maria-discuss team mailing list archive

[Jeff Dyke <jeff.dyke@xxxxxxxxx>]

Thought it was worth adding, I only added the relevant details, i hope, as
i've been dealing with scaling MariaDB and MySQL, with innodb, for a long
time in both high data throughput and millions of sessions a day, so i've
taken the normal steps to mitigate/investigate this on the server, btw is a
EC2 C5XLarge, with NVMe SSDs.  I'm not calling myself an expert, just
looking for the non low hanging fruit.

On Tue, Aug 25, 2020 at 7:24 PM Jeff Dyke <jeff.dyke@xxxxxxxxx> wrote:

> Unfortunately, I can't easily give you a repro case, or show you the
> queries.  I upgraded a replica/DW server from 10.4.latest to 10.5.5. I did
> this on Friday night.  We're not a high volume site (by design) so nothing
> would really pop up over the weekend.  We use google data studio to present
> reports of aggregated data after the daily events have run.  Any given
> report page will generate from 5-25 connections from datastudio, lots of
> joins, group bys etc.    All of the daily events and reporting and ad-hoc
> querying by our data analyst was working fine on 10.4, but on 10.5.5,
> started generating "Out of sort memory, consider increasing server sort
> buffer size" errors.
>
> That's the background, there were no query changes, no significant
> increase in data.  I changed sort_buffer_size and made a couple other
> optimizations that fit our environment and table types. And i'll have more
> information in the next few days.
>
> All that, leads me to a simple question, that I have not been able to find
> an answer to or missed something in the changelogs that would indicate this
> behaviour would have changed.  Has anyone seen anything similar.  Or does
> this ring a bell to a change I need to tune other than increasing
> sort_buffer_size.  The only thing that changed was an upgrade.
>
> Thanks for reading and giving it a thought.
>