How we check

What we check, how often, and when we call a site down.

What the checker does today and the numbers it works with, taken from the code that runs it. Where something is not built yet, this page says so.

  • 10sec

    An answer slower than this is a failed check.

  • 2failures

    In a row open an incident and send the alerts.

  • 2successes

    In a row close it and send the all-clear.

  • 1min

    Between checks while a site is failing, whatever its interval.

01 — Check types

Seven ways to check a service, and heartbeats for jobs.

Every monitor uses one of them. A check passes only when every rule set on it passes; the first rule that fails is the error you see in the alert.
  • HTTP and HTTPS
    https://

    Requests the URL with the method, headers and body you set, and times the name lookup, connect, TLS handshake, first byte and transfer. A keyword or JSON check reads the first 1 MB of the body.

    Down whenNo answer within 10 s; a status code outside the list you expect (200 to 399 unless you set one, such as 200-299,404); a keyword that is missing, or present when it must not be; a JSON value that does not match; or an answer slower than your limit (2 s by default for APIs, none for web pages). A redirect counts as an answer; with a keyword or JSON check it is followed and the final page is read.

  • TCP port
    tcp://

    Opens a connection to the host and port, timing the lookup and the connect.

    Down whenThe connection is refused, the host is unreachable, or nothing connects within 10 s.

  • UDP port
    udp://

    Sends one datagram, text or hex, to the host and port.

    Down whenWith an expected reply set: no reply containing it in time. Without one: only an ICMP “port unreachable” counts, because UDP has no handshake and silence is normal.

  • DNS
    dns://

    Looks up the A, AAAA, CNAME, MX or TXT records of a name.

    Down whenNo records of that type, or none that contain the value you expect.

  • Ping
    ping://

    Sends three ICMP echo requests. The response time is the average round trip.

    Down whenNo reply at all, or more packet loss than the limit you set.

  • Mail server
    smtp:// imap:// pop3://

    Connects to SMTP, IMAP or POP3, waits for the greeting, upgrades with STARTTLS or starts with TLS, and says goodbye. It never logs in. A TLS check reads the certificate every time.

    Down whenNo greeting, a server that refuses STARTTLS, a failed TLS handshake, or a timeout.

  • Browser flow
    Chromium

    Runs up to 20 steps in a real browser: go to a URL, click, fill a field, press a key, wait for an element, expect text, expect a URL. A run may take 60 s and one step 15 s.

    Down whenAny step fails, or the run runs out of time. The failed step, its error and a screenshot of the page are kept, with secret fields masked.

  • Heartbeat
    heartbeat://

    Not a check we make: your cron job or worker calls its own secret URL every time it runs.

    Down whenNo ping for the period you set plus a grace period (5 minutes by default, up to 24 hours).

02 — How often

As often as every minute, on every paid plan.

You pick the interval per monitor. The plan sets the shortest one you can pick.
PlanCheck everyBrowser flows
Starter10, 15, 30 or 60 minutesNone
Pro1, 5, 10, 15, 30 or 60 minutes3 tests, every 5 minutes or less often
Team1, 5, 10, 15, 30 or 60 minutes10 tests, every 5 minutes or less often
Agency1, 5, 10, 15, 30 or 60 minutes25 tests, every 5 minutes or less often
Enterprise1, 5, 10, 15, 30 or 60 minutes50 tests, every 5 minutes or less often
A failing site
Checked again after 1 minute, until the incident opens and for as long as it stays open.
Heartbeats
Looked at when their deadline passes. The period can be anything from a minute to a week.
Certificate and domain expiry
Read once a day. A mail server check over TLS reads its certificate on every check. The site's alert list, its channels and phones are warned 30, 14, 7, 3 and 1 days before a certificate ends and once it has expired; a domain 60, 30, 14, 7 and 1 days before, from Team. Each warning goes once, a renewal starts the count again, and nothing opens an incident.
Change monitoring
DNS records, the TLS certificate and page content hourly; the PageSpeed score on desktop and phone once a day. DNS and the certificate from Pro, content and PageSpeed from Team.
03 — When it is down

Two failed checks open an incident. Two good ones close it.

  1. 1

    One failed check is recorded, nothing is sent

    A single timeout is usually the network between us and the site. The check is stored with its error and the site is checked again after 1 minute, even when its interval is longer.

  2. 2

    Another region confirms it

    Before the incident opens, every other region checks the site at once. It opens only if at least one of them sees the failure too. If every region that answers sees the site up, nothing is sent and the main region keeps checking. A region that does not answer within 20 s never holds an alert back: with no answer at all, the incident opens as before.

  3. 3

    2 failed checks in a row open an incident

    The incident starts at the time of the first failed check, not the second, so its length is the real length of the outage. The alerts go out and the status page shows the site as down.

  4. 4

    2 good checks in a row close it

    The incident ends at the first good check. Everyone who got the down alert gets the all-clear with how long it lasted.

  5. 5

    Planned maintenance stays quiet

    During a maintenance window the checks still run and are recorded, but no incident opens and nobody is alerted. Status pages announce a window up to 7 days ahead.

  6. 6

    Resolved by hand, counted again

    If someone resolves an incident while the site is still failing, the count starts over: a new incident opens after the next 2 failed checks.

Incident lifecycle · one-minute checks
  1. 14:02:00Check fails

    pay.acme.com did not answer within 10 s. Nothing is sent yet.

  2. 14:03:00Confirmed

    The next check fails too. The incident opens and the status page shows Checkout as down.

  3. 14:03:02Alert sentTeams · #ops

    Microsoft Teams message in #ops, right away or after the delay set on that channel.

  4. 14:05:00Recovered

    Two good checks in a row. The incident closes after 2 min and #ops hears about that too.

The incident is counted from 14:02, the first failed check, to the first of the two good ones.
04 — Alerts

Who hears about it, how and when.

Every path keeps its own record of what it sent, so a restart of the checker neither repeats an alert nor loses one.
Email
On every plan. Sent the moment the incident opens, to owners, admins and the site admins assigned to the site, plus up to 10 extra addresses on the site. One down email and one all-clear per incident. A site with an escalation policy is emailed by its steps instead.
Phone push
To the same people, on every phone signed in to the Heedery app. Marked time-sensitive, so it comes through Focus. A tap opens the incident to acknowledge it.
Slack, Microsoft Teams, Discord, Telegram, webhooks
From Pro. Each channel can wait from 1 to 60 minutes before it posts; an outage shorter than that wait never posts. Webhooks are signed with HMAC-SHA256. A failed send is tried again with a growing pause, up to 5 times.
Escalation policies
From Pro. Up to 10 steps, each after a delay you set, notifying people, channels or whoever is on call in a schedule. The last step can repeat every minute up to every 24 hours, at most 100 times. Acknowledging from the email, the push or the chat message stops the rest.
SMS and phone calls
From Pro, through escalation policies, to people who confirmed a US or Canadian mobile number. The SMS carries the acknowledge link; the call reads the alert out twice; the all-clear comes by SMS only. Each one uses credits: Team includes 50 a month and Agency 200, used first and renewed monthly, and prepaid packs that never expire top them up. A message the carrier does not deliver or a call nobody answers gets its credits back.
Status page subscribers
Emailed once when an incident on a site their page shows opens and once when it resolves, and again for each update you publish.
05 — The figures

How uptime, the SLA and response times are worked out.

Uptime on status pages and site pages

For each day, the share of the day that was not inside an incident; the figure for a period is the average of the days that had monitoring. It is based on incident time rather than failed checks because a failing site is checked more often, so counting checks would weigh an outage several times over. Days before a site was added do not count against it.

Uptime in monthly reports and the SLA

Checks that passed divided by checks made, from hourly figures. Hours that fall wholly inside a planned maintenance window of that site are left out. A report compares the result with the target you set for the client, from 90% to 100% (99.9% unless you change it), and marks it met or missed. Across several sites, each site counts once, so a site checked every minute does not outweigh one checked every 15.

Average response time

The mean of every check’s response time over the period, from 5-minute and hourly figures weighted by the number of checks in each.

95th percentile

95% of the successful checks answered in this time or less. Failed checks are left out, so a timeout does not set the figure. The site page works it out from the last 24 hours of raw checks; a monthly report from daily summaries of every successful check.

A checker that stops

If no result arrives for two check periods and two minutes more, the site shows as unknown instead of staying green, so a stalled checker cannot pass for a healthy site.

How long it is kept

Every check, with its timings
7 days
5-minute figures
30 days
Hourly figures
90 days
Daily uptime
Kept for good

Site pages show 30 days on Starter, 60 days on Pro, 90 days on Team, 90 days on Agency. Team and up keep showing daily uptime past that.

06 — Where checks run from

3 locations.

Every check runs on schedule from Virginia, United States, the main region. Frankfurt, Germany and Singapore check too: before an incident opens, they check the site at once, and it opens only if one of them also sees it fail. A network problem between one location and your site no longer wakes anyone up.

On Team and up, each region also checks every site every 5 minutes or as often as the site is checked if that is less often, and the site page shows the response time from each on a map. Uptime and the SLA are still worked out from the main region's checks.

  • Virginia, United States
    Main region
    Checking
  • Frankfurt, Germany
    Confirms outages
    Checking
  • Singapore
    Confirms outages
    Checking
User agent
Mozilla/5.0 (compatible; Heedery/1.0; +https://heedery.com/checker)
Public addresses only
Private and internal addresses are refused, on redirects too, and a name is checked again on every request. Internal services are watched by a heartbeat: a job inside the network pings us on schedule.
Where your site's server is
On every plan the site page marks where your site's server is: the address its name resolves to, looked up every six hours in DB-IP's free database, which is updated monthly. Behind a CDN such as Cloudflare there is no one place: the CDN picks the point that answers each request, so the map shows the point that answered each region most often over the last 24 hours, with each point's share. Name the origin server in a site's settings to place it instead. IP geolocation by DB-IP

See it on your own site.

Add a monitor and the first check runs within a minute.

Start for free