Hi Bekir,

I'm very happy to answer your questions. See me comments in line.

On Sat, Jan 23, 2021 at 8:24 AM Bekir Fajkovic <bekir.fajkovic@citynetwork.eu> wrote:

The questions:
------------------------

Having the fact that Trove Victoria release provides Docker containers as a new way of database instance provisioning, i am wondering how far the project
is developed in terms of covering the different types of databases. What i can see by mainly parsing the code provided on Github, those seem to be
officially released:

- MySQL
- MariaDB
- PostgreSQL

and the rest of the planned database types are in "experimental" phase. And also, regarding certain types of databases (for example MySQL, version 5.7 and 8.0) only certain
versions of the datastores seems to be supported, but not all.

MySQL 5.7.x is supported but 8.0 is in experimental.

On the other hand, nothing regarding datastore versions supported for MariaDB and PostgreSQL seems to be
mentioned somewhere. Could someone please confirm that as well as give some more details about it?

MariaDB 10.4.x and PostgreSQL 12.4 are supported. The other versions need to be fully tested.

I successfully managed to create certain versions of datastores in my devstack environment, belonging to those 3 database types mentioned above (and based on
trovestack-generated dev guest image that is by default delivered with devstack installation), but not without some undesirable events. For example, i am able to register
PostgreSQL datastore version 12 and instantiate a database instance of that version but not version 13 and above, where i get some hostname-related errors etc.

Yes, because PostgreSQL 13 has never been tested.

Also, a question regarding the building of the production-ready guest image. As mentioned, Trovestack script is provided as a possible way of producing the images (by omitting
dev option the Trove Guest Agent binaries are deployed into the instantiated VM). How does an image produced this way looks like? From where the base image is fetched,
is it a "cloud based image" with cloud-init in it, are the automatic security and software patching features disabled in produced image, so that we do not get unexpected service
interruptions when the OS suddenly decides to start updating itself etc..

If you look at trovestack script implementation, you can see it's calling disk-image-create script from diskimage-builder, and there are some elements[2] defined in trove repo[3] for building the image.

[1]: https://docs.openstack.org/diskimage-builder/latest

[2]: https://docs.openstack.org/diskimage-builder/latest/elements.html

[3]: https://github.com/openstack/trove/tree/master/integration/scripts/files/elements

Regarding the Trove Guest Agent service - i read in some Trove books previously that there are dedicated agents for each and every database type, is it the same situation
in Victoria release, or is there an "universal" Guest Agent covering all the database types nowadays? Where is the code that adapts the Agent commands towards the database
instances placed inside the project?

Trove never has dedicated agents for each and every database type, it's using the same trove-guestagent but with different configurations for different datastores.

The backups - as i can see there seem to be some kind of dedicated docker-backup images involved in each database type. Could someone explain the internals of backup mechanisms
inside Trove Victoria release in more details?

The backup container image is created to help trove-guestagent to achieve datastore-agnostic (so we only need to maintain a universal guest image), we shift the backup and restore functionalities and needed tools to a dedicated container. The implementation can be found here[4].

[4]: https://github.com/openstack/trove/tree/master/backup

---

Lingxian Kong

Senior Software Engineer

Catalyst Cloud