I removed unnecessary tables, but that's about all simplification I was able to perform. The data and aria logs are still big.
3 archive files are uploaded to FTP (in addition to what Dreas provided before):
982872_datadir.tar.gz - already created and populated table and offending aria logs placed into the datadir. To reproduce the problem, decompress the archive and start MariaDB server on the extracted 'data' as a datadir (or run aria_read_log -a inside the datadir).
982872_aria_logs.tar.gz - separately archived aria logs.
982872_sql.tar.gz - the dump of the table (structure and data).
To reproduce the problem using the dump, create a new empty datadir, start MariaDB server, load the dump, shutdown the server, place aria logs into the resulting datadir, then either start the server or run aria_read_log -a.
On release server, it crashes as described above.
On 5.2 debug server it aborts with the failing assertion:
#6 0x00007fe752c3bd4d in __GI___assert_fail (assertion=0xdb8d72 "!new_page", file=<optimized out>, line=6355,
function=<optimized out>) at assert.c:81
#7 0x0000000000a0f72a in _ma_apply_redo_insert_row_head_or_tail (info=0x25f0708, lsn=42966603572, page_type=1,
new_page=1 '\001', header=0x25e1cda "\236", <incomplete sequence \372>, data=0x25e1ce0 "", data_length=34)
at ma_blockrec.c:6355
#8 0x00000000009c9ba8 in exec_REDO_LOGREC_REDO_INSERT_ROW_HEAD (rec=0x7fff100855f0) at ma_recovery.c:1537
#9 0x00000000009c7619 in display_and_apply_record (log_desc=0x1a662b8, rec=0x7fff100855f0) at ma_recovery.c:587
#10 0x00000000009ccd30 in run_redo_phase (lsn=42966567214, lsn_end=0, apply=MARIA_LOG_APPLY) at ma_recovery.c:2695
#11 0x00000000009c6ec6 in maria_apply_log (from_lsn=42966567214, end_lsn=0, apply=MARIA_LOG_APPLY, trace_file=0x25e1a00,
should_run_undo_phase=1 '\001', skip_DDLs_arg=1 '\001', take_checkpoints=1 '\001', warnings_count=0x7fff10085b08)
at ma_recovery.c:349
#12 0x00000000009c6b44 in maria_recovery_from_log () at ma_recovery.c:239
#13 0x0000000000956b2f in ha_maria_init (p=0x25c7718) at ha_maria.cc:3366
#14 0x00000000007d2209 in ha_initialize_handlerton (plugin=0x25c6cf0) at handler.cc:432
#15 0x00000000008a361a in plugin_initialize (plugin=0x25c6cf0) at sql_plugin.cc:1263
#16 0x00000000008a3c88 in plugin_init (argc=0x1246a78, argv=0x25988e0, flags=0) at sql_plugin.cc:1451
#17 0x0000000000680146 in init_server_components () at mysqld.cc:4089
#18 0x0000000000680cb9 in main (argc=8, argv=0x7fff100863f8) at mysqld.cc:4560
aria_read_log -a (debug version) also aborts with an assertion:
Reproducible on maria-5.2, 5.3, 5.5.
I removed unnecessary tables, but that's about all simplification I was able to perform. The data and aria logs are still big.
3 archive files are uploaded to FTP (in addition to what Dreas provided before):
982872_ datadir. tar.gz - already created and populated table and offending aria logs placed into the datadir. To reproduce the problem, decompress the archive and start MariaDB server on the extracted 'data' as a datadir (or run aria_read_log -a inside the datadir).
982872_ aria_logs. tar.gz - separately archived aria logs.
982872_sql.tar.gz - the dump of the table (structure and data).
To reproduce the problem using the dump, create a new empty datadir, start MariaDB server, load the dump, shutdown the server, place aria logs into the resulting datadir, then either start the server or run aria_read_log -a.
On release server, it crashes as described above.
On 5.2 debug server it aborts with the failing assertion:
[Note] mysqld: Aria engine: starting recovery redo_insert_ row_head_ or_tail: Assertion `!new_page' failed.
recovered pages: 0%mysqld: ma_blockrec.c:6355: _ma_apply_
#6 0x00007fe752c3bd4d in __GI___assert_fail (assertion=0xdb8d72 "!new_page", file=<optimized out>, line=6355, <optimized out>) at assert.c:81 redo_insert_ row_head_ or_tail (info=0x25f0708, lsn=42966603572, page_type=1, LOGREC_ REDO_INSERT_ ROW_HEAD (rec=0x7fff1008 55f0) at ma_recovery.c:1537 and_apply_ record (log_desc= 0x1a662b8, rec=0x7fff100855f0) at ma_recovery.c:587 LOG_APPLY) at ma_recovery.c:2695 42966567214, end_lsn=0, apply=MARIA_ LOG_APPLY, trace_file= 0x25e1a00, run_undo_ phase=1 '\001', skip_DDLs_arg=1 '\001', take_checkpoints=1 '\001', warnings_ count=0x7fff100 85b08) from_log () at ma_recovery.c:239 handlerton (plugin=0x25c6cf0) at handler.cc:432 components () at mysqld.cc:4089 63f8) at mysqld.cc:4560
function=
#7 0x0000000000a0f72a in _ma_apply_
new_page=1 '\001', header=0x25e1cda "\236", <incomplete sequence \372>, data=0x25e1ce0 "", data_length=34)
at ma_blockrec.c:6355
#8 0x00000000009c9ba8 in exec_REDO_
#9 0x00000000009c7619 in display_
#10 0x00000000009ccd30 in run_redo_phase (lsn=42966567214, lsn_end=0, apply=MARIA_
#11 0x00000000009c6ec6 in maria_apply_log (from_lsn=
should_
at ma_recovery.c:349
#12 0x00000000009c6b44 in maria_recovery_
#13 0x0000000000956b2f in ha_maria_init (p=0x25c7718) at ha_maria.cc:3366
#14 0x00000000007d2209 in ha_initialize_
#15 0x00000000008a361a in plugin_initialize (plugin=0x25c6cf0) at sql_plugin.cc:1263
#16 0x00000000008a3c88 in plugin_init (argc=0x1246a78, argv=0x25988e0, flags=0) at sql_plugin.cc:1451
#17 0x0000000000680146 in init_server_
#18 0x0000000000680cb9 in main (argc=8, argv=0x7fff1008
aria_read_log -a (debug version) also aborts with an assertion:
aria_read_log: ma_key_ recover. c:988: _ma_apply_ redo_index: Assertion `page_offset >= keypage_header && page_offset <= page_length' failed.
A release version of aria_read_log -a crashes with segmentation fault.