banshee/web-collection/web/https/page

Web page

Checks a web page the way check_http does: its status code, its response time, its size, a text it must hold or must not hold, a header it must carry, and where its redirections lead. Eight watches are offered, each one set on the probe.

Versions

Generic: any web server, over http:// or https://, whatever its software.

How it works

One GET to the page address; nothing else is sent. The collector returns:

the status code of the last answer read (200, 301, 503…);
the response time, in milliseconds, of all the requests made, redirections included;
the size of the body, in characters;
the final address: the last page read, or, when the last answer is a redirection not followed, the address it points to;
the number of redirections followed.

Redirections: the HTTP function of the collectors follows none. The collector reads the location of a redirection (3xx) itself and follows it while three conditions hold: the number set in web:redirects is not reached, the address stays on the same host (http:// to https:// on the same host is followed), and the answer is a redirection with a location. A relative location (/login, login, //www.example/login) is completed from the page that sent it. A redirection to another host is never followed: its address is the final address, and "Unexpected redirection" watches it.
The texts are searched in the body of the last answer, as it arrives (HTML included), case included. Set them on the probe, in the setting of their watch: empty, the watch judges nothing.
The header expected is searched among the headers of the last answer, its name in any case; with a value set, the header must hold it. The value found is never written in the message: a header may carry a cookie.
An account or a token: a token goes as Authorization: Bearer, and then no account is sent beside it; otherwise an account goes in Basic. Either is sent to the redirections followed too, which stay on the same host.
https:// follows the TLS profile of the probe: a certificate it refuses stops the collector as misconfigured, with its reason.
A site that refuses the connection, or does not answer in the delay of protocol:http:timeout, is critical: "Site down". Any other failure (an unknown name, no route) leaves the probe unreachable.

The expiry of the certificate is not read here: deploy network/tls/certificate on the same host and port, which watches it, with its chain and the protocols the server accepts.

Prerequisites

The address reachable from the poller.
For a page behind an account or a token, protocol:http:user and protocol:http:password, or protocol:http:token, set on the host or one of its folders. They go to https:// only: a secret toward http:// stops the collector as misconfigured.

Variables

Name	Label	Says	Default
`web:url`	Page address	the full address, `http://` or `https://`, `https://www.example/health`	—
`web:host`	Site name	sent as the `Host` header, for a site that a server hosts beside others and that is reached by an IP address; optional, the name in the address when empty	—
`web:redirects`	Redirections followed	how many redirections are followed, on the same host only; 0: none	`0`

Outputs

Name	Label	Unit	Per instance
`status`	Status code	—	no
`response-time`	Response time	ms	no
`size`	Page size	characters	no
`final-url`	Final address	—	no
`redirects`	Redirections followed	—	no

Watches

Name	Title	Kind	Defaults	Checked by default
`unexpected-response`	Unexpected response	expected value, on `status`	critical when the code is anything but 200	yes
`expected-text`	Expected text missing	judged by the collector	critical when the page does not hold the text set; no text by default	no
`unwanted-text`	Unwanted text present	judged by the collector	critical when the page holds the text set; no text by default	no
`unexpected-redirect`	Unexpected redirection	expected value, on `final-url`	no address given: it raises nothing until an address is set on the probe	no
`slow`	Slow page	above a threshold, on `response-time`	warning above 1000 ms, critical above 3000 ms, for 3 runs	yes
`too-small`	Page too small	below a threshold, on `size`	critical below 1 character: an empty page	yes
`expected-header`	Expected header missing	judged by the collector	critical when the header set is missing, or does not hold the value set; no header by default	no
`slow-unusual`	Unusual response time	off its habit, on `response-time`	normal sensitivity, never under 500 ms	no

"Unusual response time" needs the history of the measure: until the poller has learnt it, the watch says it is learning and raises nothing.

Choose an expected text that only a page in good health shows, for instance the name of a service on a health page; as unwanted text, a sentence of the maintenance page. A page that redirects on purpose (http:// to https://) is checked either by following it (web:redirects at 1 or more), or by expecting its code (301 in "Unexpected response") and its target in "Unexpected redirection".

Limits

The size is counted in characters, not in bytes.
A redirection to another host is not followed, even when it is expected: a site that moves to another domain is checked by a probe on that domain.
Only GET is sent: a page that needs a form posted first (a login) is not served.
The page is read as the server sends it: a page built by JavaScript in the browser shows here as its empty frame.