Fix it now
Event ID 1146 records that the cluster killed and restarted the Resource Hosting Subsystem because a resource stopped responding. Event ID 1230 names the resource, its type and its DLL. Fix what that resource depends on, then isolate it so it cannot take its neighbours down with it.
Get-ClusterResource | Where-Object State -ne 'Online'
Get-ClusterLog -Destination C:\temp\ -TimeSpan 5 -UseLocalTime
- Find the Event ID 1230 next to the 1146 in the System log. It prints the resource name, the resource type and the resource DLL, which is the whole answer to which component hung.
- Look at what that resource sits on. Microsoft’s own advice in the event text is to verify the underlying infrastructure – storage, networking or services – associated with the resource.
- Isolate it so a repeat does not take unrelated resources with it: in Failover Cluster Manager, open the resource’s properties, go to Advanced Policies and select the option to run it in a separate resource monitor.
- Take the role offline and online again so the new monitor takes effect.
- On the same tab, check the pending timeout against how long the resource genuinely needs to come online.
Event 1146 is recovery, not the fault. The cluster decided a hung resource was worse than a restarted hosting process and acted; the thing to fix is whatever the resource was waiting on.
If the role has been stable through a full day and a backup window, you are done. If it keeps happening, the next section explains what the hosting subsystem is waiting for.
Why it happens
Each cluster resource type is implemented by a resource DLL that the cluster loads into a hosting process and calls into: bring this online, take it offline, are you still alive, handle this control code. Those calls are expected to return within a time limit. When one does not, the cluster concludes the process is stuck.
Event ID 1230 is the cluster saying exactly that: a component on the server did not respond in a timely fashion, which caused the named cluster resource – with its resource type and DLL printed alongside – to exceed its time-out threshold. As part of health detection, the cluster will try to recover by terminating and restarting the Resource Hosting Subsystem that runs the resource. Event ID 1146 is that termination happening. Read as a pair, they tell you what hung and what the cluster did about it.
The blunt consequence is that every resource sharing that monitor goes down with the process. One slow disk or one badly behaved application agent can take unrelated resources offline at the same moment, which is why the incident often looks far broader than its cause. Running a suspect resource in its own monitor contains the damage; it does not fix the hang.
Storage underneath the resource stalled
You have this one if A physical disk or volume resource is named in the 1230, and storage events appear on the node in the same minute.
- Check the System log around the termination for Storport event 129 (a reset issued to the device) and event 153 (an I/O retried).
- Compare the path count on this node against a healthy one with
mpclaim -s -d. - Fix the storage first. Nothing you change in the cluster helps while the volume is stalling.
A resource DLL or application agent is blocking
You have this one if The same third-party or application resource type is named in the 1230 every time.
- Run that resource in its own monitor so it stops taking neighbours down with it.
- Update the vendor’s cluster agent or resource DLL, using the DLL name printed in the event to identify it precisely.
- Collect the cluster log for the window and give it to the vendor with the event.
The node is not the owner the operation assumed
You have this one if 0x80071397 appears in the same window, usually while the role is already moving.
- Treat it as a state message rather than a fault: Windows error 5015 is ERROR_HOST_NODE_NOT_RESOURCE_OWNER.
- Check whether the node is even a possible owner of that resource. The same error covers a node that is not in the resource’s possible-owners list, which is a configuration problem that will never resolve itself.
- In Failover Cluster Manager, open the resource’s Advanced Policies and confirm the possible owners are the nodes you intended.
A name resource is blocking on the directory
You have this one if A network name resource is named, and the node is slow to reach a domain controller.
- Confirm the node can reach a writeable domain controller and resolve names promptly.
- Check for a site or DNS configuration sending the node to a distant domain controller.
- Confirm the computer object behind the clustered name is healthy, and repair it from Failover Cluster Manager if not.
The resource genuinely needs longer than it is given
You have this one if A large file server or database resource is named, and it does come online successfully when nothing rushes it.
- Raise the pending timeout on that resource’s Advanced Policies tab, rather than changing cluster-wide behaviour.
- Run the resource in its own monitor at the same time.
- Write down the change and the reason, so the next person does not read it as an arbitrary number.
A resource that takes minutes to come online is worth investigating in its own right. Give it the time it needs, then find out why it needs it.
Full reference
Reading the pattern across several incidents
| What you see | Where the fault is |
|---|---|
| The same resource named in every 1230 | That resource, its DLL, or the storage under it |
| Different resources named, always on one node | Something node-wide: storage paths, drivers or a scanning agent |
| It only happens during backups | The backup agent, or the load it puts on storage |
| A third-party resource type is named | The vendor’s resource DLL rather than the cluster |
| A network name resource is named | Directory or name resolution blocking the call |
| Terminations stop when one role is moved off the node | That role’s dependencies, not the cluster |
What the two events actually print
Event 1146 says the Resource Hosting Subsystem process was terminated and will be restarted, that this is typically associated with cluster health detection and recovery of a resource, and that you should refer to the System event log to find which resource and resource DLL caused it. That last sentence is the instruction to go and find the 1230.
Event 1230 carries three fields worth more than the rest of the incident put together: the resource name, the resource type, and the DLL. The DLL is what tells you whether you are looking at an in-box resource type or a vendor’s, and it is the piece of evidence a storage or application vendor will ask for first.
The cluster status codes you will see alongside
| Code | Windows error | Meaning |
|---|---|---|
0x80071397 |
5015 | ERROR_HOST_NODE_NOT_RESOURCE_OWNER: the node is not the owner of the resource, or is not a possible owner of it |
0x80071398 |
5016 | ERROR_HOST_NODE_NOT_GROUP_OWNER: the same thing about the group rather than the resource |
0x8007138F |
5007 | ERROR_CLUSTER_RESOURCE_NOT_FOUND: the cluster resource could not be found |
The distinction between the first two is worth holding on to, because they are one digit apart and describe different objects. If you are chasing a script or a monitoring tool that reports one of them, check which it is before you conclude anything about ownership.
Isolation, and when it is the wrong answer
Running a resource in a separate monitor is a containment measure. It costs a process and some memory, and it buys you the guarantee that when this resource hangs again, it takes nothing else with it. That is worth doing for anything from a third party that has misbehaved before, and for the resource named in a repeat 1230. It is not worth doing for every resource in the cluster: you would be paying for isolation you have no evidence you need, and you would still be no closer to the hang.
Do not raise cluster-wide hang detection timeouts to make these events stop. Doing so leaves genuinely hung resources running longer before anything recovers them, and on some configurations the response to a detected hang can be a deliberate bugcheck for diagnostics. Change either only with a specific reason and a way back.
When it is none of the obvious things
- Collect the cluster log for a narrow window around the event rather than a wide one;
-TimeSpan 5on Get-ClusterLog is usually enough and keeps the file readable. - Check whether an antivirus or backup agent was updated shortly before the first occurrence. Scanning agents that are not cluster aware show up as hangs in resource calls, not as scanner errors.
- Check whether the node was patched or its storage driver changed on the same date.
- Look for Event ID 1069 and 1205 in the same window. They record the resource failure and the role failing to come completely online, and together with 1230 they show whether recovery actually completed.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
Event ID 1146 |
The Resource Hosting Subsystem process was terminated and will be restarted; the System log names which resource and resource DLL caused it | Microsoft Learn |
Event ID 1230 |
A component did not respond in time, so the named resource exceeded its time-out threshold and the cluster will recover by terminating and restarting RHS | Microsoft Learn |
0x80071397 |
Windows error 5015, ERROR_HOST_NODE_NOT_RESOURCE_OWNER: the node is not the owner of the resource, or is not a possible owner of it | Microsoft Learn |
Confirm the fix worked
- The role comes online and stays online through a full working day.
- The suspect resource shows a separate resource monitor on its Advanced Policies tab.
- No new 1146 or 1230 appears through a backup window and a scheduled maintenance window.
- Moving the role to another node and back completes cleanly in both directions.
Get-ClusterResourceshows every resource in the role Online, with no 1069 or 1205 alongside.
Questions people ask about this
Does terminating RHS lose data?
Not directly. Resources hosted in that monitor go offline abruptly and the role fails over, so applications experience it as an unclean stop, exactly as they would on a node failure. Anything not committed to disk at that instant is lost the same way it would be then.
Should every resource run in its own monitor?
No. Each monitor is a process, and isolating everything costs memory while buying nothing for well-behaved resources. Isolate the resource the 1230 keeps naming, and anything from a third party with a history.
Why does the role sometimes come back without anyone touching it?
Because that is the designed recovery: the hosting process is restarted and the role is brought online again, on the same node or another. It is why these incidents often pass unnoticed outside the event log.
Is this a licensing or edition problem?
No. Failover Clustering behaves identically across Windows Server editions, and a resource call that does not return is a storage, application or code problem on a server you already licence.
The 1230 names a resource DLL I do not recognise. Now what?
Search for that file name on the node and check which product installed it. The DLL is the most precise identifier in the whole incident, and it is the detail a vendor’s support team will want before they will look at anything else.
