Drop smartos support?

[bnoordhuis]

• https://www.joyent.com/blog/a-new-chapter-begins-for-triton-and-smartos was reason for libuv to drop support back in April (libuv/libuv@6dfcdb9)

• BUILDING.md claims it's a tier 2 platform but I'm unaware of anyone actively maintaining the port

• CI buildbots are pretty flaky

• port has probably 0 users

Suggestion: just drop it?

[richardlau]

• https://www.joyent.com/blog/a-new-chapter-begins-for-triton-and-smartos was reason for libuv to drop support back in April (libuv/libuv@6dfcdb9)
• BUILDING.md claims it's a tier 2 platform but I'm unaware of anyone actively maintaining the port
• CI buildbots are pretty flaky
• port has probably 0 users

Suggestion: just drop it?

cc @nodejs/platform-smartos @bahamat

[sxa]

Can we check the download numbers on it? It could be moved to unofficial status, although we cross-compile most of them so may require extra effort. I would imagine that even though libuv formally dropped support it would still be kept working by virtue of them still supporting SunOS, as you say it should really have someone who would be willing to maintain it ...

[richardlau]

Can we check the download numbers on it? It could be moved to unofficial status, although we cross-compile most of them so may require extra effort. I would imagine that even though libuv formally dropped support it would still be kept working by virtue of them still supporting SunOS, as you say it should really have someone who would be willing to maintain it ...

If anyone wants to crunch numbers: https://storage.googleapis.com/access-logs-summaries-nodejs/index.html

[sxa]

If anyone wants to crunch numbers: https://storage.googleapis.com/access-logs-summaries-nodejs/index.html

Broadly comparable to AIX then ;-)

[bahamat]

Hi, I’m the maintainer of a very large project using Node on SmartOS. Our codebase consists of over 500 repos, most of which are node.js. We have thousands of users. Our customers provide services that are used by millions of end users. All depending on node.js running on SmartOS. So the premise of “probably 0 users” completely false.

As part of this project, I maintain several node modules that are direct or transitive dependencies on millions of installations, all of which are developed on SmartOS first.

We provide infra to the Node.js project for build machines and various other services, for both SmartOS as well as other operating systems.

Despite this, there seems to be a concerted effort from some people to kill Node.js on SmartOS/illumos. We’ve seen reduced collaboration over time from the node project with regard to maintaining and keeping support for SmartOS, despite our efforts. After node stopped producing binaries for SmartOS, we’ve been building our own, and needed to maintain a large set of patches to ensure that happens properly.

We would gladly work closer with the node community and have proper support upstream. We currently have 21 patches that we apply for builds of node 16 (we haven’t yet had time to port these forward to node 18). We would gladly contribute all of these patches, but we’ve seen little interest in them being accepted.

As mentioned by @bnoordhuis, Triton (and all of its code, customers, and users) now has a new home, with new management, and new personnel (now with MNX, no longer with Joyent). Our project is not dead, we are actually seeing renewed interest, and MNX is investing heavily in SmartOS and Triton.

Instead of dropping support for SmartOS, I would really like this to be an opportunity to let the past be the past, and treat this a call for working closer together. We’re still working on some things related to the transition of the code base from Joyent to MNX, but within the next couple of months we’ll be fully standing on our own and would really like to take a more active role in maintaining the port and providing support for the platform.

[sxa]

That sounds great. I'm a little surprised that you've seen resistance to your patches being accepted (I don't think I've seen any of those myself) but I'd hope we'd be willing to take patches that didn't cause any problems elsewhere. I certainly don't have any negative impression of SmartOS/illumos (I'm a long term Solaris fan!) and it would be great to know that someone was willing to continue to contribute to the main codebase and keep the support for it alive and functioning in there, so I'm personally 👍🏻 on working closer together.

Has the lack of support at the libuv project been problematic for you at all?

[bahamat]

@sxa A lot of that happened long before I got involved with the project and took over maintainership, to the point where I wouldn't even be able to cite specific examples. Which is another reason, IMO to just let the past be the past, and just focus on going forward.

It may take a little bit of time, but we'll start working towards organizing what we've got and figuring out what to upstream. In the mean time, if there are any pressing needs or anything particularly broken with SmartOS, please send it my way, and we'll see what we can do about addressing those. You can also hit me up on the OpenJS Foundation Slack, if you have any questions or need to discuss something where a github issue isn't quite the right venue.

As far as libuv goes, I didn't know they had dropped support. We've been producing our own builds to go into Triton and SmartOS, essentially forever (which is one reason why there are low download numbers). That's something we'll get to, and hopefully nothing will have broken too badly there. But we'd also like to do what we can to ensure that libuv builds and runs on SmartOS as well.

[sxa]

Sounds great! It may be worth engaging with libuv particularly in light of your interest in keeping such a critical component of node working, since if there are any platform specific issues in the libuv implementation we'd generally want them to be upstreamed.

Do you run the node test suite regularly on your builds? I know there has been some mention of particular flakiness in the smartos CI jobs and it would be good to understand if you're seeing problems too or if some of it might be specific to our environment and therefore resolvable.

[cjihrig]

@bahamat libuv does have an illumos team. Would you (or anyone else) like to be added to it?

[bahamat]

Yeah, I’ll be evaluating the situation with libuv in the near future. We currently have 9 patches for libuv.

I’m not sure about the test suite, but I’ll look into that. Again, it’ll take some ramp up time for us, but we’ll start working on evaluating the test suite, and seeing how it goes both with and without patches, and getting forward ported to latest.

@cjihrig Yes, I think that would probably help. This also brings up a good point. In most cases, SmartOS support doesn’t need to be different from illumos support. If libuv (and node for that matter) simply want to stop treating SmartOS and illumos differently, and just support illumos as a whole, that probably wouldn’t be an issue from our side. In any case, I’d like to have a closer relationship, so everything else I said still stands.

[bnoordhuis]

Data point: in the last 4 or 5 years libuv received more patches from traditional Solaris users (!) than smartos and illumos combined. Hell, I personally probably fixed more bugs than the whole of the illumos community together.

The commit history for node is harder to pry apart but it doesn't look thriving either. If no one's maintaining the port and CI flakiness adds friction for contributors, it should go.

[bahamat]

I don't think that should be that surprising.

We address bugs that are illumos specific. As I mentioned, I have a code base that consists of over 500 repos. I have a lot more to do than fix bugs in libuv that will almost certainly be addressed by the libuv community. Likewise, we (Triton, specifically) don't regularly contribute to nginx, haproxy, or postgres, all of which we heavily depend on.

I don't think anybody here is expecting the illumos community to become core contributors to node.js, libuv, or any other project. Those communities are more than capable of addressing general bugs. But where there are edge cases that need addressing on our platform specifically, we're more than happy to directly contribute.

As far as each point in your original comment:

• https://www.joyent.com/blog/a-new-chapter-begins-for-triton-and-smartos was reason for libuv to drop support back in April (libuv/libuv@6dfcdb9)

This is a mistaken assumption. SmartOS isn't dead. It just has new owners (with renewed investment, I might add).

• BUILDING.md claims it's a tier 2 platform but I'm unaware of anyone actively maintaining the port

I've volunteered MNX to actively help here.

• CI buildbots are pretty flaky

Again, something we (MNX) intend to address.

• port has probably 0 users

Again, a mistaken assumption. We make our own builds for binary deployment in our own products, and node.js is included in our packaging system for SmartOS. Which means that the vast majority of downloads for SmartOS will be our own builds from our own servers. The only time someone needs to use a build not produced by us is when we don't yet have a version available.

Given the above, I'd like to move to close this issue, unless there's something more that needs to be discussed.