Create a network test

How a test works

A test is a probe the platform assigns to one or more agents on a schedule. The agents run it from where they are, report each result, and the platform keeps the series, evaluates the thresholds and raises alerts. Create one under Network → Tests → Create Test (/network/tests/create); a new or edited test reaches its agents within about a minute.

The nine test types

Type	What it measures	Reach for it when
ICMP Ping	Round-trip latency, jitter and packet loss to a host, over a configurable count, interval and payload size.	The baseline "is the path healthy" check. Cheap enough to run every minute.
TCP Connect	Whether a port accepts connections, and how long the handshake takes.	A database, a message broker, anything that does not speak HTTP — and any target that filters ICMP.
UDP	UDP reachability and response time.	DNS servers, NTP, VoIP signalling.
Traceroute	The hop-by-hop route, with per-hop latency, over ICMP, UDP or TCP up to a maximum hop count.	You want to localize a problem, not just detect it. Feeds Path Analysis.
DNS Resolution	Resolution time and the answer for a record type, against the system resolver or one you name, over UDP or TCP.	"The site is slow" that turns out to be a slow resolver; validating that a record has propagated.
HTTP/HTTPS	A full request with method, headers, body and authentication (basic, bearer, API key, or mutual TLS with a client certificate and custom CA), plus the timing breakdown — DNS, connect, TLS handshake, first byte — and assertions on status, response time, headers, body and size.	An API or health endpoint. Assertions turn "it answered" into "it answered correctly".
TLS Certificate	Certificate expiry, chain validity, the expected common name, cipher suite and chain size, with a warning window in days.	Every public endpoint you own. This is the test that pages you before the certificate expires.
MTU Discovery	The largest packet that traverses the path unfragmented, searched between a minimum and maximum size.	VPN and tunnel paths, where a silently reduced MTU shows up as mysterious stalls on large transfers.
P2P Bandwidth	Throughput between two agents for a set duration.	Capacity between sites or regions. It floods the link while it runs — schedule it off-peak.

Create a test, step by step

1. Type and target. Pick a type, then the target host (an IP or name; a URL for HTTP) and, where it applies, the port. Each type then shows its own options — ping count and payload, traceroute protocol and max hops, DNS record type and resolver, the HTTP request builder, TLS expectations, MTU bounds, or the target agent for a bandwidth test.
2. Source agents. Choose which agents run it. You can name them, or target dynamically — all agents in a region or all agents with a tag — so an agent you install next month in eu picks up every eu test on its own.
3. Schedule. A preset from every minute to weekly, or a custom cron expression (Spring format: seconds, minutes, hours, day, month, weekday). One to five minutes suits reachability; hourly is plenty for a certificate check.
4. Alerts. Enable notifications and set the latency threshold (ms) and packet loss threshold (%), then pick the notification channels. A result that crosses either raises an alert through the platform's normal routing — silences, on-call schedules and escalation apply.
5. Create Test. The first results appear on the Dashboard after the first scheduled run.

Design for localization, not just detection. A ping test tells you a path degraded; the same target with a traceroute test from the same agent tells you which hop. Pair them. And run the important ones from at least two agents, so a single agent's own host or uplink cannot masquerade as a target problem.

What HTTP tests capture

Beyond up or down, an HTTP test records the whole timing breakdown — DNS lookup, TCP connect, TLS handshake, time to first byte, total — and the TLS facts of the connection: cipher suite, certificate chain length and size. A slow handshake, an oversized chain or an intermediate about to expire are all visible from the result, not just a failed request.

Cases

• Database reachability: a TCP Connect test from the application VPC's agent to the database host, every minute, alerting on loss.
• Third-party dependency: an HTTP test to a payment provider with a status-code assertion, from two regions, to prove "it's them, not us" with timing to show for it.
• Branch to head office: ping plus traceroute from the branch agent to the head-office gateway, so a degradation comes with the hop that caused it.
• Certificate hygiene: a TLS Certificate test on every public hostname with a 21-day warning window, hourly.
• Inter-region capacity: a P2P Bandwidth test between the two regional agents, nightly.

Editing a test opens the same form pre-filled (/network/tests/edit/:testId). Assigned agents pick up the change on their next poll — no restart, nothing to redeploy.