Querying INFORMATION_SCHEMA.TABLES memory usage

[colin]

We are encountering an issue when running 5.6.26-74.0 where the memory usage of mysql grows until the process is reaped by the OOM killer.
We have a large number of INNODB tables (~90K) spread over ~20 databases and use file per table.
We noticed that the memory issue increases correspond with executions of the following query:

SELECT table_schema "Database name", sum(table_rows) "No. of rows", sum(data_length) /1024/1024 "Size data (MB)",sum(index_length)/1024/1024 "Size index (MB)" FROM information_schema.TABLES GROUP BY table_schema

The memory increase from this query doesn't seem to be freed and it increases each time the query is run.
This query is run by our monitoring agent periodically to capture usage stats.

I can reproduce this locally on OSX using the homebrewed perconaserver install (modified to use the correct version).
I also tried version 5.6.30-76.3 and the growth seems to be managed (there is still high memory usage but it reaches a peak and gets reclaimed after the above query is executed).

I noticed a similar issue was reported in https://bugs.launchpad.net/percona-server/+bug/1380895 and is said to be fixed in 5.6.26 but it didn't look like there was any regression test added so maybe this reappeared (the code looks to have changed quite a bit between that fix commit and 5.6.26-74)?

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-3470