Crash with backtrace like:

mongoc_stream_check_closed ()
_mongoc_cluster_sendv () 
_mongoc_cursor_next ()
mongoc_cursor_next ()

mongoc_stream_check_closed only checks if "stream" is NULL in debug mode; a release-mode driver will crash if the stream passed to function is NULL.

Likely cause:

• Start a cursor, complete the initial query on a primary, the cursor has a "hint", the index of its server in driver's array of servers
• Shut down the primary
• Do another operation on the cluster, forcing the driver to reconnect and change cluster state (fewer servers, different servers, a server with a cleared stream?)
• Iterate the cursor to completion - it tries to send a getmore using its "hint", which may be out of bounds now, or point to a different server?, or point to a server with a NULL stream, not sure which of these three is possible
• Segfault

There seem likely a number of ways the cursor's "hint" can be invalidated. Avoid crashes in all of them for the 1.1.x code.

Good news - this is all completely re-architected and corrected in 1.2, but 1.2. won't be released until EOQ at the earliest.