[Koha-devel] Minimal docker images for Koha

Tue Feb 18 07:29:42 CET 2020

That sounds reasonable regarding buster-slim.

I was thinking a minimal Koha in terms of something that would never really be entered, but rather re-generated if there were any problems. But I agree about it not being worth trying to get the smallest image possible. I think that's small image size is overly fetishized. (That said, I love my "from scratch" images that only have a single binary in them, which are absolutely tiny hehe.)

That's a good point about the volumes. I suppose I was thinking about it from a testing perspective, but that defeats the purpose. 

I was actually wondering earlier how many packaged dependencies Koha really has. Off the top of my head, I think the main ones are Apache httpd, Zebra/YAZ, and Mysql client libraries. It looks like "unzip" is needed as well. A number of the other dependencies like xmlstarlet, tinymce, libjs-jquery, mysql, etc are just for the Debian packages. (Actually, I think there is a mysql client binary use in one of the unit tests which seems suboptimal.) Oh, and due to your use of that static file middleware, Apache httpd isn't even needed. A person could probably run the Zebra server in a different container, share a mounted volume for the data, and then just have Zebra/yaz client software in the Koha container. Or if you went the Elasticsearch route, you wouldn't need Zebra at all, right? So in theory you could maybe get away just with "unzip" as a dependency that needs installing via a package manager? (Technically, we could use pure Perl unzip modules as well...)

David Cook
Systems Librarian
Prosentient Systems
72/330 Wattle St
Ultimo, NSW 2007
Australia

Office: 02 9212 0899
Direct: 02 8005 0595

-----Original Message-----
From: Julian Maurice <julian.maurice at biblibre.com> 
Sent: Tuesday, 18 February 2020 5:09 PM
To: dcook at prosentient.com.au; 'koha-devel' <koha-devel at lists.koha-community.org>
Subject: Re: [Koha-devel] Minimal docker images for Koha

The correct URL is indeed https://hub.docker.com/r/julianmaurice/koha
Sorry for that :)

I just tried buster-slim. The resulting image is 1.06GB. That's ~100MB less than with buster, not bad. But the debian page on dockerhub also states that "These tags are an experiment [...] and are definitely subject to change" so I prefer to stay on the non-slim variant for now.

I also tried multi-stage builds, but I didn't like the fact that it leaves an untagged image after each build.
Also I figured we might need the build tools once in the container, to install a missing Perl module for instance.
And to be honest, after the "failures" of alpine and perl-slim, I decided it was not worth it to try to build the smallest image possible.
But you're right, using multi-stage builds we can further reduce the final image size.

And if you really really want the smallest image, I think you should use alpine and recompile perl with the correct flags to avoid the MARC::Charset problem. I estimated that you could have an image of 800MB or less.

About mounted volumes, from what I understand volumes are for mutable data, and it's considered bad practice to use them for anything else.

I would really like to see a non-Debian based image of Koha. It should be relatively easy to do so, by replacing the apt-get command with the corresponding one for OpenSUSE.

Le 18/02/2020 à 01:25, dcook at prosentient.com.au a écrit :
> Hi Julian,
> 
> Awesome to see you playing with Docker. I love Docker and I've been thinking about assembling similar Docker images for Koha, but haven't had the time, so hurray for you doing it!
> 
> I couldn't get your first URL to work as I think that's an authorised 
> user link. Is it safe to say that this is the public link? 
> https://hub.docker.com/r/julianmaurice/koha
> 
> Did you try the "debian:buster-slim" image? I keep meaning to look to see what makes these Koha images so large. With an interpreted language like Perl and Koha's immense functionality, I think it would be challenging to get a small image size. RedHat's registry.access.redhat.com/ubi8/perl-526 image is 658MB and that's without anything application specific even installed.
> 
> That being said, I think I see some places where you could improve your Dockerfile. Have you checked out multi-stage builds at https://docs.docker.com/develop/develop-images/multistage-build/? In your final image, you shouldn't need packages like gcc, git, *-dev, make, cpanm, etc. You can install those in a "builder" image and do the Koha build there. Then you can copy your Perl libraries and built Koha into the final image. That should also help you cut down on the number of layers you have.
> 
> Nice work identifying those system perl dependencies. I keep thinking how we should minimize those. It should be reasonably easy to fix those.
> 
> Actually, if Koha and its Perl dependencies are localized, you could override those with a mounted volume... which makes me wonder why you couldn't use a more minimal Linux...
> 
> I've actually been meaning to make some OpenSUSE Koha images which would help test out the concept of a minimal Koha running on something other than Debian as well, although one may as well use Debian, since that's the supported OS for Koha...
> 
> David Cook
> Systems Librarian
> Prosentient Systems
> 72/330 Wattle St
> Ultimo, NSW 2007
> Australia
> 
> Office: 02 9212 0899
> Direct: 02 8005 0595
> 
> -----Original Message-----
> From: Koha-devel <koha-devel-bounces at lists.koha-community.org> On 
> Behalf Of Julian Maurice
> Sent: Tuesday, 18 February 2020 4:59 AM
> To: koha-devel <koha-devel at lists.koha-community.org>
> Subject: [Koha-devel] Minimal docker images for Koha
> 
> Hi all,
> 
> I've been playing with docker lately, and I tried to build a minimal docker image for Koha. Here are the results.
> 
> My goals were:
> * Install only required "things" to get Koha up and running, and 
> nothing else (no testing or dev tools),
> * No external dependencies except CPAN
> * Follow Docker best practices as much as possible
> 
> The resulting images are here:
>     https://hub.docker.com/repository/docker/julianmaurice/koha
> 
> and the Dockerfiles are here:
>     https://github.com/jajm/koha-docker
> 
> A few things worth mentioning:
> 
> * I tried to build the smallest image possible by using alpine or perl 
> slim images at first but it was not that great, because the perl 
> version shipped with those images is missing some libs, which cause 
> MARC::Charset to build a database of several hundreds MBs (which is 
> only 5MBs with a standard perl version). So I chose a more standard 
> image
> (debian:buster) as base.
> 
> * Koha doesn't work well when running with a perl version different than the system perl installed in /usr/bin/perl. For example, the updatedatabase doesn't work when called from the web installer. This is because Perl scripts are called directly as executable files, and shebangs contain '/usr/bin/perl'. Same problem from misc/translator/translate which calls tmpl_process3.pl.
> 
> * I tried to make the Koha installation as self-contained as possible.
> Almost everything is installed as a non-root user in /home/koha, including Perl dependencies.
> 
> * It doesn't need a reverse proxy such as apache or nginx. The necessary URL rewriting is handled in PSGI file. The container expose two ports, one for intranet, the other one for OPAC.
> 
> * Each Perl dependency is installed in its latest version, so expect things to break. I can only confirm that the webinstaller, basic cataloguing and search/indexation work. I did not test anything else.
> 
> * There are docker-compose.yml files in the github repository to get Koha running quickly with mariadb, memcached and elasticsearch.
> 
> * Zebra is not installed
> 
> * Images weigh ~1.15GB uncompressed (koha sources included)
> 
> If you made it this far, thanks for reading :) And if you want to use 
> these docker images, you should start by reading 
> https://github.com/jajm/koha-docker/blob/master/README.md
> 
> --
> Julian Maurice
> BibLibre
> _______________________________________________
> Koha-devel mailing list
> Koha-devel at lists.koha-community.org
> https://lists.koha-community.org/cgi-bin/mailman/listinfo/koha-devel
> website : http://www.koha-community.org/ git : 
> http://git.koha-community.org/ bugs : http://bugs.koha-community.org/
> 

--
Julian Maurice
BibLibre

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 484 bytes
Desc: not available
URL: <http://lists.koha-community.org/pipermail/koha-devel/attachments/20200218/0b39fc89/attachment.sig>