Is this issue reproducible internally as well Or only happens for external connections?
Are you using vGPU based pools? Does it happen with all desktop pools Or specifically with windows 10 1909 pool only?
Are you able to reproduce the issue with PCoIP as well Or does it happen only with Blast?
How many vCPU you have assigned to VMs? Please note 2 or more CPU is recommended for blast protocol.
flags:0x0 indicates that the agent should wait for the client to attempt a reconnect rather than terminate the session.
To further troubleshoot first isolate the issue as per above suggestions and provide blast-worker and blast-service both log files from the problematic VM.