Thirty-seven kinds of check. One place to read them.

Nothing to install anywhere. Add a host, choose the checks, and Gryphon reaches it the way your customers do.

Ping

Judged on packet loss rather than on a single lost packet. A host that goes silent is only called down once a TCP probe agrees, so a firewall that drops ICMP does not read as an outage.

HTTP and HTTPS

The host's own address, answering with any 2xx.

A page's content

One specific page, with the method, the status codes that mean up, and text the page must contain — so a 200 that renders an error page is still caught.

SSL certificate

Verified as a browser verifies it: trusted, complete, and issued for this host's name. Then a warning 30 days before it — or the intermediate it was issued under — expires, and a problem at 7, on any port rather than only 443.

Domain expiry

A warning 15 days before the registration lapses and a problem at 7, read from the registry rather than from the zone.

DNS

Is the nameserver answering, and is it answering with the record you expect. A, AAAA, CNAME, MX, NS or TXT, over UDP or TCP, against the resolver of your choice.

Postgres

On a port open to the network, with a user and password or with neither: a bare handshake still tells a server that is up from one that is still recovering.

MariaDB and MySQL

The same, on the port you give. The handshake completes properly rather than leaving a half-open connection in the server's log.

Redis

A ping to the instance, with a password when it wants one.

Any TCP port

Is the port open and accepting connections. For anything that speaks a protocol Gryphon has never heard of.

Send and expect

Open the port, optionally send a line, and check what comes back: 220 from SMTP, SSH- from SSH, +OK from POP3, your own string from your own service. With TLS if it speaks it. How to set one up

Heartbeat

The other way round: your backup, cron job or queue worker calls its own URL each time it finishes, and Gryphon raises a problem when that call is late or the job reports a failure. The one way to notice a job that has quietly stopped running. The job can run on any machine that can reach Gryphon, including one behind your firewall with no ports open to the world.

If a firewall stands in front of what you monitor, the addresses these checks come from are published.

With the Gryphon agent: one binary on Linux or Windows, a menu bar app on macOS.

Disk space

Per filesystem, with your own warning and problem levels for each check, and the last 24 hours drawn as a trend with those levels marked on it.

Memory

What is actually available, not what is merely unallocated, so cache is not mistaken for pressure.

CPU and load average

Load read against the core count, so the same threshold means the same thing on a 2-core box and a 64-core one.

Postgres, MariaDB/MySQL and Redis

Bound to localhost, so the port never has to be opened to the network. A password, if the check signs in, is stored encrypted and reaches the agent only over HTTPS. Or with no credential at all.

HTTP, HTTPS and ping

Sent from the monitored host, so they reach what the outside cannot: private addresses, internal names, an admin page behind the firewall, a self-signed certificate.

TCP port and send-and-expect

The same two port checks, dialled from inside: the host itself by default, or any address on its own network. For the ports that are never opened to the network, such as an internal mail relay or a service's admin port. With TLS if it speaks it.

Your own scripts

Any script or Nagios plugin you put in the agent's script directory. Its exit code is the status and its first line is the message. What may run is decided on the host, never from Gryphon.

File freshness

Whether last night's backup, dump or export actually landed: how long ago a file was written, or the newest file in a folder, optionally only names like *.sql.gz. Too old is a warning and then a problem at the ages you set; missing, or smaller than it should be, is a problem at once. No change to the job, and the agent looks only in folders the host's administrator names.