tokudb table rows decrease to 0, table has 10m rows in it - no indexes being used on any query plan

[Ross E Carver]

I think this is a bug relating to an earlier bug, or perhaps it was not actually fixed, so I'm re-submitting it as a new/existing bug. The earlier bug is:

It appears there is a google group discussion about this, but no official open bug here.

https://groups.google.com/forum/#!topic/percona-discussion/hjmx3snYX4A

Seems like a pretty critical problem since its causing huge lock contention on updates and query plans result in full table scans. Below is the behavior I'm seeing on a table which has a high number of updates. This is on: Server version: 5.7.16-10-log Percona Server (GPL), Release 10, Revision a0c7d0d

mysql> select TABLE_ROWS from information_schema.tables where table_name='lw_item_discovery';
+------------+
| TABLE_ROWS |
+------------+
| 0 |
+------------+
1 row in set (0.01 sec)

mysql> select count(*) from retail.lw_item_discovery;
+----------+
| count(*) |
+----------+
| 10174538 |
+----------+
1 row in set (3.33 sec)

mysql> set session tokudb_analyze_mode=TOKUDB_ANALYZE_RECOUNT_ROWS;
Query OK, 0 rows affected (0.00 sec)

mysql> analyze table retail.lw_item_discovery;
+--------------------------+---------+----------+----------+
| Table | Op | Msg_type | Msg_text |
+--------------------------+---------+----------+----------+
| retail.lw_item_discovery | analyze | status | OK |
+--------------------------+---------+----------+----------+
1 row in set (0.01 sec)

mysql> select TABLE_ROWS from information_schema.tables where table_name='lw_item_discovery';
+------------+
| TABLE_ROWS |
+------------+
| 0 |
+------------+
1 row in set (0.00 sec)

mysql> set session tokudb_analyze_mode=TOKUDB_ANALYZE_RECOUNT_ROWS;
Query OK, 0 rows affected (0.00 sec)

mysql> analyze table retail.lw_item_discovery;
+--------------------------+---------+----------+------------------+
| Table | Op | Msg_type | Msg_text |
+--------------------------+---------+----------+------------------+
| retail.lw_item_discovery | analyze | status | Operation failed |
+--------------------------+---------+----------+------------------+
1 row in set (0.01 sec)

[George Ormond Lorch III (gl-az)]

https://tokutek.atlassian.net/browse/FT-732

[Sly Ripper (sly-ripper)]

This still isn't fixed in 5.6.35-80.0-1.jessie, it's happening with 'updates' and 'on duplicate key updates'.

A table with 3,020,666 rows reported 2,799,582 yesterday and 2,662,729 today.

TOKUDB_ANALYZE_RECOUNT_ROWS is a temporary fix.

[Phil Sweeney (3-launchpa9-9)]

I'm also still seeing this in MariaDB 10.1.22 (which includes TokuDB 5.6.35-80.0)

[jocelyn fournier (joce)]

I confirm I have the problem with 5.6.35-80.0 (perhaps a regression, I don't remind having issue with earlier versions including FT-732 fix)

[Phil Sweeney (3-launchpa9-9)]

For those still having issues, please be aware some further row count fixes are due soon via https://jira.percona.com/browse/TDB-2

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-398