Options for the Future of the Domino Open Liberty Runtime

Nov 18, 2019 10:57 AM

Tags: java liberty

I've been thinking lately about what I can do with the Open Liberty Runtime project. If you're not familiar with it, the gist of it is that it gloms the Open Liberty Java/Jakarta EE app server onto Domino to allow deploying up-to-date WAR-based apps "to Domino", sharing the server's access to NSFs.

In its current form, it's serving me reasonably well: this blog is running on it, for example, and writing the SFTP server to target it has been a delight. I've been tracking a set of issues about it, mostly relating to improving the administration/deployment side as it is in ways that would improve my current use. There are some more-out-there ideas like running apps from NSFs, but for the most part it's been focused on "a JEE server attached to Domino".

Slightly-Tweaked Model

But what I've been thinking about lately is tweaking the model a bit to be less like "a new HTTP runtime" and more like a series of related servers, which would then be proxied to. This is in line with what Liberty has been trending towards anyway. Though it started out as essentially a cleaned-up WebSphere monolith, the fact that it's been targeted at cloud-service use has meant that it's been aggressively tuned to allow you to include only the specific features you need, with the idea that there will be many instances of Liberty running, one per app. This is also seen in their relentless push for faster startup times, something that only matters if you plan to start up instances of the server very frequently.

My Runtime project technically supports this model currently. Though it uses a single Liberty runtime download, the main Servers list is geared towards creating multiple running servers with their own app sets, which end up being distinct processes:

Open Liberty Admin Servers view

So one could hypothetically put dozens of servers in here and use it as I've been describing. Other than providing that capability, though, the tooling doesn't really help you much: in particular, it would do nothing to alleviate the problems of conflicting port mappings or unifying all of these apps into a single front. I did a little work making a reverse-proxy webapp, but it's really meant for the case of having one Liberty server with all of your apps, and wouldn't have any meaning in a many-server setup.

One option I've been considering here is giving the server side of this some knowledge of and control over the HTTP ports used by the individual Liberty servers, so that it could assign random ones automatically. Then, I'd also be able to add on a dynamically-configured reverse proxy that would know about these different running servers and could route requests automatically.

Other Runtimes

On top of that, there's nothing about the project that's really inherently tied to Open Liberty as such. It currently assumes that that's the runtime, but the only way it interacts with it is by pouring files out to the filesystem and then running some known shell scripts. There's nothing stopping such a setup from also gaining a bit of knowledge about, say, Node, Swift, or any number of other runtimes. Things would get progressively weirder the further you get from Java when you want to access the Domino runtime, but hey, that's what the C API is for!

Reinventing the Wheel

While mulling this over, though, I did realize one thing: I'm essentially describing a jankier version of Kubernetes. For the most part, the problem of running disparate apps with their own runtimes and needs managed by a central server is solved by that and Docker. This project would be a bit different in that the apps would automatically inherit a Domino runtime and would also (usefully) maintain clustered/replicated configuration by way of the Domino server backbone, so it's not exactly the same thing. And, while HCL is pushing for a Dockerized future, the pieces aren't there yet and won't fully be for a while: while Domino-on-Docker is a thing technically, "many Notes-runtime apps with many server IDs" for NRPC use is a licensing minefield, and the gRPC bindings are currently painfully limited in both capability and language support.

So I likely would be best off just letting time (which is to say, HCL) solve the fancier problems and just focusing on the Runtime's original job of being a better Java app server for Domino. I do think it would be useful to better support the one-server-per-app approach, especially for a hypothetical case of wanting to deploy an XPages app in one Liberty server but then a JSP or JSF app in another, and that'll take a better reverse proxy.

I think it's good to mull over, though. This already provides a good path to better app dev with Domino, and smoothing that out more will make my life all the better and will hopefully be useful to others.

Commenter Photo

Ben Langhinrichs - Nov 18, 2019 11:18 AM

It's an interesting dilemma, and I have no great answers. I will mention that I was faced with a similar issue in 2005, and had to decide whether I should let time (meaning IBM then) solve the problem I faced and focus on my core business. Spoiler: IBM never did, so I was glad I didn't assume they would. Now, HCL is not IBM (thank goodness), but I would still be a little wary of assuming too much, as they do face a lot of competition for their time and energy, and have shown some similar difficulty harnessing the resources to solve the fancier problems. Just saying.

Commenter Photo

Ben Langhinrichs - Nov 18, 2019 11:19 AM

Minor point about this blog: When I post a response, it wraps in the middle of words. Not sure why. (See my response above.)

Commenter Photo

Jesse Gallagher - Nov 18, 2019 12:40 PM

That's a good point about the "fancier problems". While I think that there's a lot of promise in this sort of app dev becoming more standard for Domino-touching development, I think I'd better prove my case more before just assuming it'll show up in the product on its own.

As for the comment, that's the fault of the blog's CSS: I'm over-compensating for the edge case of a comment containing a way-too-long word that would break page layout by being aggressive with hyphenation. In theory, browsers should be doing it only when necessary or logical, but they're much, much too trigger-happy about it in practice.

New Comment