[haiku-depot-web] Re: Test deployment [was: Re: Syndication with Web Site Users?]

  • From: Andrew Lindesay <apl@xxxxxxxxxxxxxx>
  • To: haiku-depot-web@xxxxxxxxxxxxx
  • Date: Mon, 20 Jan 2014 22:43:23 +1300

Hi Oliver;

- - -

I've made RPMs with a maven plugin before, but not using the regular RPM tooling. I could take a look at it, but it may take a little while.

- - -

I have tended to use Postgres (PG) for my professional work for a long time now. I find it very good although I would agree that the dump-restore is an issue for larger upgrades on major versions.

Aside from personal preference, I guess the only two concrete technical reasons why I would like to stick with PG rather than use a modern MySQL database;

1) PG supports deferred referential integrity constraint checking which MySQL still does not. The ORM that I am using has an in-memory transactional model wherein it is not intended that the application controls the sequence of the database statements executed. This fact means that I'd either have to code knowing this limitation or would need to disable referential constraints -- shudder!

2) Another big plus for PG is that the automated schema migration system I am using works really nicely with Postgres' transactional DDL in that you cannot get partially applied migrations.

The application will use a connection pooling system inside the application server to open multiple sockets to Postgres and I am not aware of a problem with this in relation to a standard Postgres deployment and its threading model.

If it is not too much inconvenience, I would rather continue to use Postgres, but I understand your position and would rather do the work now to make things easier for you and others going forward to fit in if it is a problem. Let me know if you need this on MySQL and I'll work on seeing how I can achieve that.

cheers.

On 20/01/14 6:15 am, Oliver Tappe wrote:
Hi there,

On 2013-12-16 at 16:02:09 [+0100], Ingo Weinhold <ingo_weinhold@xxxxxx>
wrote:
On 12/16/2013 10:56 AM, Andrew Lindesay wrote:
I have had a look at OBS this evening and can see how it works.  It will
be a bit of a learning curve for me; would it be more efficient if I
were to document how to build + launch the application as a java process
and leave the OBS packaging up to yourself or others who run the system?

Sure, you do that. Unless Niels feels inclined to do so, I guess we
won't deploy anything on the Haiku server before Oliver is back from
vacation (in about a month). And I'd like to wait for his opinion on
things anyway.

I agree that OBS is what we should use and it'd be ideal if you (Andrew)
could provide an rpmspec-file. If you don't feel comfortable doing that, I
can do that when adding the project to OBS.

The basic idea is to deploy a test installation of the web service onto
vmrepo (the VM providing our git repositories) soon. Having had a look at
the README, the only requirement of the web service which I'm a bit
reluctant to install on vmrepo is postgres. The reason for that is that the
installation of postgres that we use on vmdev (Trac) has been difficult
from an admin's point of view (postgres needs to be dumped and reloaded for
system upgrades). Additionally, it seems that postgres doesn't provide a
threaded setup, which may or may not be a problem, depending on how the
DB-connection sharing works from inside the Java process. That's why I
would prefer either MySQL or sqlite. That being said, I'm not against using
postgres if there are reasons for doing so. Please share your views!


--
Andrew Lindesay

Other related posts: