I have a Windows 2012 R2 server setup as a print server.
Background info:
- 2 vCPUs 6GB RAM running in a VMware environment.
- 32 printer queues all using the HP Universal driver. (had 6.2.1, now using 6.2.0 for parity with driver installed on clients)
- Max of 400 concurrent client machines (also VMs) connecting. Supporting 800-1200 users in a VDI environment.
- All 32 queues are deployed via Group Policy.
- Currently 30 queues are set to allow "authenticated users" the ability to print. (another team is working on building the user lists to only allow appropriate offices access to their printers)
This server was running without problems for about two months. Within the past 30 days we have needed to restart the spooler service or the entire server quite often (three times in the last 8 hours).
When the problem happens the spooler will consume 1.9 GB of RAM, but total RAM utilization will be at 98% (Rammap reports ~3GB in the nonpaged pool and ~2GB in process private). Spooler service generally is running 40-80% CPU. Clients will not see printers presented and report extremely long spooling times. If I catch this early enough I can stop the spooler service, wait for the process to drop out of the taskmanager, and start it back up. If it has been hung for a while I need to reboot the server. Stopping the service doesn't do much. watching the memory utilization has the process 'giving back' extremely slowly.
I did some digging and came across a posting about services/drivers having handle leaks. this now has me watching the handle count for the spooler service. When I stopped the service earlier today it had 840,000 handles. That's over 400 times the recommended maximum that I was reading.
I have not been able to find the behavior that causes this memory condition. Any leads?
