Scaling issues with remote blockstore implementation
We have a user who is setting up 40 nodes each with their own clone of an iSCSI blockstore. They had trouble with some of the clients coming up without their blockstore mounted due to a timeout in iscsiadm
as it tried to login to the iSCSI session (issue #513 (closed)).
In debugging this, I found an even worse problem where trying to instantiate a new instance of a 40 blockstore experiment shortly after terminating a previous instance, let to failures on the server due to VLAN inconsistencies (issue #515 (closed)).
While these are two separate bugs, they stem from the same basic problem: the blockstore server "faux vnode" setup is serialized and not terribly fast.
Edited by Mike Hibler