Why are RDSHs hanging in a remote desktop farm?

Question

Active Directory

Colleagues, the situation has led me to a dead end. There is a terminal farm based on Win Srv 2012R2 with a connection broker in high availability mode.
The farm consists of:
7 RDSH servers,
2 clustered Hyper-V virtual machines with domain services,
2 of the same virtual machines with connection brokers,
MFC Role "File Server" of two nodes
MFC MS SQL Cluster of two nodes.
Roaming profiles and redirected folders (or rather paths to them) have been added to the DFS namespace and are located on the File Server clustered role with a SCSI disk. Why did I focus on this: the problem is that about once a day (less often than 2) all 7 RDSH servers just hang tightly and stop responding to any input. If the user logs in/logs out while doing this, the user sees the expanded message "User Profile Service Running". This can last long enough until the dullness passes. For those who have an active session, applications that directly or indirectly access the user's files over the network, such as explorer or messenger, which saves received files to the network redirected Downloads folder, fall.
There are no errors in the logs, except for the crashed applications mentioned above, as well as that one or another service did not respond in the allotted time. It would be possible to unambiguously sin on the aforementioned role of the cluster, but when you try to open it from another host on the network, everything opens and is copied / deleted. Yes, and it happens that RDSH servers with such stupid things cannot be accessed using mstsc /admin.
Actually, I appeal to people with rich experience in deploying such farms. The logs are clean, all services are working, but for a week I see the same picture at about the same time (at night, in the range of 1 to 3 hours, there are no backups at this time). Maybe somehow the role of the file server in the cluster is not correctly deployed? Maybe you can't use it at all to store roaming profiles and Folder Redirection? I stupidly had no options left, I would like to google, but I don’t know what. Today, at the moment of such a plug, I transferred the role of the file server cluster to another node of the cluster. And it was suspiciously long (about 30 seconds). Maybe because it's in it, or maybe because there was this stupid thing and I tried to access the RDSH servers with 7 test accounts at the same time. After the transfer of the role, everything seems to have been chewed. But this solution does not suit the problem with empty logs on the cluster nodes. Tell me, what magazines to spy on? Where to hang out?

Answer 1

For those who signed up, and just for history. The reason is found, and I will still think about the elimination.
The regularity of the manifestation (every day between one and 2 am) suggested that this happens according to the scheduler. So:
brought out the one I was interested in:

LastRunTime        : 16.10.2016 1:00:00
LastTaskResult     : 0
NextRunTime        : 17.10.2016 1:00:00
NumberOfMissedRuns : 0
TaskName           : Storage Tiers Optimization
TaskPath           : \Microsoft\Windows\Storage Tiers Management\
PSComputerName     :

In the actions of which it appeared:
Nothing but defragmentation with a high priority. Forcing this task to run demonstrated all of the above problems. Due to the slow disk subsystem on the file server, the user software and user profile services simply worked out I / O operations for an extremely long time, and if they were repeatedly clicked with the mouse, they completely hung up.
The properties of the task say that it is not recommended to change or delete it. I found info on this topic on the flow, I will think what to do with it. For the curious link:
https://technet.microsoft.com/ru-ru/library/dn7891...