Node.js Core Technical Committee, 16 Nov 2016

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2016-11-16 Node.js Foundation Core Technical Committee (CTC) Meeting

Description

https://github.com/nodejs/CTC/issues/33

A

Welcome to node.js foundation, core technical committee meeting for november 16 2016 as ever we'll start with a stand-up onna. Take it away.

B

Yeah not much I've been catching up a bit on things, I'll and it two or three of my peers that have been lying around for a while. That's that um next up is Bradley. Who is not here so Ben, hey.

C

Nothing special to report really just as usual. A lot of the first line. Hotline support, I, don't know who's next on. There's a list in front of me.

D

I think I'm, the next person who's here I was doing issues in for quest and I just did the whip, you v1 dot, n dot, one release and open the PR and committed for that next to is Evan, which I think we skipped Nikita.

E

Nothing significant just some security investigation, Lucas system, nothing else to the boss, I, don't know we're next, but probably later zone yeah.

F

From a just small PR, some comments of the are delish not much. ah Next is.

G

Jiraiya, hey um I: did it some code coverage for process next, six and process warnings um also some miscellaneous PR, an issue stuff and then, as of like I, don't know very recently say like 20 minutes ago, we put the bot script to attempt for request back for its into production. So this should automatically add certain labels. You can see that in nodejs, / github block github dash blonde / pool / in 90 I'll. Put that in the chat too.

G

So, what's that what that's going to do is if you have a pull request, if it can not land on v7 allowed that don't land on so if they can't land on the current current I can't land on the current branch allotted, don't land on label and then, conversely, or sort of opposite for LTS branches, it will, it will only do LTS watch labels if a kennel and cleanly the next is Tracy.

G

You kun I, don't think it's actually here. I should look at that arm, so that be Fodor is Fodor here.

G

But our is not here so James James is also not here. um Michael no Josh grant yeah.

H

Just going over some libraries and tools and stuff that work with inspector and helping out a bit the version management group that we're trying to get going next is Michael Dawson.

I

So we recited some support on resizing re X machines too, so they have the resources we'd originally intended. There are some contributions to help make the code coverage job. The report look better so validated those and push them through general keeping up with issues and commenting I 6th apart config and Jenkins. So we can now get emails on jog failure set up the next benchmarking meeting.

I

It started an issue just to see about putting the membership, like we've been doing with the bill group or a group I'm going to work on maybe I stable, POC attended the LCS and diagnostics meetings and just miscellaneous reviews and lands which may be already covered, but on to Brian White.

A

Brian White is not here, but he did put in that. He reviewed PRS and reply to issues. I also see that James showed up. While we were doing this so James. Are you prepared to do a stand-up, or should we keep going hi.

J

Yes, it's really good for me, I was in Japan for note rest of for few days and I've been working on the HT HT to be two implementations sit for me.

K

J

K

So uh nothing major to report from my side next would be Jen.

K

You rod, which I don't think is here that hasn't a high either shigeki and Lucas. Aren't here so next would be Michael I. I I'm not sure how to pronounce a name targus. Oh.

L

Yeah, yes, can you hear me yeah yeah, you can say Michael, it's fine, so hello, everyone I've been reviewing some poor requests and working on making v8 five dot. Five ABI compatible with our current version, five dot 4 and that's it.

A

Next on the list is miles.

M

Hello, so let me see what I wrote down so I've been working on back porting stuff to the four and v6 mentoring, a new potential back border Gibson, another IBM er, just kind of like hung out with me, while I was doing a back boarding session and started an invite to other people.

M

If anyone else on this call is interested in screen sharing with me, while I do a back porting session, please just let me know more than happy to have you involved I mentored and on boarded a new carrying the gold mine, collaborator, George Adams, another IBM employer employee he's landed a whole bunch of great commits on their general review of issues and PRS and also participate in the initial version management meeting, which happened yesterday. Next up would be Trevor as Trevor around.

N

Yes, nothing really to note, then I just responded to 94 67. Also I found a couple of performance. Regressions haven't looked deep into it, but when I was rebasing some code from six on a master, I found that use of cost um slowed things down and looked like the class like class constructor. Not just normal constructors also got a hair bit slower, so I'll look into that sometime, but I was able to work around it. One of them was just to go back to using far instead of constant that solved. The issue next is rich.

A

A

We're going to talk about this, but only briefly the ctc meeting schedule I, you know, I, had him I had a proposal to add a third meeting time. Ben had a counter proposal and that got me deciding that. I should finally stopped trying to manually process the spreadsheet and figure things out and instead write some code to come up with options based on criteria.

A

So I'll probably try to finish that up this week or something and then and then hopefully this will all be a lot less painful but we'll see um and then it's the usual test fixes watching Jenkins and you know pestering the right or wrong people when stuff doesn't work and PR an issue comments. I believe that is the end of the stand-up. Is there anybody who did not did not provide a stand up with.

A

Great, let's move on to the agenda the public portion. This meeting is probably going to be pretty fast. We just covered the first thing: the CTC meeting schedule proposal, I I'd like to propose that we just for now stick with the two times we've been rotating through until you know we get and and and and maybe a lot back channel with been a bit to kind of come up with something that will distribute the pain a little more to North America, to make it less painful on other time zones.

A

Does anybody want to talk about this, or is it okay to move on to the next topic.

G

Which, which, like scheduling schedule, are we following right now like what so.

D

E

G

Clear on like what next meetings time is going to go or the meeting after that, okay.

A

If you go to the CTC repo, the readme has a table for all of the currently scheduled meetings for november and december. Oh okay,.

G

A

And it and adjust it just currently it it just rotates through two times the two times we've been bouncing back and forth through with I, don't think, there's anything else in there yeah we haven't yet added a third time. That's what this is kind of about uh and maybe there's something to be said for taking a breather and letting it bounced between two times for a month before try to mix it up a little bit for ya the other schedules there.

A

Anything else I.

F

Was going to say, maybe be useful to send my calendar invites to everybody because I know like keeping up the time is difficult for me and, if stir the day so I'm sure it's more difficult for people who it's normally during than I.

A

Trevor expressed a similar concern: God Devon.

H

William cap key actually has a calendar that he maintains for the foundation. Maybe we can get him to drop these on there and then everyone can just subscribe to that calendar. I.

A

Think he actually already has these almost counter go ahead. Evan, oh yeah! He knows those I didn't realize there.

F

Already was warm and.

A

If that, if that doesn't work for people, uh you know if we, if we can come up with one calendar option, that will work like if I send a google calendar, invite two people I'm happy to set those up. I mean there's, there's already so much tedious involved in setting up a meeting that just one more thing isn't going to even register her.

A

Alright, well, this will be on the on the agenda in coming weeks. Let's, let's move ahead speaking of William khaki, he just wanted a sort of a public service announcement about issue 32 in the CTC repo, which is about reviving the documentation working group. So if anyone listening or if anyone on the call today or anyone reading the minutes thinks they can provide a valuable contribution to documentation and is motivated, maybe head over there and drop a note in and I'll see what we can do. Does anybody else want to add anything on this topic.

G

It wasn't like part of this blocked by the fact that we never got like guides to build from the core repo I even know who was working on that, but anyways mr. online drama.

F

I think that part of the trouble was that the docs working group was wanting to move the documentation out of core repo. Oh true,.

G

F

Think that there was like we couldn't gain consensus on that. So I think that that was part of a stall yeah.

A

I have to imagine that you know it's even even not, even though they did, even though the docs working group didn't you know, get that to happen, that there was understandable. You know arguments the other way. I have to imagine that, there's that it's still possible to have a functional working group that you know cares about the docs and is actively working on on stuff yeah.

I

I think that's true: it shouldn't be a showstopper.

A

Okay, I'm going to move along to the next topic. Unless someone tells me the stop, stop stop, which is the reverting of the buffer deprecation warning, and I believe that we are going to move that to the private segment. So unless again, someone tells me to stop I'm going to move along.

A

uh Tls fix leak, fixed the leak of right, wrap, plus TLS wrap. This is issued 96 26 and after remind myself who put this on the agenda.

A

C

Not here, but it's about a memory leak in the TLS module and I believe Fodor wants to expedite back porting it to the v6 and few for branches. I.

M

Think that we've resolved it actually in the issue I proposed that we get this into a v7 release ASAP and cut an RC with the commits next week on v4 and v6, which was the current plan anyways, which would allow us to still keep a December 6 release date for the official releases and also gives it two weeks to bakes were not. You know releasing something that hasn't gone out in a current release. Fedora said it would be.

M

Okay and one of his colleagues said an RC works for them as far as something that they can move forward with. So I think that we are good moving forward with the current plan, which does not change our release. Plans at all for LTS right.

I

And whoever whoever objected to, basically, you know pulling a patch forward bit there I forget who that was there: okay, now, because I.

E

Thought somebody.

I

Had just had an objection to the fact that we'd be putting out a patch that to openssl hadn't actually put out in one of their really. This.

C

Is the super tissue okay.

A

Great- and it looks like miles was the one who suggested that we might want to talk about that today. So if miles is happy, that's good enough for me. Unless somebody else has something to add on that topic.

F

You hey models. Are you going to cut the v7 release or I.

M

Was not planning to cut the v7 release? I was assuming that there was a v7 release coming down the pipeline because of the lib UV stuff. So I staged that change in V, 7x staging and I was just going to make sure to communicate with whoever was coordinating that release.

F

Ok, Oh Jeremiah. Do you want to take this one or you me, as I ever for.

G

What next week, I.

F

Mean what kind of top right do you think that we need all the smiles I.

M

Was under the impression that the rush was more around getting it out in the LTS releases, but I would prefer if we could get a v7 release out next Tuesday, because if we have the RCS go at that time, that gives us exactly two weeks for it to bake before we promote both of those LTS releases.

G

G

I can probably do I print release them aim for monday, tuesday. If.

F

You have you can't then speak out tomorrow.

M

Okay, excellent. Thank you.

A

Okay, I love happy resolutions I the last one is the.

A

Issue 94 67 that Trevor alluded to earlier, don't call it to vm from a sink wrap and I guess. This is a.

A

Chris Madsen put this on the agenda wanting to presumably have there be a plan of action, whether it's land, this and then land something else later or don't land this or what so I guess I've trevorr been wants to provide the current status of where the conversation is and if we at you know what we need to discuss or if we are even qualified to discuss it here and now.

C

Depends on how many people here are interested in a async rep. So the issue is that, due to the way yet garbage collected works, it's no longer allowed to call into DVM during a garbage collection cycle, and we currently do that when I hope, I get it right. So when you're using a single rep, you can install hooks on one of those hooks is to destroy hook, and that runs when a weekly, reachable persistent objects are garbage collectors and that was never safe, but it kind of yeah escape notice, and it's currently crossing crisis.

C

We've received three separate book, a book report so far, so my stopgap measure is to simply remove the destroy cook for now and if someone wants to step up and come up with a better fix and I believe Trevor, Knight more or less than agree on what the proper fix is. Well, then, if someone was working at their most welcome but as a like I said, as a stopgap measure, I'd like to remove it for now, oh yeah.

N

A

Have a talk, no.

N

You go ahead fresh dirt, where we're talking I.

I

Was just gonna ask a quick question which was the like is removing it going to break stuff versus if we just make it do nothing. I know it's. That's not going to be good for people either removing.

G

It will brace the fist right there, there's not much code that relies on current acing crap, but it does exist. Hey.

N

The fact that bug reports were filed means it is being used, which makes me happy so no, there are. There are a couple resolutions for this and.

N

Then so 1 I've gone over this code a hundred times well to working on that final PR and I've. Seen that the all the place all the handles and request that we make week, don't have to be because once they're done, they're done I feel like we did it in the past so that, if the object, if the handle was accessed after it was used, it wouldn't crash because of try and do lookups on a class.

N

It didn't exist, but those safeguards have been put in now, whereas, where, if the class is in there, it just simply returns, I, don't know anything. So we could just remove all week. Miss and manually delete them when, when we're done with them, another option and I need your expertise on this, but I think it's safe to call in to the vm from the from the set set. Second pass call back call so like on first GC or yawn first GC pass.

N

We basically call the destroyed call back through the second pass, call back, I think that'd be safe. No.

C

I figure right but I, don't know. Perhaps a little comment. I was under the impression that the vi team wants to face that out. mmm I, don't know why I think we're I think that early yeah.

K

I had the same deja vu, feeling I think at some point. They were actively trying to face it back, but I'm not sure if it's happened or if it's stone. The plan I think it's there for now, but we good to make sure that it's actually intended to be there going forward. Okay,.

N

That would be code code why's that be simplest to implement, because we we could also put it into a buffer, because, right now the users are allowed to know when a buffer object is cleaned up and, as somebody made mention an hour ago in the issue, modules like node week are now unstable because they rely on that same timing. So if you call a javascript function during during the time that note buffer alerts you to that, then it can fail as well.

N

So it might be worth also tweaking note that no buffer API to not call until the second pass call back to make that safe and then the last the last one is last option is just to throw everything into a UV idle tea and basically call all the callbacks on set immediate. Because I didn't I didn't it- realize this fully until late last week that even when you close a handle, it's still not called immediately.

N

The disruptor isn't called because the the class isn't deleted until the UV close call back is called, and I have no freaking idea. I don't know: what's going on with a vm and GC when that phase of the event loop is running, then maybe you can weigh in on that.

C

Don't know I don't know, yeah, don't know exactly what your question is: okay,.

N

So if, in some in some cases, when you, when you close a handle it will, it will excuse me, it will call the destruct ER immediately, which will call the javascript function immediately, that's safe, but in other cases it puts the handle through UV clothes and then the destructors called after the UV closed convex callbacks called and those callbacks aren't called until the last phase of UV run and I can't remember what that specific thing is called.

N

But at that point you we've completely exited out of all JavaScript stacks and I'm, not quite sure like what else might be going on at that time. Does that make sense right.

C

Yeah I think I missed and what you mean yeah, so it's from the UV close call back. So at that point, there's only there only see stack frames underneath the vm is in a consistent state. It's not busy garbage collecting, so it's safe to call in to the vm against them. It's only now yeah only with weekly persistent angels. When those are collected, that's yeah, that's the playing point right now: okay, great.

N

Then so those are the three options remove all week handles completely I use second times call back if it's available and throw everything into a UV, Idol t that needs to be thrown into UV entity. So.

G

To clarify for the UV Idol t option, it would only be destroyed, call box that would be thrown into it not before us for a minute. No.

N

No justice right and that's just the around GC, so only destroy and I. This should be documented. Destroy, is not always called the exact moment that, like the user, closes the handle right because we wait until UV has properly cleaned that the resources in the UV close call back. So we've never guaranteed the the timing on UV close we've only we've only treated the destroyed call back as a any resources. You have tied to this ID it's time to clean them up.

G

Okay, um so this poses I think a little bit of an interesting problem that we haven't run into yet with our current processes. So a sink rap is currently an experimental API, which means it could be changed or removed at any time by technicality. As far as my understanding goes, so we could remove that for now to make it not crash.

G

The question is: if that is a good idea, arm, I would probably lean towards if we really need to go that route to just making them know ups instead or that.

N

Would be more, maybe they would.

G

Be in that case anyway, because we're not trying to call it, I'm not sure. No, that.

N

Would be more dangerous because users, if users hook into it, they will be expecting it and probably using it to clean up like IDs attached to a map that have resources and if it's never called, then their process is just going to bloat and die I yeah.

G

I mean their process is kind of just gonna die anyways, though, because it's gonna throw.

N

But not at all cases like this is not every process that uses this will die I.

N

I have been using it a ton in its development, and I have never run into this scenario. I don't I mean.

G

I mean if we remove the destroy hug mmm yeah. I guess I guess removing the destroy cook would be essentially no op amp, because it, whether you call a user function or not, I guess.

N

The reason a real solution interests me is because this is, I have my coat up right now.

N

This is exactly what I'm coding to it: I have all the old stuff in then you brought this up, and so this was one but like almost very last thing to throw it up and say: okay, the final PR is done, but if we need to rush this in to fix these issues because the final PR is bigger or require more review, I would prefer that if it would feel silly to remove it, just add it back in and I'm fine one offing it and just getting it immediately.

N

I know how to implement any of these I'm curious from everyone. What they feel is the most appropriate solution. Well,.

C

If one of the options is to remove all week handles to me, that seemed like the best option. If only for the performance gains I mean we can you'll have more overhead than strong persistent handles. Okay, all.

N

Right I can do that.

A

So this I think was put on the agenda because there was concern that there's going to be a service stalemate where we're nine 467 wasn't going to land and something else wasn't going to land for a long time or something like that. It sounds like at least between Trevor and Ben. There's some agreement on what the approach will be over the next week or something is that accurate, yeah.

C

Yeah I think we're already pretty much an agreement. It's more a matter of time frame, I think so. My main problem is that people are reporting the reporting crisis which I would like to see fixed.

C

So we have to stop gap measure, that's basically just removing the destroyer hook, so there may be a better fix. But if it's not going to be tell you lies in the next step, say one or two weeks, then I would still push for removing it. Destroy you for now. Okay,.

N

So is the UM wins? The next release, weird approx, Tuesday.

G

N

By next Tuesday all right, so then what I can do is I can put one or two days of work into removing all the week handles, and that will give me a good estimate on how long it will take to complete if it's more time than I can get done for the next release, then I'll just implement the using UV idle see, that's a very straightforward fix, and I can definitely get that done before the next release. So either way we can have a fix to prevent crashes before next release.

G

Okay, I'm, just as a question like is the crash only when people use async wrap.

C

Yeah, but the thing is so all three book reports we've received so far from people that we're not aware that I actually were using a Singh rap, because it was a dependency of dependency that well that that used to destroy hook.

G

C

G

Mean that's that's a little bit tricky but I think like okay, if it's crashing on an experimental API, that's like a little.

N

Bit less you know, I saw just that. um Somebody commented in 94 67 just before the meeting began that they had been experiencing crashes in node week, module because of this exact same issue, whereas before it might have not been officially safe, but it was safe enough to call on the JavaScript from that waiting. I.

A

I think your minds referring to a sink hooks as the experimental API. Yes,.

N

I know I'm referring to whether it's crashing, if that's the only place, is crashing and I'm saying no, there is our modules in the ecosystem, where it's also crashing in terms of core.

N

It's the only javascript exposed API, but also the buffer API, where we allow the callback that is also a place where can crash, we don't guarantee safety of calling into JavaScript, throw.

A

N

A

uh Because we have a plan of action, I'd like to UM I'd like to, unless there's more, that needs to be talked about right now, I'd like to I'd like to propose that we move on just to make sure we have enough time for the buffer issue in the private segment. Is that okay with everybody, yep yep?

A

Okay, I will. I will note that someone made a comment about. Maybe if we don't want to get bug reports about experimental API is crashing. Maybe you know in production systems, maybe we can not turn on experimental things by default. I, don't know how I'll feasible that that actually is, but one of the.

M

Market idea, but.

A

M

I speak to that really quickly, rich really.

A

Quickly, go ahead to.

M

Ever wrote that I think it's important to keep those experimental features, even though we don't support them, they've been a way of finding lots of bugs they fix it so that when the new releases come out, we don't find those bugs then like this. So.

A

This is the system working kind.

G

Of yeah, that's it like the thing is is like we don't really get that many like bug, reports or stuff. It seems from RCS. So um if you want to do in like a minnow, that's only built by some releases I feel like we're still not at a point where people use those that much yeah.

A

Okay, that's that's! That can be a large conversation because there's lots of point counterpoint, so I probably should have not brought it up, but um uh it is time. Oh you know we didn't do we didn't go over the previous meeting, but we're just gonna. Let's just move along to QA on public channels. I believe is what we're up to now. So any questions or comments on the youtubes or anywhere else, Jeremiah.

G

William said that he updated the calendar, so I guess, like the new meetings, are on William copies calendar. What we talked about earlier so yeah, if you have any questions, put them in the youtube chat or an IRC, will be watching for a couple minutes. In the meantime, the next CTC meaning is actually reaching my no better when it is what time it is.

A

Oh yeah, four hours earlier than this one started this next cct meeting will be in one week from today, but starting at 1600 UTC.

G

And then the next toc meeting is apparently tomorrow at this same time as this meeting, is it.

H

Well, diem has moved up an hour, wanna.

G

Knows that so p.m. pacific, yes, diagnostics, meeting, apparently there's one scheduled for sometime in december. If you look at the node j, / diagnostics, repo on probably more information there, benchmarking there's a meeting I think that's this friday november, the 18th for p.m. eastern time- and I think that's it for scheduled upcoming meetings- know that many viewers, no questions. I think we can probably wait just a minute and then head into the private section.

A

A

It's like college radio, get air.

G

All right, well, I, think we a bunch of stuff to discuss in the private section. So let's move into that. Okay.

A

G

Everyone for watching, yes,.

A

Thank you. Everybody.