On Wed, 16 Oct 2019, 11:07 Reindl Harald, <h.reindl@thelounge.net> wrote:


Am 16.10.19 um 11:59 schrieb Gordan Bobic:
> I have seen failures when upgrading from 10.0, with the latest 10.1 ->
> 10.2 -> 10.3 -> 10.4 unless I issued a clean shutdown between the latest
> 10.1 and 10.2 as recently as last week, with the latest RPMs for each
> version.
>
> You are trying to make an argument analogous to "smoking can't be
> harmful because I've been smoking for 50 years and I'm not dead yet".
>
> What is your sample size?

* 10 years
* 12 setups with innodb
* MySQL 5.0 -> 5.1 -> 5.5
* MariaDB 5.5 -> 10.0 -> 10.1 -> 10.2 -> 10.3

Mo Sep 02 2019: 10.2.26 -> 10.3.17

I did 6 such upgrades for customers in the last month alone. Granted, I only re-tested for this issue on one of those 6 because it was on a setup easy to snapshot and rollback, I do such things most months (only when upgrading from <= 10.1 to => 10.2, obviously) since I first got bitten by it a few years ago. And I have yet to see it work without a clean shutdown.

So clearly there must be at least an element of luck / write load in play. And I don't put much faith in luck.


and frankly i expect whatever software to be able to read it's old data

Famous last words...