Comment 10 for bug 1177206

Revision history for this message
Vlad Lesin (vlad-lesin) wrote :

So the crash on "prepare" step happens due to bad generated data on "backup" step. That is why I tried to make the test which would generate such data that would lead to the bug on "prepare" step. I tried the following things:

1) Test with randgen - just launch randgen with percona_qa rules and do incremental backup. The test is attached to this post;

2) buf_flush_list() was modified such a way that it just returns if certain global variable is set, the variable was set just after fil_create_new_single_table_tablespace() was invoked for the newly created compressed table to avoid pages flushing after the tablespace creating, in this case page's recv list contained MLOG_PAGE_CREATE command which invokes page_create_low() and set PAGE_N_HEAP field so the bug was not reproduced.

It seems log file begins to be tracked from wrong start_lsn, but I don't have the ideas how it can be reproduced.