Linux server monitoring: Monitor every server, not just your containers.
systemd services, PHP-FPM pools, Node apps under pm2, Go binaries, Java, Python workers, and the containers next to them. One agent per server, one command, in any language. When something breaks, WatchLog says why in plain words.
curl -fsSL …/install.sh | sudo sh -s -- --token wl_ct_••••
Install with one command, a deb, an rpm or docker run.How the server install works
- nginx.serviceServiceactive46 MiB
- php-fpm: pool wwwProcess20 instances · connections failing1.1 GiB
- mariadb.serviceServicerestarting · OOM-killed2.0 / 2.0 GiB
- cron.serviceServiceactive3 MiB
- redisContainerrunning · healthy210 MiB
Likely causemariadb.service is down. pool www’s connections to it fail since 14:02, when mariadb.service was OOM-killed at its 2 GiB memory limit.
- nginx.serviceServiceactive46 MiB
- apiProcesscrash-looping · exit 11.0 GiB
- workerProcess2 instances380 MiB
- [email protected]Serviceactive3.1 GiB
- redisContainerrunning · healthy210 MiB
Likely causeThe new version fails since its last change. api, a pm2 app, exits with code 1 about 20 s after every start, since its executable changed on disk at 14:01.
- caddy.serviceServiceactive38 MiB
- orders.serviceServicerestarting · OOM-killed512 / 512 MiB
- billing.serviceServiceactive96 MiB
- sync.pyProcess1 instance · started by hand64 MiB
- natsContainerrunning45 MiB
Likely causeOOM-killed at its 512 MiB memory limit. orders.service was OOM-killed at its MemoryMax three times in 20 minutes; systemd restarted it each time.
An example. Names are illustrative.
Services, processes and containers
- Services
Every systemd unit
State, restarts, exit codes, CPU, memory against MemoryMax, throttling at CPUQuota, tasks and I/O. A failed backup job at 03:00 is an alert, not a surprise.
- Processs
Grouped the way you think
pm2 apps by name, supervisord programs, PHP-FPM pools, gunicorn and uwsgi apps, a jar, a script someone started with nohup. Your own rules for the rest.
- Containers
Docker, next to the rest
Containers and Compose projects on the same server, in the same table and the same map. Kubernetes has its own views.
- PHP-FPM
- Node.js and pm2
- Go
- Java
- Python: gunicorn, uwsgi, supervisord
- Ruby
- .NET
- Rust
- a binary started with nohup
Language doesn’t matter: WatchLog watches processes, cgroups and the network, never your code.
What every server gets on day one
- Crashes, from the kernel
Every exit and its code or signal, for any process, with or without systemd. Crash loops and OOM kills are named, with the limit that was hit.
- Why it broke
Probable cause in plain words: a new release, a memory limit, a dependency that is down or a CPU limit, with what was checked.
- Who talks to whom
A live network map from eBPF: nginx to PHP-FPM to MariaDB, your Go service to Redis. HTTP status and latency for plain HTTP/1.x, without touching code.
- Your databases and caches
PostgreSQL, MySQL and MariaDB, Redis, MongoDB, RabbitMQ, Kafka, nginx, Traefik and PHP-FPM, installed on the server or in Docker, found by their service’s name.
- Alerts that work on day one
A service failed, a crash loop, an OOM kill, a full disk, a PHP-FPM pool at pm.max_children. To Slack, Telegram, email or a webhook.
- Each server’s own events
New in 0.9.2: a service failed (with systemd’s result and exit code), started, stopped or restarted; a process group’s failed exit, crash loop or release; an OOM kill with the limit it hit. Grouped like Kubernetes events, filtered by server and workload.
- Logs when you need them
The last lines of a unit’s journal, a pm2 app’s or a supervisord program’s log, on demand, with secrets masked. Never stored, never indexed.
One product for all three
The same dashboard, alerts and probable cause for your Kubernetes clusters and Docker hosts.
Kubernetes
Docker
Linux servers
Questions
Do I need Docker or Kubernetes?
No. The agent installs as a normal systemd service, from one command, a deb or an rpm. If Docker is there, its containers show too.
Do I have to change my code?
No. WatchLog reads processes, cgroups and network traffic. It works the same for PHP, Node, Go, Java, Python or a compiled binary.
Does the agent run as root?
No. It runs as its own user, watchlog, with five capabilities. Command lines, arguments, environment values and process ids never leave the server.
Does it monitor pm2 apps and PHP-FPM pools?
Yes. pm2 apps are found by name and each gets its own row, CPU, memory, restarts, logs and connections. A PHP-FPM pool shows as a process group, and with two lines in the pool’s file WatchLog also reads its busy workers, waiting requests and pm.max_children.
Is this APM or tracing?
No. You see each process’s resources, crashes, connections and HTTP status and latency, not the functions inside it.
Which Linux distributions?
Ubuntu 20.04+, Debian 11+, RHEL, Rocky and Alma 8+ and Amazon Linux 2023, on amd64 and arm64. Without systemd, processes still show. The network map needs Linux 5.8 or newer with BTF.
What does it cost?
Self-hosted is free and open source (AGPL-3.0) with no server limit. WatchLog Cloud Free covers 5 nodes or servers; Pro is $3 per node or server a month, $15 minimum.
Your servers, understood in five minutes.
Free and open source to self-host, with no limit on servers. Cloud Free covers 5 servers; Pro is $3 per server a month, $15 minimum.