|
The relevant code is:
1695 if(master_dcb == NULL || master_dcb->session == NULL)
|
1696 {
|
1697 skygw_log_write(LE,"[%s] Error: Master server DBC is NULL. "
|
1698 "This means that the connection to the master server is already "
|
1699 "closed while a query is still being routed.",__FUNCTION__);
|
1700 return qtype;
|
1701 }
|
1702 CHK_DCB(master_dcb);
|
1703
|
1704 data = (MYSQL_session*)master_dcb->session->data;
|
1705
|
1706 if(data == NULL)
|
1707 {
|
1708 skygw_log_write(LE,"[%s] Error: User data in master server DBC is NULL.",__FUNCTION__);
|
1709 return qtype;
|
1710 }
|
The crash occurs on line 1704.
Since it was checked on line 1695 that master_dcb and master_dcb->session are non-NULL, there are basically two options:
- Either master_dcb->session points at unallocated memory, or
- master_dcb->session has been set to NULL, by some other thread, after the check on line 1695.
The former would require that not only has the session object pointed to by master_dcb->session been freed, without master_dcb->session having been set to NULL, but also that the memory would have been reclaimed by the operating system. That seems unlikely.
So, the likely reason is that master_dcb->session has been set to NULL by another thread after the check was made on line 1695, but before master_dcb->session->data is accessed on line 1704.
This is the kind of issue that the fix of MXS-329 is expected to deal with.
|
michael_error1 (1).log