segfault in "handle_delayed_insert"

[Michael Widenius]

>>>>> "elble" == elble  <elble@stripped> writes:

elble> only a problem when running with --debug turned on, and using
elble> 'insert delayed'...

elble> _my_thread_var() was returning null, and
elble> DBUG_ENTER() dosen't like that.

elble> here's a patch against 3.23.4-alpha:

elble> *** sql/sql_insert.cc   Tue Sep  7 19:00:45 1999
elble> --- sql/sql_insert.cc   Mon Oct  4 14:56:02 1999
elble> ***************
elble> *** 760,766 ****
elble>   {
elble>     delayed_insert *di=(delayed_insert*) arg;
elble>     THD *thd= &di->thd;
elble> -   DBUG_ENTER("handle_delayed_insert");
  
elble>     pthread_detach_this_thread();
elble>     /* Add thread to THD list so that's it's visible in 'show processlist' */
elble> --- 760,765 ----
elble> ***************
elble> *** 779,784 ****
elble> --- 778,784 ----
elble>     }
elble>   #endif
  
elble> +   DBUG_ENTER("handle_delayed_insert");
elble>     if (init_thr_lock() ||
elble>         my_pthread_setspecific_ptr(THR_THD,  thd) ||
elble>         my_pthread_setspecific_ptr(THR_NET,  &thd->net))


Fixed. Thanks!

elble> i'm trying to track down a lockup when using 'insert delayed' a lot.

elble> i have a chunk of code that inserts rows constantly to a table.
elble> every ten minutes, it forks off a child to do selects from that
elble> table. the parent returns to continue to insert rows, but into
elble> a new table. when the child is done processing, it drops the
elble> 'old' table.

elble> but the drop never completes, the server locks up pretty tightly.
elble> you can establish a connection, but only if you're on 'localhost'.
elble> you can run one command (e.g. show processlist), but then you can't
elble> run any more unless you reconnect. when you reconnect, the old
elble> 'show processlist' is in state 'cleaning up' or somesuch...

elble> ring any bells?

Hi!

DROP TABLE is rather special as it has to ensure that there isn't any
insert delayed threads running on the table, so this could be a source 
to problems,

Could you try to modify the test program 'fork_test.pl' to reproduce
this problem?

Regards,
Monty