Troubleshooting
PI-to-PI interface service will not start from FactoryTalk Historian
Short answer
When the PI-to-PI interface service will not start, the fault sits in the local launch, access to the two historians, or the FactoryTalk Historian license, which must permit native OSIsoft clients. Capture the exact startup error, work the three gates in fixed order, then prove the destination with values, timestamps and quality.
On this page
- Capture the failure exactly as it happens
- Three gates decide the fix
- What not to change while the service is down
- Gate one: prove Windows can launch the service
- Gate two: prove the interface reaches both servers
- Gate three: prove the source license permits the client
- Worked triage: interface configured, service will not start
- Prove the destination with values, timestamps and quality
- Questions asked
When the PI-to-PI interface service will not start after configuration, the fault sits in one of three places — the local service launch, access to the two historians, or the FactoryTalk Historian license — and not in the tag mapping. Capture the exact startup error before changing anything, work the three gates in fixed order, and fix the first one that fails. A running service is not acceptance: prove the transfer to the receiving PI Server with values, timestamps and quality.
Capture the failure exactly as it happens #
The PI Interface Configuration Utility (PI ICU) reports the problem as a "Failed to start service" message carrying the service name, and that message covers two different faults: Windows never launched the executable, or the interface launched, could not reach a server, and exited. The Windows Services console shows both as a stopped service. Record four things before changing anything:
- The service name and the exact Windows message at the moment of the failed start.
- Every interface message between the start request and the process exit, with timestamps.
- Whether PI ICU itself faults. A control that was registered on the machine and whose file was later moved or deleted produces its own startup errors; restore the ICU installation so the registered control and the physical file match.
- The start time, so the Windows and interface logs can be aligned.
Read the earliest relevant entry, not the final service-stopped line: the last message is the consequence of the first.
| What you observe | Where the fault sits | First check |
|---|---|---|
| Windows rejects the start immediately | Service registration, executable access or service-account logon | Service properties and Windows event detail |
| The interface launches, then exits | Startup configuration or connection initialization | Earliest interface message before exit |
| The service stays up but nothing updates | Reads, writes or point selection | Matched source and destination points |
Three gates decide the fix #
A configured interface proves the configuration work was done, nothing more. Between a configured interface and data moving from FactoryTalk Historian to the receiving PI Server stand three gates:
- Local launch. Windows must start the installed interface executable under its configured service account.
- Server access. The interface must authenticate to both historians: read on the source, write on the destination.
- Source license. The FactoryTalk Historian license must allow connections from native OSIsoft clients.
The connection arrangement uses a PI trust, which identifies the incoming interface connection and maps it to server-side access. A trust authenticates; it does not entitle. Fix the first failing gate and leave the gates that pass alone.
What not to change while the service is down #
- Recreating destination tags. Mapping matters once the service runs and reaches both servers; it cannot repair a service that never launches.
- Changing buffering settings. The PI Buffer Subsystem is a delivery function across interruptions, not the connection method between two historians.
- Restarting in a loop. Collect the failure once, apply one fix, restart once, and compare; every blind restart buries the timeline.
- Widening trusts or permissions. Broadening a trust to force a connection on an OT network is the one change that adds risk while fixing nothing — correct the connection identity instead.
Gate one: prove Windows can launch the service #
Open the service properties and confirm the executable exists at its configured path on that machine, that the startup arguments reference the configuration you built in PI ICU, and that the service account can log on: an account logon failure stops the start before any interface code runs. For faults that appear after a .NET-level problem, restarting the PI Buffer Subsystem service and the PI Interface Service from the Windows Services console is the documented first move. Resolve both server names from the interface host, not from your engineering workstation, and keep the Windows service identity separate from the identity the PI servers see — an interactive test under your login can succeed while the service fails under its own account.
Gate two: prove the interface reaches both servers #
The interface does not start unless it can connect to the PI Data Archive, so test that path first, with the apisnap utility, run from the interface host. Then check both ends separately:
- Source. Compare the incoming connection's actual identity with the trust definition and confirm the read access that trust carries; a trust entry that exists is not a trust that matches.
- Destination. Confirm write access on its own. Success reading FactoryTalk Historian grants nothing on the receiving server.
- Startup timing. A failed-to-start message during startup is often timing: the interface starts and begins its history recovery while the service controller has already reported failure, so let recovery finish and re-check the service before treating the message as a fault.
When other clients fail against the same server at the same time, the fault is the server or the network path rather than this interface, and the same evidence discipline applies when troubleshooting historian trend client connection failures.
Gate three: prove the source license permits the client #
Read the entitlement from the properties of the Historian SE Server in the FactoryTalk Administration console and confirm it covers connections from native OSIsoft clients. If the wording leaves doubt, have Rockwell Automation or an authorized distributor verify it against the installed license before more configuration work; no trust setting substitutes for this entitlement.
Worked triage: interface configured, service will not start #
The inputs, each from the plant floor or the product documentation:
- Symptom as reported: the interface configuration is finished in PI ICU, and starting the service fails.
- Documented behavior: the interface does not start unless it can connect to PI Data Archive.
- Documented requirements: a PI trust for the interface connection, and a FactoryTalk Historian license permitting native OSIsoft clients.
- The three gates, checked in fixed order.
The rule in words: count the interface's own messages between the Windows start request and the process exit; that count places the failure at exactly one gate.
The arithmetic, written out:
- Interface messages before exit = 0 → no interface code ran → gate one. Fix the executable path, the startup arguments or the service-account logon.
- Interface messages ≥ 1 and the earliest is a failure to connect to the PI Data Archive → the process ran and could not connect → gate two. Confirm the path with apisnap, then fix reachability or the trust.
- Interface messages ≥ 1, connections reported open and no data moving → gate three. Read the license entitlement in the FactoryTalk Administration console before touching any trust.
- Service running, no failure messages and the destination stale → no gate failed; move to proving delivery.
The threshold is the zero-message case. Zero interface messages before exit keeps you on gate one, and no trust, tag or license change can fix an executable the service controller never launched. Work one gate per restart, and restart once per fix.
Prove the destination with values, timestamps and quality #
A started service is the first acceptance check, not the last. Give the link a health point first: in PI ICU, on the PItoPI > Debug tab, check "Interface Status Tag on Receiving PI Server" and specify a tag name — the interface then maintains its own status point on the destination, separate from your data. Then:
- Select a few source points with identifiable recent changes and record their values, timestamps and quality.
- Read the matched destination points and compare value, event timestamp and quality, not the latest displayed number.
- If reads succeed and updates are missing, review the destination write diagnostics.
- To compare both sides over one window, export historian trend CSV at one-minute intervals from source and destination and align the timestamps.
Once the link is healthy, recovering the history the outage missed is a separate task, and the same evidence discipline carries into backfilling PI from an existing plant historian.
Questions asked
Does the FactoryTalk Historian side need a license for this transfer?
Yes. The FactoryTalk Historian license must allow connections from native OSIsoft clients, and this is the gate teams skip most often because everything else looks configured. A PI trust authenticates the connection but grants no entitlement; if the console wording leaves doubt, have Rockwell Automation or an authorized distributor read the installed license.
Why does the interface stop when the PI Data Archive is unreachable?
It is designed that way: the interface does not start unless it can connect to the PI Data Archive, so an unreachable archive looks exactly like a broken service. Run the apisnap utility from the interface host to test the path before reinstalling anything; once connectivity returns, the same service should start without configuration changes.
The service reported a start failure but data appeared anyway. What happened?
A timing overlap. The interface can start and begin its history recovery while the service controller has already reported the failure, so the message can precede a working transfer. Let recovery finish, re-check the service state, then prove delivery with values, timestamps and quality before treating the message as a fault.
How do I watch the interface health on the receiving PI Server?
In PI ICU, open the PItoPI > Debug tab, check Interface Status Tag on Receiving PI Server and specify a tag name. The interface then maintains a status point on the destination that you can trend next to your data, which answers whether the link is alive separately from whether the data is fresh.