When a self-referencing spider table creation requires connection to send queries to the data node, it could cause a hang on an MDL deadlock, until lock wait timeout lapses. Here are two examples:

CREATE TABLE t ENGINE=Spider COMMENT='WRAPPER "mysql",srv "srv",TABLE "t"' AS SELECT 1; 

CREATE TABLE t ENGINE=Spider COMMENT='WRAPPER "mysql",srv "srv",TABLE "t"'; 

Note that the second example requires table discovery which results in spider connecting to the data node to query for the table structure.

In both examples, a set global lock_wait_timeout= 1; reveals that the hang is indeed waiting for the lock wait timeout.

The spider self-referencing detection is not called because the deadlock happens before call to any spider handler methods. To see this, place a set global lock_wait_timeout= 1; before either example. Run mtr --rr on the testcase Place a breakpoint at my_error and my_message on the condition that the error number is 1205 ER_LOCK_WAIT_TIMEOUT. Continue till the breakpoint is reached. Place another break point at mysql_parse, as well as rbreak ha_spider::.. Then do reverse-continue. The mysql_parse breakpoint is reached.

One solution is to temporarily set lock_wait_timeout to a small value when doing the "pre-query" in the first example, and table discovery query in the second, and suggest checking self-referencing in the error message, just like MDEV-29676.