- Weekend Domino-Apps-in-Docker Experimentation
- Executing a Complicated OSGi-NSF-Surefire-NPM Build With Docker
- Getting to Appreciate the Idioms of Docker
For a couple of years now, first IBM and then HCL have worked on and adapted community work to get Domino running in Docker. I've observed this for a while, but haven't had a particular need: while it's nice and all to be able to spin up a Domino server in Docker, it's primarily an "admin" thing. I have my suite of development Domino servers in VMs, and they're chugging along fine.
However, a thought has always gnawed at the back of my mind: a big pitch of Docker is that it makes not just deployment consistent, but also development, taking away a chunk of the hassle of setting up all sorts of associated tools around development. It's never been difficult, per se, to install a Postgres server, but it's all the better to be able to just say that your app expects to have one around and let the tooling handle the specifics for you. Domino isn't quite as Docker-friendly as Postgres or other tools, but the work done to get the official image going with 11.0.1 brought it closer to practicality. This weekend, I figured I'd give it a shot.
It's worth taking a moment to explain why it'd be worth bothering with this sort of setup at all. The core trouble is that running an app with a Notes runtime is extremely annoying. You have to make sure that you're pointing at the right libraries, they're all in the right place to be available in their internal dependency tree, you have to set a bunch of environment variables, and you have to make sure that you provide specialized contextual info, like an ID file. You actually have the easiest time on Windows, though it's still a bit of a hurdle. Linux and macOS have their own impediments, though, some of which can be showstoppers for certain tasks. They're impediments worth overcoming to avoid having to use Windows, but they're impediments nonetheless.
But back to Docker.
For a little while now, the Eclipse Marketplace has had a prominent spot for Codewind, an IBM-led Eclipse Foundation project to improve the experience of development with Docker containers. The project supplies plugins for Eclipse, IntelliJ, and VS Code / Eclipse Che, but I still spend most of my time in Eclipse, so I went with the former.
To begin with, I started with the default "Open Liberty" project you get when you create a new project with the tooling. As I looked at it, I realized with a bit of relief that there's not too much special about the project itself: it's a normal Maven project with
war packaging that brings in some common dependencies. There's no Maven build step that expects Docker at all. The specialized behavior comes (unsurprisingly, if you use Docker already) in the
Dockerfile, which goes through the process of building the app, extracting the important build results into a container based on the
open-liberty runtime image, bringing in support files from the project, and launching Liberty. Nothing crazy, and the vast majority of the code more shows off MicroProfile features than anything about Docker specifically.
Bringing in Domino
The Docker image that HCL provides is a fully-fledged server, but I don't really care about that: all I really need is the sweet, sweet libnotes.so and associated support libraries. Still, the easiest way to accomplish that is to just copy in the whole
/opt/hcl/domino/notes/11000100/linux directory. It's a little wasteful, and I plan to find just what's needed later, but it works to do that.
Once you have that, you need to do the "user side" of it: the ID file and configuration. With a fully-installed Domino server, the data directory balloons in side rapidly, but you don't actually need the vast majority of it if you just want to use the runtime. In fact, all you really need is an ID file, a notes.ini, and a names.nsf - and the latter two can even be massively trimmed down. They do need to be custom for your environment, unfortunately, but at least it's much easier to provide just a few files than spin up and maintain a whole server or run the Notes client locally.
Then, after you've extracted the juicy innards of the Domino image and provided your local resources, you can call
NotesInitExtended pointing to your data directory (
/local/notesdata in the HCL Docker image convention) and the notes.ini, and voila: you have a running app that can make local and remote Notes native API calls.
I uploaded a tiny project to demonstrate this to GitHub: https://github.com/jesse-gallagher/domino-docker-war-example. All it does is provide one JAX-RS resource that emits the server ID, but that shows the Notes API working. In this case, I used the Darwino Domino NAPI (which I really need to refresh from upstream), but Domino JNA would also work. Notes.jar would too, but I think you'll need one of those projects to do the
NotesInitExtended call with arguments.
Dockerfile for the project goes through the steps enumerated above, based on how the original example image does it, and was tweaked to bring in the Domino runtime and support files. I stripped the Liberty-specific stuff out of the pom.xml - I think that the original route the example did of packaging up the whole server and then pulling it apart in Docker image creation has its uses, but isn't needed here.
Much like the pom.xml, the code itself is slim and doesn't explicitly refer to Docker at all. I have a
ServletContextListener to init and term the Notes runtime, as well as a
Filter implementation to init/term the request thread, but otherwise it just calls the Notes API with no fuss.
I haven't yet tried this with larger projects, but there's no reason it shouldn't work. The build-deploy-run cycle takes a bit more time with Docker than with just a Liberty server embedded in Eclipse normally, but the consistency may be worth it. I've gotten used to running a
killall -KILL java whenever an errant process gloms on to my Notes ID file and causes the server to stop being able to init the runtime, but I'd be glad to be done with that forever. And, for my largest project - the one with the hundreds of XPages and CCs - I don't see why that wouldn't work here too.
Normal Domino Projects
Another route that I've considered for Domino in Docker is to use it to deploy NSFs and OSGi projects. This would involve using the Domino image for its intended purpose of running a full server, but configuring the INI to just serve HTTP, and having the
Dockerfile place the built OSGi plugins and NSFs in their right places. This would certainly be much faster than the build-deploy-run cycle of replacing NSF designs and deploying the plugins to an Update Site NSF, though there would be a few hurdles to get over. Not impossible, though.
I figure I'll kick the tires on this some more this week - maybe try deploying the aforementioned giant XPages .war project to it - to see if it will fit into my workflow. There's a chance that the increased deployment times won't be worth it, and I won't really gain the "consistent with production" advantages of Docker when the way I'm developing the app is already a wildly-unsupported configuration. It might be worth it if I try the remote mode of Codewind, though: I have some Liberty servers that Jenkins deploys to, but it'd be even-better to be able to show my running app to co-developers to work on something immediately, instead of waiting for the full build. It's worth some investigation, anyway.