QA: This is tricky to reproduce. The secret is to:
1) execute select * from information_schema.columnstore_extents; // This sets the reference to shared memory in mysqld
2) restart columnstore "systemctl stop mariadb-columnstore", "systemctl start mariadb-columnstore" // This causes the shared memory reference to be invalidated
3) do something to change the extentmap // This will only be reflected in Other columnstore processes, not mysqld, which has an invalid reference
4) execute select * from information_schema.columnstore_extents; // Access the invalid reference
This last select will not reflect any changes made or may hang or crash.

Here's a concrete example:

Start with no columnstore tables and the system running.

MariaDB [mysql]> select * from  information_schema.columnstore_extents;

Empty set (0.094 sec)

(In another console)

systemctl stop mariadb-columnstore

systemctl start mariadb-columnstore

Run test000.sh to fill up the extent map

MariaDB [mysql]> select * from  information_schema.columnstore_extents;

Empty set (0.094 sec)

(In another console)

systemctl stop mariadb

systemctl start mariadb

Will cause a refresh of the shared memory reference

select * from  information_schema.columnstore_extents;

.

.

.

984 rows in set (0.150 sec)

Build verified: 5.4.1-1 (Drone #870)

Reproduced the issue in 1.5.3-1 and verified the fixed in 5.4.1-1.

For additional data extents, cpimported 10g lineitem source data.