Monitors
A monitor is one thing you want watched: a URL, a host, a port, a DNS name, a certificate, a domain, a job that must call home, a user journey, a page in a real browser, a database or a mail server. This page lists every type, its options and how results are judged.
Monitor types
| Type | What it checks | Details |
|---|---|---|
| HTTP(S) | Availability, status code, response time, optional SSL and content checks | below |
| Keyword | A word or phrase must (or must not) appear in the page | below |
| API | JSON/XML endpoint with method, headers, body and response validation | below |
| Ping | ICMP reachability, latency, packet loss and jitter | below |
| Port | A TCP (or UDP service) port accepts connections | below |
| DNS | A record type resolves to the expected value; changes | below |
| SSL | Certificate validity, chain, hostname, expiry and TLS version | below |
| Domain | Domain expiry, registrar and nameservers through RDAP/WHOIS | below |
| Heartbeat | Your cron job or service calls us; silence means down | below |
| Transaction | A multi-step journey (log in, add to cart, check out), each step asserted, values carried from step to step | Synthetic monitoring |
| Browser | A real headless Chrome loads the page: load time, paint, layout shift, requests, failed requests, console errors | Synthetic monitoring |
| MySQL / MariaDB, PostgreSQL | Sign in, run a read-only query; connections, replication lag | Database and mail server monitors |
| Redis, MongoDB, Memcached | PING / hello / stats; memory, connections, hit ratio, role, replication | Database and mail server monitors |
| SMTP, IMAP, POP3 | Greeting, STARTTLS or implicit TLS, certificate, optional sign-in | Database and mail server monitors |
| FTP | Greeting, explicit or implicit FTPS, optional sign-in and a path that must exist | Database and mail server monitors |
Common settings
| Setting | Meaning |
|---|---|
| Name | Shown everywhere, up to 200 characters |
| Target | URL, host or IP, depending on the type (see below). Private and reserved addresses are refused by the SSRF guard unless the administrator allows private targets; the cloud-metadata address 169.254.169.254 is never allowed |
| Interval | How often the check runs: 30 s, 1, 2, 5, 10, 15, 30 or 60 minutes. Your plan sets the fastest interval you can pick; a faster value is clamped to it |
| Timeout | 5–120 seconds (default 30) |
| Confirmations | 1–10. On a single location: that many consecutive failures before the status flips to Down. On several locations: that many locations must see the failure (see below) |
| Locations | core (the main server), any public probe the administrator runs, and the location pools region:<Region> and auto:<N> (see below), up to the number your plan allows |
| Group, tags, notes | For your own organisation of the list |
| Public | Lets the monitor be embedded (/embed/<token>), shown as a badge (/badge/<token>.svg) and opened at /m/<token> without signing in |
| Alert channels | Which channels hear about this monitor. Leave empty to use the organisation's default channels |
| Alert on degraded | Also notify when only some locations fail. Set through the API (config.alert_degraded); the panel form has no switch for it yet |
Changing the type or the target resets the status to Pending and schedules a check at once. Pause stops checks and drops any that are already queued; Resume schedules one immediately.
Confirmations and locations
Every result is stored with the location it came from. The status is decided from the newest result of every location:
- One location: a failure increases the failure counter; the monitor is Down once the counter reaches Confirmations. One passing check resets it and puts the monitor Up.
- Several locations (paid plans): the newest result of each location within two intervals is compared. All up → Up. At least Confirmations locations down (or all of them when there are fewer) → Down. Some down but fewer than that → Degraded (a regional problem; it is recorded and, when Alert on degraded is on, alerted).
A probe that goes offline, or that the administrator is draining, is swapped for the least busy online probe of the same region, or for the main server, so checks never stop. Results keep the code of the probe that really ran the check.
Location pools
Instead of fixed probes you can pick pools in the Location pools card of the monitor form (shown when the administrator runs probes in at least one region; not for heartbeats):
- Any probe in a region (
region:<Region>, for exampleregion:Europe): each check runs on one online probe of that region, the least busy one at that moment. A pool counts as one location of your plan. When no probe of the region is online, the main server runs the check. - Automatic spread (
auto:<N>, offered from 2 up to the number of regions): each check runs on N probes from N different regions, again the least busy probe of each region. It counts as N locations of your plan. Regions you already use through a fixed probe or a region pool are skipped.
"Least busy" compares the jobs waiting for a probe and the queue it reported in its last heartbeat with its capacity (max jobs × weight). Pools can be combined with fixed locations, and the confirmation rule above treats auto:<N> as N locations. Through the API, send the same tokens in locations:
"locations": ["core", "region:Europe", "auto:3"]
An unknown region or an auto value outside the allowed range is refused when you save.
HTTP(S)
Fetches the URL and judges the answer. Options:
- Method GET, POST, HEAD, PUT, PATCH, DELETE or OPTIONS, with an optional body. A JSON body gets
Content-Type: application/jsonautomatically, anything elseapplication/x-www-form-urlencoded, unless you set the header yourself. - Headers one per line,
Name: value. - Basic auth user and password (the password is stored encrypted; leaving the field empty on edit keeps it).
- Bearer token or an API key header (config
bearer,api_key_header+api_key; encrypted at rest, shown as***in the API). - Expected status codes, see the syntax below. Empty means 200–399.
- Follow redirects (on by default, max 5 hops by default, configurable). Redirects are followed by hand and every hop goes through the SSRF guard again. A redirect loop is Down.
- Ignore SSL errors for self-signed certificates.
- Max response time in ms: a slower answer counts as Down ("Response took 1200 ms (limit 300 ms)").
- Expected headers one per line,
Name: value(value matched case-insensitively as a substring; empty value = header must exist). - Custom user agent.
- SSL check: reads the certificate from the handshake, stores the expiry date and warns the configured number of days (1–90, default 14) before it runs out.
- Detect content changes: hashes the visible text of the page and raises a Content changed alert when it differs from the previous check.
- Keyword (optional on HTTP too): a word that must be present or absent.
The result records the status code, response time, time-to-first-byte and other timings, final URL after redirects, the IP used, the size, the Server and Content-Type headers.
Expected status codes syntax
Comma-separated list of single codes, ranges and classes:
200
200,301,302
200-299,404
2xx,3xx
2xx means any code from 200 to 299. The check is Up when the code matches any entry.
Keyword
Same request as HTTP(S), then the page body is searched for the keyword (up to 500 characters). Choose must be present or must be absent ("Maintenance" absent, "Add to cart" present). Case-insensitive by default; config case_sensitive makes it exact.
API
For JSON or XML endpoints. Same request options as HTTP(S) (method, headers, body, auth, bearer/API key), then the body is validated.
Response format: json (default) or xml.
JSON rules, one per line: a dotted path, an operator and a value.
status == ok
count > 3
data.items.1.id == 2
data.version matches ^2\.
data.missing !exists
tags contains production
error != true
| Operator | Meaning |
|---|---|
exists / !exists | the path is (not) present |
== / != | string comparison; booleans compare as true/false, objects and arrays as JSON |
> < >= <= | numeric comparison |
contains | substring, or element of an array |
matches | regular expression (PCRE, without delimiters) |
Paths use dots for both object keys and array indexes (data.items.0.id). $ or . is the root. A body that is not valid JSON is Down.
XPath rules (XML), one per line: //status == ok, //count exists, //name contains foo, //x != y. The first node matched by the path is compared.
Body regex (config body_regex): the whole body must match the pattern.
A keyword can be combined with the rules.
Ping
ICMP echo through the system ping binary. Target: a hostname or IP (resolved and validated first). Options: count 1–10 (default 4), per-packet timeout 1–10 s, max packet loss % (Down when reached), max average latency in ms. The result stores sent/received, loss %, min/avg/max and jitter.
Where PHP has no shell access (shell_exec disabled, common on shared hosting) the check falls back to a TCP connection on ports 443, 80 then 22 and says so in the details.
Port
Opens a TCP connection to host:port (give the port in its own field or as host:port). Known ports are labelled (HTTP, SSH, SMTP, MySQL, cPanel, WHM, DirectAdmin, Plesk, RDP, …). For SSH, FTP, SMTP, POP3 and IMAP the banner is read; for Redis a PING is sent; for MySQL the server version is read. Expected banner (config expect_banner) makes the check Down when the banner does not contain the text.
UDP (config protocol = udp): port 53 sends a real DNS query and expects an answer; port 123 sends an NTP request; any other UDP port is best-effort (only an ICMP "port unreachable" counts as Down, because most UDP services do not reply to an empty packet).
DNS
Resolves the hostname for one record type: A, AAAA, CNAME, MX, TXT, NS, SOA, PTR, SRV or CAA. Leave Resolver empty to use the system resolver, or name one (1.1.1.1, ns1.example.net): a small built-in UDP DNS client asks that server directly, so "what does Cloudflare answer" works everywhere PHP runs.
Expected values (one per line or comma-separated): with match mode any (default) at least one expected value must appear in the answer (substring match, so google.com matches an MX aspmx.l.google.com); with exact the answer set must equal the expected set. No record at all is Down.
The answer set is hashed; when it changes between checks a DNS changed alert is raised (once per hour at most).
SSL
Connects with TLS to host (port 443 unless you give another) with SNI and reads the certificate: validity, chain trust, hostname match, expiry date, issuer, subject names and TLS version. Down when the certificate is expired, does not match the hostname, is not trusted (unless Ignore SSL errors is on), or the handshake fails. Minimum TLS version (config min_tls, e.g. 1.2) makes older protocols Down. Warning days (1–90) sets when the daily expiry warning starts; config expiry_is_down turns "expires within the window" into Down.
Domain
Looks the registrable domain up through RDAP (the registries' JSON service), with WHOIS on port 43 as the fallback for TLDs without RDAP. It stores the expiry date, registrar, status and nameservers. Down when the domain is not registered, has expired, or has a status such as redemption, pending delete or client/server hold. Expected nameservers (one per line) must all appear in the registry record. Warning days (1–365, default 30) sets the daily expiry warning; config expiry_is_down makes the window itself Down. Some registries do not publish an expiry date: the check then says so and stays Up.
Heartbeat
Nothing is probed. Your cron job, backup script or service calls a URL on its schedule; silence means Down. After saving, the monitor page shows the heartbeat URL:
https://your-monitor.example/hb/<token>
Set the expected interval (30 s to 7 days) to how often the job runs. The monitor is Down when no call has arrived for interval + grace (grace defaults to half the interval; config grace_sec overrides it). It is evaluated once a minute by the housekeeping job, not by probes, so it needs no location.
A cron job that pings after it finished:
0 3 * * * /usr/local/bin/backup.sh && curl -fsS https://your-monitor.example/hb/<token> >/dev/null
The job can also report a failure itself; the monitor goes Down at once and the message is kept:
curl -fsS "https://your-monitor.example/hb/<token>?status=fail&msg=disk+full"
GET and POST both work; status=error is the same as fail. The endpoint is rate-limited to 600 calls per minute per address, answers JSON ({"status":"ok",...}) and refuses paused monitors with 403. The old address /public/heartbeat.php?token=… still works.
Alerts raised by monitors
Besides Down / Up / Degraded, monitors raise SSL expiring (also for database, mail and FTP monitors that use TLS with certificate checks), Domain expiring (once a day while inside the warning window), DNS changed and Content changed. Each of them is an alert, an in-app notification, a webhook event and a message to the mapped channels. See Alerts.
Results and history
Raw results are kept for the retention period of your plan (30 days on the free plan, up to 2 years). Older results are rolled up per day (checks, up/down counts, average/min/max response time, incidents, downtime), and the roll-ups are kept for 2 years, so the 90-day and 365-day uptime figures survive the pruning. The results list, the uptime bars and the response-time chart (bucketed so a 30-second monitor is still readable) are on the monitor page, and the same data is available through the API.
একটি ক্লিক আমাদের এই ডকস উন্নত করতে সাহায্য করে।

