Applies to: Windows Server 2019, Windows Server 2022 and Windows Server 2025, subject to the workload and edition requirements below.
Define the failing connection before testing
Network troubleshooting is most efficient when the failure is specific: which source computer, destination name, protocol, port and time are involved? “The server has no network” can mean a DNS failure, one blocked application port, an incorrect route, a failed VPN or a genuine link outage. These problems require different evidence. Record the user-visible error and whether the issue affects all clients or only one path before changing configuration.
This guide uses built-in PowerShell tools to narrow a Windows Server connectivity fault without immediately resetting adapters or disabling security controls. Its examples target an illustrative HTTPS application at app01.corp.example on TCP 443. Substitute the actual endpoint and required port. A successful TCP connection still does not establish that TLS, authentication or the application itself works, so keep transport and application checks distinct.
Preserve a baseline from the affected server
Run diagnostics on the computer that experiences the failure. A nearby workstation may use another DNS server, VLAN, proxy or route. Record the server name, current time and adapter identities. Use a reliable alternate management path if the failing interface carries your remote session; avoid a repair command that disconnects you before its result can be observed.
hostname
Get-Date
Get-NetAdapter | Select-Object Name,ifIndex,MacAddress,Status,LinkSpeed
Get-NetIPConfiguration
Get-NetConnectionProfile
Get-DnsClientServerAddressLook for a disconnected adapter, an unexpected address, an incorrect subnet mask or a missing gateway. An automatically assigned address can indicate failed DHCP, but first confirm that DHCP is actually intended. On multihomed servers, identify the interface associated with the failing destination. A Public profile on a domain network can be a symptom of domain-location or DNS problems and may also affect firewall behaviour. Do not relabel it simply to bypass a rule.
Test name resolution independently
Resolve the exact name used by the application and inspect all returned addresses, including IPv6. Compare them with the expected service endpoint. A load balancer, split DNS or stale record can point different clients at different destinations. If the application uses a short name, test the FQDN too and review suffix search behaviour rather than assuming both names identify the same host.
Resolve-DnsName 'app01.corp.example'
Get-DnsClientCache | Where-Object Entry -Like '*app01*'
# Query a specific approved resolver when comparing DNS responses:
Resolve-DnsName 'app01.corp.example' -Server '10.40.0.10'A direct IP test can help separate DNS from transport, but HTTPS and integrated authentication may require the proper hostname. Connecting by IP can cause a certificate-name error even when the service is healthy. Do not edit the hosts file as an undocumented permanent fix for a DNS problem. If cache state is relevant, preserve its evidence before clearing a targeted test condition; flushing every resolver cache on production systems is rarely the first useful step.
Inspect route and source-address selection
A server with several adapters, VPNs or static routes can choose an unexpected source address or next hop. Ask Windows which path it selects for an actual resolved destination. Check the most specific prefix and both interface and route metrics. An extra default gateway can produce asymmetric traffic even when the outbound connection initially appears to leave the correct adapter.
Find-NetRoute -RemoteIPAddress '10.90.0.20'
Get-NetRoute -AddressFamily IPv4 | Sort-Object DestinationPrefix,RouteMetric
Get-NetIPInterface -AddressFamily IPv4 | Select-Object InterfaceAlias,InterfaceMetric,ConnectionState
Test-NetConnection -ComputerName '10.90.0.20' -DiagnoseRouting -InformationLevel DetailedConfirm the remote side has a route back to the chosen source network. A local route can send requests successfully while replies follow another gateway. For IPv6 destinations, inspect the IPv6 route table rather than expecting an IPv4 change to matter. Do not add a broad default route to solve one missing subnet path. A narrowly scoped route is appropriate only after the topology owner confirms its next hop and return path.
Test the actual service port
Use Test-NetConnection with the destination name and real TCP port. Detailed output identifies the resolved address, selected source/interface and TCP result. Compare the result with an approved known-good source to determine whether the fault is local, path-specific or shared by the destination.
Test-NetConnection -ComputerName 'app01.corp.example' -Port 443 -InformationLevel Detailed
# Optional supporting probe; ICMP may be blocked:
Test-Connection -ComputerName 'app01.corp.example' -Count 4TcpTestSucceeded True proves a TCP handshake to the selected endpoint, not a valid TLS certificate or a successful login. A False result does not identify which firewall or host rejected the connection. ICMP failure is similarly inconclusive when policy blocks ping. For UDP services such as DNS, RADIUS or IKEv2, a TCP port probe is not an equivalent test; use a protocol-aware request and correlate server logs or a targeted capture.
Check the destination listener and policy
If you administer the destination, verify the expected service is running and listening on the intended address and port. A listener bound only to loopback cannot accept remote clients. Associate its process ID with the application and confirm it belongs to the intended service rather than an unexpected program.
# On the destination server:
Get-NetTCPConnection -State Listen -LocalPort 443 |
Select-Object LocalAddress,LocalPort,OwningProcess
Get-NetFirewallProfile | Select-Object Name,Enabled,DefaultInboundAction,DefaultOutboundAction
Get-NetFirewallRule -PolicyStore ActiveStore -Enabled True |
Where-Object Direction -eq Inbound | Select-Object DisplayName,Action,ProfileReview the specific firewall rule’s port, program, service, source scope and active profile, including centrally applied policy. Avoid assuming that one Allow rule overrides an applicable Block rule. Do not disable every firewall profile as a routine diagnostic. A narrow temporary rule, if appropriate, must have a defined owner, source scope and removal plan; often logs and policy inspection can establish the problem without changing exposure at all.
Correlate intermittent faults with time and workload
For a failure that comes and goes, record repeated samples with timestamps and preserve application, System and network-device evidence from the same interval. Check adapter errors and discards, link flaps, VPN reconnects, DHCP renewals, host resource pressure and upstream maintenance. A successful probe after the incident does not disprove the user’s earlier failure.
Get-NetAdapterStatistics
Get-WinEvent -FilterHashtable @{LogName='System';StartTime=(Get-Date).AddHours(-2)} -MaxEvents 200
1..6 | ForEach-Object {
Get-Date
Test-NetConnection 'app01.corp.example' -Port 443 -InformationLevel Quiet
Start-Sleep -Seconds 5
}Keep sampling bounded and avoid high-volume probes against a fragile service. For packet capture, limit scope and duration and protect the resulting files because they can contain addresses, identities or application data. In a VM, correlate guest observations with hypervisor and virtual-network metrics. A guest cannot directly prove whether a shared physical uplink or host-level policy dropped its packets.
Apply one evidence-based repair and verify recovery
Choose a repair that addresses the observed layer: correct a DNS record, restore the intended route, start the actual listener, fix a scoped firewall rule or resolve an upstream network fault. Make one material change at a time and retain the previous value. Broad reset commands can erase useful state, remove static configuration or disconnect unrelated workloads without explaining the original failure.
Repeat the same source-to-destination application test after the change and confirm previously working paths still work. For intermittent faults, observe through the relevant workload window rather than stopping after one successful probe. Document the root cause, changed setting, verification and rollback. If the evidence points outside the server, provide the network owner with timestamps, source and destination addresses, protocol, port and observed path; this is more actionable than a screenshot saying the Internet is unavailable.
Related guidance and official references
Add a specific static route; Create a scoped firewall rule; Diagnose high CPU usage.