OpenJS Foundation OpenJS Foundation Collaborator Summit, Berlin, May 2019, 5 Jun 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Node.js Release Working Group

Description

OpenJS Foundation Collaborator Summit, Berlin, 2019

A

The current researching kind of varies between one per week may take a little bit less.

B

We release mere moments, trying to do the number of conflicts in Turkey, which means that we assume as a fisheye lens. We have a look at it and then we open the bank report Wizards and we're patch. So we do and Pacific everything is not breaking change from the signature to then, when the original care is they've on top and then.

C

Being or even through, while reviewing some of the repairs to ask as possible.

B

B

As long as we do that that's fun, because um yeah I would actually say is more case where we care less about that than before, because we do beg for it everything release team backwards, mmm whoever did Kara or release team ever much like in this case. You just have to make sure it's backwards from some carries. We can directly say it's making sense, because breaking changes helplessly everything it does is great, but most PRS can normally do like.

B

Okay, you switch like this part and then the rest of the hold still same-same but I mean and the breaking changes calm.

B

And then you have less confidence, because in the beginning, when we do release find a major, it's easy going looks like everything went, cool, the more release you do. The more conflicts hit, one of the rights to always prevent is blaming chairs out of order here. Another corner.

A

We know just maybe.

A

When we hit me legs, we have amazing back requested.

A

So now we're attend events where we can pull in really changes. People either ask for or likely request the branches, like. That's a really trying to do in that heat event all its birth. You have that forever.

B

So yeah there's no way it's that's what we can do I'm going.

B

Will become easier also for the maintenance.

C

So I'd assume open.

C

A

A

B

Everything everything is configured something. The question is: is there a checklist.

A

D

That's one of the things that I'd like to do eventually is like on, say, I'm just a bit. I.

A

B

Create worse sometimes,.

B

They change something in the now that so you actually.

B

C

B

C

Correct order which the commits can land that we do.

C

B

This is even less confident, because now we have like it almost always right.

E

D

Really increasing incrementing log of the PR numbers like.

B

Can't you just go back or no, no, no, we land in carries yeah.

C

Here are number orders when was open but, like some PR explained right away other's techniques years right.

D

But even if, like anyone that takes weeks, we, the ones in the future still be no.

C

B

Like Lenape are now you like turns out first PR.

B

C

Rafael is joining the zoom chat and I'm just gonna. Let him know that there's it's a pretty small group here. There's only two people from the release working group- that's F, creates and Reuben rich water, and then it's me Tierney. While he and me know.

B

B

What we are working on is.

D

You chose an example also.

E

Just to clarify what you say when you landed me here here, but major.

D

C

Come a little trouble hearing it in here, but it's I'm just a little bit trouble. Is this.

C

This looks like it would might be the microphone.

B

So- and this is actually it just means- there is a conflict in some way in this period that the releaser cannot resolve on its own immediately, because sometimes they're, like tiny conflicts, that I just we solved right away while ending the commits mmm like if it's obvious, you know.

F

B

Then then you can do it and you run the test and everything passable. But if it's not that way and it's conflicting and then you go into the PR and market as Beckford requested with the current release it you're preparing and then many old bank or it has to be open for it, take ported to 11. They are back for two tubers makes. That means it was requested, and it is now back ported which is a bit redundant because we don't normally use it.

B

Then we have don't land on version, which means sometimes we say explicitly. This should only be back ported until a specific version which could be important for allocated v8 change and because I mean some change might be December minor, but it could be like the slow down on old versions, but the performance improvement on new versions so, depending on the v8 version, as an example yeah and then.

E

You reach a man.

B

E

B

F

B

B

So that's the don't land and that's always during the lending process. Already normally sometimes someone might come also later on, but that's super rare, not only we just don't have it and then we have the version, whatever version it is and we use it for open back court, be ours. Mm-Hmm.

E

C

Is not in a long knowingly.

B

Which means that when you manually backward something, this label would be attached to it. Okay,.

A

E

A

People couldn't tie to this well like brackets. Yes,.

B

Except in the title, it's always good to dude and and like almost no collaborator, knows how to use these labels.

E

It kind of sounds like y'all are really overworked, and improving process is a nice to have.

B

Yeah so yeah, because, for example, one very typical mistake that people do is that they and let's say mmm they- want to have a PR BEC imported on a specific release line. Then they actively at the back court requested label. Now that does exactly the opposite of what I actually want.

A

B

Renaming the label would be yeah.

D

C

Because all those we've discovered on this product naming labels is hard and I submit to you. The offer ready.

B

Exactly it was like multiple things, but oh my god, and it's still not really a lot of people use it.

E

C

E

A

Mean to get us off topic.

A

Things today, I'm not sure how will we socialized it? You can search PRS and hit that court requested to tend or X, and you can actually you could try and picking up one of those PRS and then basically open the diff and try to apply that if to the Tendo ex released line. So normally, sometimes it's as simple as a few variables. If that the names changed, you.

F

A

Need to reflect that change back on your branch.

F

A

Typically, the owner of the original PL will do the backcourt for it because they're the most familiar, but it's not. Everyone keeps up with it and there's something around student years. So.

C

Would backporting a forgotten unloved PR be a good first or second contribution for someone, because my my feeling has always been back. Ports tend to require, following a bunch, more instructions about what what branch to open it against and then you know how to resolve the the end of the conflict that resulted back port being requested in the first place, but I also kind of wonder if having a bunch of instructions to follow is actually good. For first time person they're.

A

C

That's much better than just being left to their own devices.

A

Because you've got the code there, so you.

E

A

Don't necessarily need for a trend feature. You just need to know like out how to apply that exact. Therefore, that code change to the unreleased line, which can be quite a nice intuitive way of starting to contribute to cool okay.

C

B

Some on that, okay, um because it's actually often challenging to know like especially let's take ten to alpha when you bet port, some PRS or, like you 14050 2:8. This is so immense that and then you have like often the worst backward carriers are C++ code. So people have to be very knowledgeable about the actual code that was changed and not only about the change itself, but also about former implementation and about.

C

Whatever change is causing a counselor, yellow bleah as well.

B

And then and to know how to resolve that conflict, and not only the conflict because sometimes changed my Glenn, but then a code change in another place. You change the logic that you have to identify why it is breaking. You had to run yeah.

F

B

F

B

Quite challenging so.

C

B

For the request, yeah.

C

For the ones I've been sitting there like, there are certainly back ports where it's just like. Oh, it's just a white space diff that needs to be sorted.

F

C

There are easy ones for those those tend to just get done.

B

A

When you're you're up late preparing a release, though 200 commit sin, and you can see like a load of white space changes or lots of written conflicts, I'll just have to back ported labor long, because I think I cannot stay up. Another half now just to get this ended when I've got a hundred more, but.

F

A

You've got the issue where I'm now waiting for that commit to be back ported or the order goes like it's. A compromise I'm either sacrificing the order in which it lands or I can't land any more commits, and when we're only doing one minor per quarter for an Audi s, release line yeah.

E

A

We can't wait for.

B

A guess it is going to become easier when we have the automation process and then.

D

Is this thing work done.

B

No okay and I mean.

D

I I can probably do.

B

Either automation or.

G

D

This is something I've one do for a while. I can do automation or releases so.

B

Do you choose, for example, was and like we meet asbestos multiple times, but time is so limited for all these, so what you will also work with a lot of station yeah.

F

B

And like Michael, and it is doing a lot of work there and they shed Lao and me, and the thing is with all the releases you have to run in CJ, Martin yeah, and then there is a lot of flakes in there and a lot of PRS that you have to update. You have to resolve the conflicts, my modules breaking when you have to update the list you have to up to the code yeah, it's a lot triage.

A

And Simmons a big time, six yeah.

B

So it's just very, very time-consuming and we needed and to release more versions. So the time was invested into releasing new versions instead of automated music yeah. But.

F

If you miss something about during the release, you know right away, or it has been time that you lived and it they passed. The new Liza mistake, and then you have.

D

You have to be Simon version.

F

Okay and then does that happen very often so.

B

I just did recently, but it was like I mean that's happening.

C

It's the exception, not the rule, but it.

B

Certainly happens.

D

There is really.

C

Something on a Tuesday and then we realized, oh, my gosh. We should not really something.

B

Very strong there are very different opinions of the days so, but the main problem about this is all comes down to when the individual person s yeah I,.

C

Think that should override all the concerns about. Oh, we don't want to do release on from.

A

E

Especially especially.

C

And and counter-intuitively, perhaps that's especially true with security releases. We we want to get them out when we can get them out and yeah. You know like people cuz like we, we we don't surprise people with them. We tell you know you know we post a week in advance, we're gonna do a security. You were a new security release on some on or around such-and-such a day at such-and-such a time and.

D

C

Know sorry, if it's on a Friday and you think they shouldn't be done on Fridays matter.

A

Like you know, we're we're telling.

C

You in advance and.

C

Then, of course, there's Sam there Sam's proposal actually like have a regular cadence for security release, is a very, very good idea, because the release team can set aside time and the users can set aside time and it'll be like Patch Tuesday. Everybody will know when it's coming a long, long long way in advance, so I guess the.

D

Question I have is like I would love to do both, but there's one automation and releases okay first, so so.

C

You put in the cart way for the horse you're trying to figure out which one to do right.

D

C

Yeah and then figure out what you're going to get like.

B

D

C

Or yeah yeah you're, not you're, not gonna, be able to automate without actually doing yeah. We're probably be my yes.

A

Can do the releases reason.

A

B

Like I, also like just recently, I was angry about myself. When did a small mistake in and the multiple changes was just early on and well, the notable changes is missing, a name from it. Well it because it changed in release, notes in a memo and I copied something wrong.

B

Yeah and because of that mistakes happen, I guess I'm the only one who noticed but.

E

So, as far as process goes, you said that miles has been onboarding you yeah, and so the way you envisioned helping newer people get familiar with it. Did you mean that they would pay you back on the existing process or you were going to start your own meetings to.

A

Any of us kid and what I've started doing is scheduling time at work when I'm allocated to work on backporting and things yeah.

E

A

I think Shelley does look similar amongst us and then we went for a phaser stifling resume whenever we were doing it. Yeah.

B

A

Were talking and just shouting what we're doing talk free, okay,.

E

A

Have time right.

D

Cool cool I.

G

Hope this problem is some early position. What is the motivation behind back putting everything in the upper branches to the healthiest branches? Ideally, we should be looking at only the stable ones or the ones which are actually required by users. Yeah.

A

I think it's an complex cutting forwards. So if you back for everything, you're less likely to get conflicts when something that comes in yeah.

G

But then that works against the ideas philosophy, for example, in the the bleeding edge we see something breaks. Quite often we see that and we are actually inheriting all those breakages into the areas we.

A

Do have time frames so your any special things back once it's been in current, like X number of weeks, yeah and- and there are some it's kind of like a release, a case-by-case basis. There are some PRS and eight miles and myself definitely with that and we've seen people have posted issues on it and we haven't pulled that back. So it's not like a black and white. Absolutely everything gets pulled back. There is some cherry, picking of which should go in yeah.

A

It's not clear, is I, think our documentation. So it's based on a best case. We pull back or we can yeah.

G

So is it possible to classify the features which are landing at in the master in a way that these changes and the family of changes? What is happening over there is potentially backward porting graduate so that you can actually start back, but.

A

Yeah I prefer that way myself. If it was when you open this I was an author you're like should this 10:8 and if you added labels at that point, it would be easier for us because we would know what needs to come back. That.

G

Way, you're reducing your work and Lord and start back coating as soon as possible, so that you don't face the problem of complexities.

B

And done in the other way around so normally we say by default every every minute and we'll be back for it. But if you, if anyone expresses concerns about it during in the PR or even later, they add the don't land on version really slap, the it creates wrong land and that also automatically implicitly implies that all conversions below that one and should also not be taken into account mmm normally.

F

B

Every explicitly, but it is actually from the release team done implicitly for all versions below, hmm and so we have something like that. We also have the base for LTS, which means like nth moment as the same thing. Then there is a at least I think, two weeks time frame, you.

A

Think it yeah two.

B

Or three yeah and and the PR has to be released already on a different release. I'm on the current one before it may be, big parted at all. I want the LTS ones and I'm I, don't know, but it was there any in like did. He have any complains or concerns, because quietly I mean aware that yeah.

G

There wasn't any complaint as such, but the way I look at it is LDS by default should not change at all. It should focus only on field reported issues, and the issue is resolved in the latest branches latest releases to put back for it, because it's already fixed and has some amount of.

G

Test cases that's wrong on that and proven to be stable. Then you can graduate to the LDS yeah.

A

I was kind of speaking to some other similar saying. Why do we put minus in LTS release lines for like nine to ten months? Because if you didn't and people would happen incentive to me back to the latest LTS, you use the new features that would be less work for us, because we'd only have to have that cool yeah fixes, oh I, think we would still backup features if people raise cases for them that those features would obviously conflict at all. Yeah.

G

I, don't have the numbers, but I see issues coming from the field which are from the healthiest. So if you actually look at in a systematic manner, we can see that some of these issues are coming from backwater issues, backwater, peers, yeah, for example, something which works on 10.15 bricks in ten or fifteen. Sixteen. So definitely it come from back. Both yeah.

B

A

B

Something this could happen with a fix as well.

G

It could happen with the fix as well, but yeah that's a classical regression, but whereas in this case it was a feature which was not intended to touch this area of core at all and.

G

Yeah, it's all regression but you're, not addressing the users pain point, as you are not addressing a reporter issue. You're, you know entering a feature fixing some of those problem, but eventually causing some other use case to break.

G

A

Think he'll be a pretty dramatic change to say we let we kind of be saying it goes straight into maintenance after coroner come battle, will only get like victors to regression. Bug fixes. Maybe I I, maybe have some exception, saying that some good plan that we tell follows yeah.

B

And I guess it's an interesting point. Definitely and we could discuss it, but then I guess it should be a bigger round, because this sounds like a huge thing. They've changed like like about every really fine, we've had so far yeah.

A

B

Not even want you beside this as the release group, but more like all collaborators sure.

G

Yeah, but it's a clear trade-off between how much time you are spending on development effort, I mean you reduce you. You tend to do back, put as much as possible to reduce the development effort, but on the other side the side effect is going to be consumers. The.

A

One element somehow when I mentioned this earlier, was that and we want people to be able to try out the features on older released lines before they upgrade.

G

A

We don't backcourt some of the minors by any additional features. They wouldn't have that opportunity to test something out or they're unloading.

G

That's going to be more disruptive than us.

G

Because, right now everything happens in the master branch I.

G

Cannot I mean.

C

C

Know I mean I mean I finally, I find I, certainly. Finally, the argument that.

C

Not you know that by by keeping the LTS branches as up to as in sync as possible with master, it greatly simplifies.

C

You know the the maintenance of those branches in that you know new things land much more easily and there aren't big impossible, plus plus heavy back ports that need to happen. On the other hand, I also understand the expectation you know. The default expectation of an end-user is when they is when they have a branch. That's marked LTS. This is not a node thing.

C

This is you know what LTS is to most people is something that will you know that will be supported for a long time and on and and and I mean it's really all, though okay, so they support for a long time and undergo minimal changes and certainly no breaking changes, although we're not talking about making breaking changes LTS, but then, on the other hand, I do think that the main thing for most people is that it's supported for a long time that they that they're never they're not going to have to endure breaking changes, but not not not necessarily that there won't be.

C

You know a lot of updates, so I, don't know so I mean I mean and ultimately, this the release working groups decision, but I do also completely appreciate. Rubens comment that, like this is not the type of decision I want just the release group.

F

We want to support not the new features that would be spotty for over. Yes, that we care about not the.

C

F

Like it's, okay,.

C

It's okay: if there's a lot of little documentation updates in the world.

D

You pick a release and that's what you stick with for a while, and then you eventually upgrade.

B

Actually, like mostly fests, are not about the bad things, but about.

C

Also I mean I mean.

D

The whole okay I just on that is it from the same people where is it from so? What I'm saying is? Is it from indifferent people asking like individual features, or is it like one group asking for a new feature over and over and over it's different persons yeah so that that that doesn't exclude what I or what I I'm asserting because like? If someone wants one feature, they might get that an investing on effort and then they want upward yeah.

D

So if it was a same group coming multiple times, I would say: okay, that's an indicator that they want to consistently update. But if it's like people out randomly asking for, for you know this feature of that feature, that's not necessary signal that they're, like try all trying to update fast and, like I, mean with my experience of both Microsoft and outsource. Like people. Stick, they the conversion and that's where they stay.

B

C

There's another there's another thing that we haven't actually touched upon here, which is that there's not all there's actually like, like we're talking about like current and then LTS, but there's actually, two flavors of LTS right, like we like, like for the first year, LTS active and then there's LTS maintenance and.

B

C

B

C

B

C

Point the point: the point is that whatever they're named, you can make the case that the you know that yeah once it's in once it's in maintenance once it's in its final year, we're not back we're, not gonna bother like we're gonna release only when there's a security fix or a critical bug. Otherwise we're not touching it. In fact, that might already be what we.

A

C

And so, and, and so that would be consistent with what you're suggesting, but then the LTS active line could be something that is I mean yeah I mean it's right there, the name right active. It's not. You know it's it's getting more than just critical fixes and security. I.

F

F

C

Only do it for the current really it was career. This is only current and LTS. So so it's at most four branches, because there's really there's a brief window where there's three L Tia, where there's two maintenance, LTS branch, isn't one active LTS branch, but usually it's it's see, I. Usually it's three branches.

B

C

Well, I mean so there's a lot of inputs here right like there's what what do end users and customers expect. There's what to the companies who support a lot of end users, customers, the IBM's, the the near forms, the note sources, the Microsoft's. What do they expect and then there's?

C

What can we do, that doesn't make the volunteer releasers completely miserable yeah and then from these things, and these things might be at odds so I think it survey is a good idea, but I also think that, like you, know, kind of like actually actively thinking about what, if what, if what we need and what our customers need are very different things, how do we resolve that? Because yeah I suspect it's more likely that that's the case than that? It's not the case. Well,.

B

C

That's the other thing like what.

B

I really hear is pretty much the space I know your is your suggestion, but for me, I think the main things that we could do as many as releasing workgroup is either. We keep it at the moment that we beg, hurt everything after a specific time and we try to actually get everything in there like really everything we were just talking about earlier. We started to beg for things and were majors in the zember pathway, so we add a specific commit on top of this ever maker.

B

That makes a summer patch to move to breaking changes um and on the other hand we could an extent the maintenance phase of the LTS release, which is pretty much exactly that that we only beg poured some fixes, like only patch commits and security visas, and then it would mean okay, we have to do less releases overall, which was less commit, so that's also, okay and but I, don't know so. Here's little banners. You know I.

C

Think I know what I would recommend if I were on the release. Working group which I not but I, would I would say, do what you do. What you were planning on doing, like you know, backport everything I'm, not back port but land, everything in the same order. You know the breaking change parts and but like actively seek feedback like from people like people, people might think they be bothered by something that when it actually happens, they're not bothered and vice versa.

C

They might think they want something that when they actually get it, it causes them problems. I mean it might not be a bad idea to pursue something. You know something you know message it. Let people know it's happening and so just see what happens. It's.

G

C

It's not like you can't decide to just stop doing that. You can absolutely any point. Just stop.

E

Doing something, though, once it's already announced to people seems like there would be inertia and they would push back people don't.

D

Like it when I'm try together, don't like.

C

E

C

E

C

People have you taken away from this case. Is the release working.

D

Group, like you.

C

D

C

Should probably consult in one conversation, go ahead and keep talking.

D

Like people don't have signals for when a superior lease happens, it's very not straightforward, like a bunch of individuals, tweeted and nodejs by retweeted, by email,.

B

Now guess that that's the only thing so that comes down so different part, because our communication with changes is not great. Like last week, I've been on a conference and a.

C

Clock doesn't know about.

B

B

And it all came down to documentation and knowledge about changes. Yeah like people did not know that something was in a specificity, so they don't know that the teacher was actually in no chess at all I really.

D

Like us to like making a list of what's in and I, think it exists somewhere, but it's like not something we use yes, but.

B

D

What's in each release,.

B

So we do have the change mark, but that's something not consumable yeah for people and and I guess that comes a little bit in that direction, but it's not really about release group. In this case release group is mainly focusing on doing movies, but this is mainly about bringing knowledge and information to the user, and this is something we really have to.

B

Understand this is how it's.

E

B

Phone as it gets.