Blog
A ping only proves a host answers ICMP. Learn what HTTP checks catch that ping misses, when ping still helps and how to layer ping, TCP and HTTP monitors.
In the ping monitoring vs HTTP monitoring question, the two checks answer different things. A ping (ICMP echo) tells you a host is reachable on the network, while an HTTP check tells you a web service actually returns a correct response to a real request. For websites and APIs, HTTP monitoring is the check that should alert you. Ping is a supporting signal for servers and network gear, and it is most useful when you layer it under TCP and HTTP checks.
The rest of this guide covers what each check proves and why ping gives false results in both directions. It also covers where ping is still the right tool and how to combine the layers so a failure tells you where to look.
A ping sends an ICMP echo request to an IP address and waits for an echo reply. If a reply arrives, you know three things: a route exists from the probe to the host, the host's kernel network stack is running, and nothing on the path is dropping ICMP. You also get round-trip time and packet loss.
$ ping -c 5 203.0.113.10
PING 203.0.113.10: 56 data bytes
64 bytes from 203.0.113.10: icmp_seq=0 ttl=54 time=21.4 ms
64 bytes from 203.0.113.10: icmp_seq=1 ttl=54 time=21.9 ms
...
5 packets transmitted, 5 packets received, 0.0% packet lossThe reply is generated by the operating system kernel, not by any application. Nginx, your app server, your database and your TLS configuration play no part in it. A server can answer pings within milliseconds while every service on it is dead.
Because ICMP never touches the application, a ping monitor stays green through most of the outages your users actually notice:
That last point matters more than it seems. Ping monitors usually target an IP or resolve the hostname once. Your users resolve the hostname every time, and a broken record takes the site down while the host keeps answering ICMP.
A successful ping proves the host's network stack is alive; only an HTTP check proves your users can get a working response.
Ping also produces false alarms in the opposite direction. Many environments drop or limit ICMP by design:
If you want ping to work for monitoring, allow echo requests explicitly and rate-limit them rather than opening ICMP fully:
# iptables: allow inbound echo requests, limited
iptables -A INPUT -p icmp --icmp-type echo-request -m limit --limit 5/second -j ACCEPTOne caution: do not "fix" this by blocking all ICMP elsewhere. Blocking ICMP wholesale breaks path MTU discovery on IPv4, and IPv6 depends on ICMPv6 for basic functions such as neighbor discovery. Block echo requests if you must, not ICMP as a protocol.
An HTTP check walks the same path a browser or API client takes: DNS resolution, TCP connect, TLS handshake, sending the request, receiving the status code and the body. Each step can fail independently, and a good HTTP monitor reports which one did. You can see the same breakdown with curl:
curl -s -o /dev/null -w "dns:%{time_namelookup} connect:%{time_connect} tls:%{time_appconnect} ttfb:%{time_starttransfer} total:%{time_total} code:%{http_code}\n" https://example.com/healthOn top of reachability, an HTTP monitor can assert things ping never could: an expected status code, a maximum response time, a keyword in the body, a valid certificate. For API endpoints these assertions are the core of the check, which we cover in detail in the guide on monitoring API endpoint uptime with status codes, latency and body checks.
Ping is not obsolete. It is the right check whenever the thing you care about is the network path or a device that has no application to query:
For ping monitors, treat a single lost reply as noise. A reasonable starting point is to consider the host down only after several consecutive failed checks, and to alert on latency relative to the host's own baseline rather than a fixed number, because a host on another continent has a very different normal than one in the same region.
The useful setup is not ping or HTTP, it is a stack where each layer answers one question. For a typical web server, build it like this:
The combinations map to causes like this:
ping tcp http likely cause
FAIL FAIL FAIL host down, network path or upstream provider
OK FAIL FAIL service stopped, port closed, firewall rule changed
OK OK FAIL app error, proxy 502/504, TLS, wrong content, DNS on hostname
OK OK OK healthy (or your check is not strict enough)
FAIL OK OK ICMP filtered, ping result is not meaningfulThe last row is the one people forget. If ping fails while TCP and HTTP succeed, the host is fine and ICMP is being dropped somewhere. Fix the firewall rule or remove the ping monitor; do not let it page anyone.
All layers should run from more than one location. A single probe with a bad route reports an outage that only exists for that probe. Requiring agreement from several regions before alerting is one of the most effective ways to reduce false positive alerts, and it applies to ping just as much as to HTTP.
If you use m0nitor, the layers above map to separate monitor types: ping (ICMP), TCP port, and HTTP(S), alongside DNS, SSL certificate expiry and heartbeat checks. The uptime monitoring and monitor types overview lists them. Checks run from distributed probes in several regions, and an outage is confirmed by multiple locations before an alert fires, which filters out single-route network blips. The same layering works with any monitoring tool that supports these check types.
No. Ping only shows that the server's network stack answers ICMP. A website can return errors, serve an expired certificate or have a broken DNS record while ping stays green. Use an HTTP check against the public hostname as the primary monitor and keep ping as a diagnostic layer if the host answers it.
Ping replies come from the operating system kernel, not from your web server or application. If ping succeeds and the site fails, the problem sits above the network: a stopped service, a proxy that cannot reach its upstream, an application error, a TLS problem or DNS pointing the hostname somewhere else. A TCP port check narrows it down further.
Allowing ICMP echo requests, ideally rate-limited, is generally fine and makes ping monitoring possible. You can restrict them to known monitoring sources if your probes publish their IPs. Avoid blocking ICMP entirely, since that breaks path MTU discovery on IPv4 and basic IPv6 operation.
For websites and APIs, yes as the alerting check, because it covers what users experience. Ping still has a place for routers, VPN gateways and servers without HTTP endpoints, and as context next to an HTTP alert to tell a dead host from a dead service.