Documentation for automated readers
A curated documentation index is available at: https://grafana.com/llms.txt
A complete documentation index is available at: https://grafana.com/llms-full.txt
These indexes can help with page discovery before fetching individual documents.
This page is also available in Markdown, which may be easier for automated readers and AI tools to parse than HTML. The Markdown version is available at https://grafana.com/docs/grafana-cloud/observe-and-act/testing/synthetic-monitoring/set-up/set-up-private-probes/network-and-resource-requirements.md, or by sending Accept: text/markdown to https://grafana.com/docs/grafana-cloud/observe-and-act/testing/synthetic-monitoring/set-up/set-up-private-probes/network-and-resource-requirements/. For broader documentation discovery, the curated index is available at https://grafana.com/llms.txt and the complete index is available at https://grafana.com/llms-full.txt.
Network and resource requirements
Before setting up a Synthetic Monitoring private probe, review the connections it needs and estimate CPU and memory for your checks. Use these requirements to prepare the host or container environment and its firewall rules.
Firewall configuration
Private probes initiate connections to Synthetic Monitoring services and monitored targets. Configure the probe host’s firewall to allow the outbound connections in the following sections. If a monitored target has its own firewall, configure that firewall separately to allow inbound connections from the probe host.
You don’t need to expose the probe’s diagnostics port to the internet for checks to run. Keep port 4050 private. For information about collecting agent metrics, refer to Monitor private probes.
Synthetic Monitoring services
Find your stack’s row in the regional endpoint table, then allow outbound connections to these destinations:
- The API server URL, over TCP port
443. The agent uses this gRPC endpoint to connect to Synthetic Monitoring. - The Secrets proxy hostnames listed in that row, over HTTPS on TCP port
443.
Allow DNS resolution for these hostnames using your environment’s DNS resolver. The API connection uses TLS; a successful TCP connection alone doesn’t establish that a proxy or firewall permits the gRPC connection.
Prometheus and Loki instances
The agent sends check metrics to Prometheus and execution logs to Loki. Allow outbound connections to both configured services, not only to the API server:
- Log in to Grafana Cloud and open your stack.
- Navigate to Testing & synthetics > Synthetics > Configuration.
- On the General page, scroll down to Data sources.
- For the Prometheus server URL:
- Click the Prometheus data source. For example:
grafanacloud-STACK_NAME-prom. - Copy the Prometheus server URL under Settings > Connection.
- Click the Prometheus data source. For example:
- For the Loki URL:
- Click the Loki data source. For example:
grafanacloud-STACK_NAME-logs. - Copy the URL under Settings > Connection.
- Click the Loki data source. For example:
- Allow outbound HTTPS to the hostnames and ports from both URLs. Don’t use these URLs as the agent’s
API_SERVERvalue.
If you can’t open the data source settings, ask your Grafana administrator for these destinations.
External script imports
- If you use k6 scripts that import external modules, for example, k6 modules from
jslib.k6.io, allow outbound connections from the probe host to the module hosts.
Note
MultiHTTP test executions make outbound connections to
jslib.k6.iowhen run. If you run MultiHTTP checks on private probes behind a firewall, allow outbound connections from the probe host tojslib.k6.io.
For HTTPS imports, allow TCP port 443.
Monitored targets
Allow outbound connections from the probe host to each monitored target. The required protocol depends on the check type. For example:
- ICMP: for ping and traceroute checks.
- UDP or TCP: for DNS checks.
- TCP: for HTTP and MultiHTTP checks.
Use the ports configured in the check, such as TCP 443 for HTTPS. ICMP checks can also require process or container permissions; allowing network traffic doesn’t grant those permissions.
For scripted and browser checks, review the destinations used by script requests, redirects, and browser page dependencies. Allow outbound connections to those hosts and ports as well as the initial application URL.
If a target’s firewall blocks these connections, configure the target’s firewall, not the probe host’s firewall, to allow inbound connections from the probe host over the required protocol.
For k6-based checks that need to reach 10.0.0.0/8, also review private-network blocking. Changing a firewall rule doesn’t override the agent’s blocklist.
Usage statistics
If usage statistics reporting is enabled, allow outbound HTTPS to stats.grafana.org on TCP port 443. To disable reporting before starting the agent, follow Anonymous usage statistics reporting.
Calculate resource requirements
Resources used by private probes may vary based on check duration, execution overlap, script behavior, and the applications being monitored, so it can be hard to predict them in advance. Use the following estimates as an initial sizing guide.
CPU values use millicores: 1000m is one CPU core, and 200m is 0.2 cores. Memory values use mebibytes: 1024Mi is one gibibyte.
Choose one baseline for the most resource-intensive check type you run.
| Check types on the probe | Baseline CPU | Baseline memory |
|---|---|---|
| API endpoint checks only | 50m | 64Mi |
| MultiHTTP or scripted checks, with or without API endpoint checks | 200m | 256Mi |
| Browser checks, with or without other check types | 1000m | 1024Mi |
Group the checks assigned to this probe by check type and note their configured execution intervals. The additional estimates below use executions per minute, not the number of check definitions. Account for how often each check runs on this probe when estimating its workload.
Add the following estimate for each execution per minute:
| Check type | Additional CPU per execution/minute | Additional memory per execution/minute |
|---|---|---|
| API endpoint: HTTP, Ping, DNS, TCP, or Traceroute | 5m | 2Mi |
| MultiHTTP or scripted | 20m | 32Mi |
| Browser | 100m | 256Mi |
Example: five HTTP checks
Suppose your probe runs five HTTP checks, each configured to run every 30 seconds. Each check contributes two scheduled executions per minute, for a total of ten executions per minute on this probe.
Using the estimates above:
- Baseline for API endpoint checks:
50mCPU and64Mimemory. - Additional CPU:
10 × 5m = 50m. - Additional memory:
10 × 2Mi = 20Mi. - Total starting estimate:
100mCPU and84Mimemory.
This example illustrates the calculation, not a guaranteed minimum or a Kubernetes resource limit. Long-running checks and overlapping executions can change resource demand. Measure resource use with representative checks before using the estimate in production.
Validate resource use
Monitor CPU usage, CPU throttling, memory usage, and out-of-memory (OOM) events. In Kubernetes, set resource requests and limits using measured demand and your cluster policy rather than copying the estimates directly. Collect host or container resource telemetry alongside the agent metrics.
Next steps
Continue with Set up a private probe to register a probe, or install the agent if you already have a probe authentication token.
Was this page helpful?
Related resources from Grafana Labs


