直接来自公开仓库,而不是你的 OpenUptime 站点,所以以 root 身份运行的就是公开的源码。想先看一遍再运行,可以先下载:

curl -fsSL https://raw.githubusercontent.com/becoolme/openuptime/main/apps/agent/install.sh -o install.sh
less install.sh
sudo sh install.sh --key oua_… --endpoint https://your-openuptime.example

脚本做了什么

  1. 检查系统是否为带 systemd 的 Linux、是否以 root 运行、密钥格式是否正确,以及上报地址是否以 https:// 开头。
  2. 从 GitHub 最新的 release 下载与本机 CPU 架构对应的 agent,同时下载该 release 的 SHA256SUMS。校验和对不上,或者文件无法运行,就中止安装,什么都不装。
  3. 把密钥和地址写入 /etc/openuptime-agent.env,只有 root 可读。
  4. 安装 openuptime-agent systemd 服务,设为开机启动并立即启动。

脚本只会创建下面这些文件:

路径说明
/usr/local/bin/openuptime-agentagent 程序,单个静态二进制文件
/etc/openuptime-agent.env密钥和地址,权限 0600
/etc/systemd/system/openuptime-agent.servicesystemd 单元文件

确认运行状态

systemctl status openuptime-agent
journalctl -u openuptime-agent -f
openuptime-agent --version

agent 启动时会输出一行日志,之后只在出错时记录。

添加更多服务器

同一个项目的所有服务器共用一个密钥,在每台服务器上运行同一条命令即可。服务器通过 /etc/machine-id 识别,所以在同一台机器上重新安装会保留历史数据。

从同一个镜像克隆出来的机器可能拥有相同的 machine id,会被当成同一台服务器。每台服务器详情页的系统一栏里能看到这个 ID。请为每台克隆机重新生成:

sudo rm /etc/machine-id
sudo systemd-machine-id-setup
sudo systemctl restart openuptime-agent

每个项目每小时最多新增 30 台服务器。每台服务器背后都有一个负责离线告警的心跳监控,它不占用套餐的监控数量。套餐限制服务器数量时,统计的是工作区内的所有项目。

选项

安装脚本支持以下选项:

选项说明
--key oua_…项目的 agent 密钥,安装时必填。
--endpoint URLagent 上报的地址。后台复制的命令会自动填入当前站点地址。必须使用 https://,只有 localhost、127.0.0.1 和 [::1] 允许用 http://,除非加上 --allow-http。
--name NAME服务器在 OpenUptime 中的名称,不传时使用主机名。不能包含引号、$、反引号和反斜杠。
--tags A,B服务器的标签,用逗号分隔,最多 20 个,每个不超过 32 个字符。
--allow-http允许上报到其他机器上的 http:// 地址,适用于没有配置 TLS 的内网自托管实例。此时密钥和指标会以明文在网络上传输,脚本也会打印警告。通过 HTTP 打开后台时,复制出的命令会自动带上这个参数。
--binary-url URL从这个 https:// 地址下载 agent,而不是从 GitHub 下载。也可以用环境变量 OPENUPTIME_AGENT_URL 指定。
--sha256 HEXagent 应有的 SHA-256 值。不传时,脚本读取 agent 同一目录下的 SHA256SUMS。
--uninstall停止服务并删除 agent,详见卸载。

--name 和 --tags 只在服务器第一次上报时生效。之后以服务器详情页里的设置为准,用其他值重新运行脚本也不会改动。这两个选项需要 agent 0.1.6 或更新版本,旧版 agent 仍以主机名注册。

agent 本身支持以下参数,也可以通过服务的 env 文件里的环境变量设置:

参数环境变量默认值
--keyOPENUPTIME_KEY无,必填
--endpointOPENUPTIME_ENDPOINThttps://openuptime.app
--nameOPENUPTIME_NAME主机名;仅在服务器注册时使用
--tagsOPENUPTIME_TAGS无;用逗号分隔,仅在服务器注册时使用
--ignore-interfacesOPENUPTIME_IGNORE_INTERFACESlo,docker,veth,br-,virbr,cni,flannel,cali,vxlan:要排除的网卡名前缀
--once以 JSON 输出一次上报内容后退出,不会发送
--version输出版本号后退出

安装后想修改设置,编辑 /etc/openuptime-agent.env 然后重启服务:

sudoedit /etc/openuptime-agent.env      # OPENUPTIME_KEY=oua_…
sudo systemctl restart openuptime-agent

想确切知道 agent 会发送什么,运行 openuptime-agent --once。

无法访问 GitHub 时

在仓库的 apps/agent 目录下自行编译(GOOS=linux GOARCH=amd64 CGO_ENABLED=0 go build),或从 release 中复制 agent,通过 HTTPS 放到服务器能访问的地址,再用 --binary-url 指定。安装脚本也一并复制到服务器上运行。

脚本会先校验下载的文件再运行它。可以在 agent 旁边放一个 SHA256SUMS 文件,按 sha256sum 的输出格式写上 agent 文件名对应的一行;也可以自己传入校验和:

sudo sh install.sh --key oua_… --endpoint https://your-openuptime.example \
  --binary-url https://mirror.example/openuptime-agent-linux-amd64 \
  --sha256 <64 hex characters from sha256sum>

上报哪些数据

  • CPU:所有核心的总体使用率、型号和核心数。
  • 内存和交换分区:已用、缓存和总量。
  • 负载:1、5、15 分钟平均负载。
  • 磁盘:所有已挂载的本地磁盘和网络共享(ext4、XFS、Btrfs、ZFS、NFS、CIFS 等)的已用和总容量,以及物理磁盘的读写速率。虚拟文件系统和容器文件系统不计入。同一个设备挂载在多个位置时,每个挂载点都会列出,但总量里只算一次。
  • 网络:每个网卡的 MAC 地址、全部 IPv4 和 IPv6 地址,以及接收和发送速率。默认排除回环、Docker、网桥和 overlay 网卡。
  • 系统:主机名、发行版、内核、架构、启动时间和运行时长。

每次上报最多列出 64 个磁盘和 64 个网卡,每个网卡最多 32 个 IPv4 和 32 个 IPv6 地址。主机上更多时(比如有几百个 ZFS 数据集或虚拟机 tap 网卡),只列出前面的部分。网络速率仍然统计所有网卡,磁盘总量只统计列出的磁盘。

点击服务器可以在侧边面板查看当前状态,打开详情页可以查看最近 1 小时、6 小时、24 小时或 7 天的图表。指标数据保留 7 天。

agent 只从 /proc 读取系统统计信息,另外读取 /etc/machine-id 和 /etc/os-release。它不会读取日志、进程、用户文件或服务器上存储的其他任何内容。

上报间隔

服务器默认每 60 秒上报一次。可以在服务器详情页的设置里单独修改:30 秒、1 分钟、2 分钟或 5 分钟。agent 会在下一次上报时采用新的间隔。比间隔早很多到达的上报会被确认,但不会存储。

在托管服务上,每 30 秒上报一次需要付费套餐;每个套餐对一个工作区(所有项目合计)的服务器数量也有上限。服务器不占用监控数量。自托管部署可以使用所有间隔,服务器数量也不受限制。

离线告警

大约三个间隔没有收到上报,服务器就算离线:默认间隔下是三分钟。每台服务器都有一个带 server 标签的心跳监控,届时会创建事件,恢复上报后自动解决,所以告警会像其他监控一样发送到项目的告警渠道。

在服务器详情页或列表的侧边面板里,可以在离线告警中设置:不需要关注的服务器可以关闭告警;可以选择发送到哪些渠道(不选就是全部渠道);重启或迁移之前可以静音 1 小时。覆盖这台服务器的维护窗口也会让告警静音。

在线服务器的 CPU、内存或最满的硬盘达到 90% 时,列表会标为占用过高。这只是提示,不会发送告警。筛选栏旁的眼睛按钮会遮住服务器页面上的所有地址并隐藏所有标签,方便共享屏幕,并记在当前浏览器里。

安全性

  • 只上报。agent 只向外发送数据,不监听任何端口,也无法执行后台下发的命令、脚本或更新。它从服务端读取的只有上报间隔。仓库里有一项测试,一旦 agent 的代码开始运行程序或监听端口就会失败。
  • 非特权运行。服务以临时系统用户运行(systemd DynamicUser),启用 NoNewPrivileges,系统目录只读(ProtectSystem=strict),/home 也只读。只有安装和卸载需要 root。
  • 下载经过校验。脚本运行 agent 之前,会用 release 的 SHA256SUMS 核对文件;下载地址和上报地址都只接受 HTTPS。
  • 密钥需要保密。拿到密钥的人可以在项目里注册服务器,也能冒充项目里的任意一台服务器上报,包括伪造数据来掩盖故障。它不能读取数据,也不能修改设置。只有工作区所有者和管理员能查看或轮换密钥。在服务器上,密钥保存在只有 root 可读的文件里,不会出现在进程列表中。
  • 密钥泄露时请轮换:进入服务器 → 添加服务器 → 轮换密钥。所有使用旧密钥的服务器都会停止上报,agent 每 5 分钟重试一次。把新密钥写入 /etc/openuptime-agent.env 并重启服务,或重新运行安装命令。

更新 agent

重新运行安装命令即可。它会下载最新的 release,替换程序文件,用你传入的密钥和地址重写 env 文件,然后重启服务。服务器的历史数据会保留。用 openuptime-agent --version 查看版本。

卸载

在服务器上运行:

curl -fsSL https://raw.githubusercontent.com/becoolme/openuptime/main/apps/agent/install.sh | sudo sh -s -- --uninstall

它会停止并禁用服务,删除脚本做了什么中列出的三个文件,并重新加载 systemd。服务器上不会留下其他东西,服务使用的临时用户也会随之消失。卸载不需要密钥。

然后在后台删除服务器:打开服务器详情页,点击删除服务器。这会删除它的历史数据和心跳监控。请按这个顺序操作:如果 agent 还在运行时就删除,服务器会在下一次上报时重新出现。每台服务器的详情页和添加服务器对话框里也有这条卸载命令。

不下载脚本、手动卸载:

sudo systemctl disable --now openuptime-agent
sudo rm -f /etc/systemd/system/openuptime-agent.service /usr/local/bin/openuptime-agent /etc/openuptime-agent.env
sudo systemctl daemon-reload

常见问题

现象处理方法
only Linux is supported、systemd is required、unsupported CPU architecture这台机器不在支持范围内,见环境要求。
run as root要通过管道交给 sudo sh -s -- …,而不是 sh。
download failed服务器无法访问 GitHub,或者 release 里还没有 agent。请检查出站 HTTPS,或使用 --binary-url。
checksum mismatch下载到的文件和 release 列出的不一致,没有安装任何东西。请重新运行命令;如果一直失败,不要从这个来源安装。
could not download SHA256SUMS、SHA256SUMS has no line镜像上没有 agent 的校验和文件。请补上,或者传入 --sha256,见无法访问 GitHub 时。
--endpoint must start with https://请使用 OpenUptime 站点的 HTTPS 地址。自托管实例如果只提供 HTTP,请在前面加一层 TLS 代理;如果两者之间的网络可信,也可以加上 --allow-http。
服务在运行,但服务器一直没出现查看 journalctl -u openuptime-agent。确认服务器能访问 OPENUPTIME_ENDPOINT 中的地址,它必须是你打开后台时使用的地址。
the agent key was rejected密钥已被轮换,或项目已被删除。把当前密钥写入 env 文件后重启服务。
HTTP 402、server_limit工作区的服务器数量已达到套餐上限。删除不再需要的服务器,或升级套餐。降级后,最早添加的服务器会继续上报。
HTTP 429一小时内新增了超过 30 台服务器。agent 会继续重试,稍后即可加入。
两台机器显示成了一台服务器它们的 machine id 相同,见添加更多服务器。
缺少某块磁盘或某个网卡运行 openuptime-agent --once 查看上报内容。名称以被排除前缀开头的网卡不会上报,可以修改 OPENUPTIME_IGNORE_INTERFACES。

Last updated 2026-11-18最后更新:2026-11-18