GitLab Pair Programming - Frontend, 5 Jun 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Pair Programming - Debugging memory leak in GitLab Jest suite

Description

**Result:**
- https://gitlab.com/gitlab-org/gitlab/-/merge_requests/33973

**Highlights:**
- 0:00 Eipi explains the problem and current debugging approaches
- 7:45 Eipi shows how he's been investigating memory issues
- 25:00 Eipi has to hop off but we were so close to seeing what was blowing up
- 30:00 Paul starts saving a memory snapshot locally and praises Demoscene while ranting about JavaScript
- 36:00 We start looking at the snapshot and aren't sure what "shallow size" vs. "retained size"
- 37:50 After researching we summarize what it means for us to have a huge gap between "shallow size" and "retained size"
- 53:30 Start on the fix with 5 minutes before Nico has to hop off
- 1:06:20 Summarizing our effort

A

Okay, so hi everyone, this gesture this session, you know as we come to the end of justice. Just the sessions become more interesting right, so today we're just going to look into a memory leak because.

A

Because the the problem is so what happened? Tim has created a merge request to become so that we become better citizens and remove emoji Jason from our j/s file from our manger. Espanyol fire, sorry brain need to sort things and unfortunately hit his tests. The new test that he introduced or refactored run into an out of memory problem, which is weird. You know at first I thought hey.

A

Maybe we had some borderline things, so, let's just increase the max memory, sizeof node, then, unfortunately didn't help, and so we still run out of memory and funnily enough, if I try it locally. I also run out of memory, so that doesn't seem to be anything see. I, specific I. The things that I have done now is locally I have started the jest tests, I had started the Justice, they run out of memory and so I restarted.

A

The just tests followed an article that said: hey, you know you can attach the chrome memory tools to jest and I basically took three memory. Snapshots and I started, comparing them and funnily enough, the biggest stuff doesn't seem necessarily related to the changes Tim made. So Tim you just find that one test in isolation I would have run all yes. So the idea is that I've run all of the tests in sequence, which helps with detecting these memory leaks, and it's just like that. Just is taking more memory overtime, so you can see it unfold.

A

You can see it right now in my logs, because I have deleted them. I have stopped the digest thing, but we can try it again, but my situation.

B

A

B

Absolutely I can we walk through some of these changes that that might also get everybody else looking at it some context, yes,.

A

So in these changes, what we're basically refactoring is that, instead of loading, our JavaScript, our our emojis directly via of epic, we are basically importing them reloading the JSON file. You know we have like this emojis that Jason and instead of loading it directly the fat pack and having it in our bundle, we're loading the Jason asynchronously.

A

If we don't have it yet, if we do have it, we put it into if we loaded, we put it into local storage and we load it from there. So you don't even need to load that Jason if it's not there right, so that actually means that we removed this so yeah it doesn't. It doesn't really have to do well.

A

C

A

So the base basic.

B

Basically, yeah I was looking at some of the other files. It seemed interesting case krola that that Awards handler one, if you go down so we're doing before we would import okay.

A

So we actually had the same same thing before so okay, so the thing is, we are currently dynamically importing emojis, which we probably could make soon at some point, but we're all doing that anyway. So here we just reflected a little bit when we init this emoji thing.

A

So yeah yeah. So it's it's. You know the fun thing is in the tests. I, don't think that this is causing that this is necessarily causing a memory leak. I think it could just make memory leak more apparent, because if we, if we look at the following thing, so let me find the tests spec emoji spec, so the big things yeah.

D

Can you enhance your screen and hence make it larger.

A

A

Para, better.

D

B

A

B

When he's like, yelling enhance or no he's not yelling isn't.

A

It like see his eye that they zoom in and then it's like. Oh, you can read the license plate and ass.

A

Okay, so the big thing that we're doing in this test that we're doing differently now is like so under the hood. The whole emoji thing works as a web component, which is unfortunately not natively available in our JSON version, because we are like 27 versions behind rights of yeah, Jason, 11 and JSON 16, which is fairly recent introduced a possibility.

A

So my assumption is that maybe this thing made the memory leak more apparent. Not necessarily you know, because maybe maybe just by registering a few more bites into J's Dom window- and you know now something not being garbage collected properly froze out of proportion. That at least was my assumption.

A

I've basically started just with this cool command, where you can add the chrome, utils right and then chrome, debugging utils, and then you can take memory snapshots and compare them. So the only thing that this essentially does differently and we could test it easily. If we just you know, X describe the emoji and we run the tests. It shouldn't lead to anything right.

A

But if we or we go back to the master branch and if you also see like memory there, it could just be that you know this is basically the drop in the in the in the bucket. You know that caused it to overflow, essentially right for whatever reason: yeah I.

B

Don't, let's try it I think that'd be interesting, is let's try running just without this emoji spec, because I think some of the emoji stuff is actually maybe globally loaded. So.

A

My test now would be so you my question would be I'm just going to delete the test file. That's one! If that's that's! Okay, but right, that's the easiest way to get rid of it completely right, reusing source control, I think so we should be yeah. What is this source control you're talking about? Well,.

B

I hear that okay.

C

Love like RM minus RF,.

A

So I'm just going to close this Chrome session here and start a fresh one and you know in order to save as time I will attach directly to the thing wow.

A

That's so cool cool, so you can see here. It's now saying: hey! You know you have to debug. Please run it see how to do that. So this is this. Is the command I? Don't know if you don't want to document it at some point, but you can see here, hey you have note just running right, then you go on inspect mode. You are here the bugger paused. So basically, what we're doing is a hey we're not doing anything right now.

A

So let's start it and then we can watch it for a bit and the really cool thing is: you can see here the memory, so you can see that the memory increases right and I mean this is expected and you can see like how all the tests are running and I mean for this analysis, and you can see that the heap size grows even without the emoji specs yay.

A

So this is a good sign that the emoji specs didn't necessarily cause this, and it should flap now my hopefully, hopefully it should flatten out, but just for the sake of arguments I'm going to take.

D

I'm going to take memory.

A

Snapshot which will stop the execution by the way right and then we can take memory snapshot in, like let's say, 1020 tests. When you see that the heap actually has grown to one point, five gigabytes, and then we can compare it, and my assumption is that it doesn't have to do anything with the emoji specs and that it actually, at least from the memory snapshots that I analyzed before, has probably done something to do. The vial of view view test utils or something how.

B

We try this on master. Just is just a double check, comparing it against master.

C

Lucas now you made poll not sleep for a couple of weeks until this memory.

B

Opened up drugs even more.

B

Now he's driving.

D

Glad to hear that we're not over talking about flattening a curve.

B

You want, if you need to hop off, we can.

B

Wait a second okay.

B

Yeah this one master and see.

B

A

Let's just double check to see if the, if this mi sorry, if I'm repeating something if this Emma's problematic or not, but to be honest like because I deleted the one spec that actually changed, I deleted the one spec that actually changed, you know it shouldn't be a problem right I mean.

B

But it's also this this production code I think this shared emoji code gets loaded, it's possible that he gets loaded for every test. So, oh, it could be the production game as well, because we have emojis. We need this me, oh geez, to run as part of our comments.

B

Oh look! It's it gets kind of going down. It seems like it's. It seems like it's hanging out. Yeah.

A

Let's, let's, let's have a look, I mean it would be weird every glass now I guess this can take some time. No I'm, not yeah I mean the thing. Is we don't need to run every test to see if this, like you know, grows out of proportion? Let's just let's, just let's just you know, maybe I'm I'm pulling out this.

A

So we can know what did I do. I want to have this as a separate window. Give me my window. That's the problem. If you're on Mac I'm, not on Windows.

A

Session to window.

B

Ok, yeah! That's how much memory it's using I, don't know if it's just cos get Jess is doing a lot in parallel. If JavaScript, it's just not that good of a language I.

A

Mean in each garbage collection collected memory, language, you can have memory leaks right and you know, starting to dig in I, actually found that you know there have been memory leaks with just in the past. Just it's actually really good that it has like these tools in it. Actually, you can see we are probably leaking memory because otherwise, there's like no, you know we're increasing from 120 Meg's over there now at 930. You know, you know, I, don't know I'm Smith.

B

A

Makes this is this, is this? Is the saddest, so I mean I cool yeah I can actually force a garbage collection right. Don't do it because I.

B

Not you can't just see it go all the way down. Well, man, we went over, we hit a thousand oh now, it's down wow force it yeah, because I 400.

D

We just use a. We just have to create a script that clicks that button halfway through this problem. So right.

C

So while this ran, they usually the culprit for memory, liquor, single thongs and stuff attached to global buses, in this case, probably with a window object. Yes, you think we could have a did any of those in the mr that tipped the scale like that as a last drop. One more single tone imported, like Axios but I, don't know axis- is not 100 megabytes to.

A

Be honest like it could, potentially, because the emoji specs also ran as the first ones right, because the last ones changer I actually don't know how just the terms determines the order, but I assume. That was the reason, because the last time I run it. So it could be because this global document registering thing is like I, don't know currently and that's why I wanted to bring it here of some just experts.

A

Do you guys know if we have like one Jase Tom instance, that's reused, or do you know if we like you know, because this could be related? If we put something into group into the global DOM and yeah.

B

It should not be it's my understanding that it is not reused, it gets set up and the test set up and that's something that happens. A test setup happens but happens just once so yeah.

C

But can you be something that we created.

B

We somehow managed to I really want to take us on a master and see if, let's, let's see if we can see if it's tied to this mr.sun I mean yes, I'm, really frank, because Jess doesn't load things like web pack does, and so jess has different third-party libraries for reading some of the things and we're already doing some really special importing here that maybe we just triggered something that's making one of those third-party libraries just upset I mean that.

A

One of the one of the interesting things and I think we at some point we need to they just anyway and I think we should update. J is dumb because we have like they actually using this pulley for for the emojis right. So if you, if you look here, we use this document dressed with just the element. It's actually you know like, and it's actually using like you know, extending HTML element and then you register at the bottom.

A

You register like this element, so we have like these gia emojis right, and this is just like the polyfills for the browser that don't support. It actually got a question which browsers do support it and maybe we could get rid of it. Dropping ie, 11, never. Nevertheless, I think the problem might because I tried, you know, be like hey, you know, I, don't care, I, don't care about the underlying logic.

A

When we unit test the thing it's probably enough to test the class, but the problem is I moved the class to a different file, important the fire and J's Tom just dies because it doesn't expect you to extend the cheer emoji. The HTML element right, a newer versions of J's Tom, don't die, but the problem is that the newer versions of J's Dom are not currently hooked up to our Jess stuff right. You can't just update J as Domon. Fortunately, you know so it goes hand in hand with just upgrade and I'm. Just like.

A

Oh me, I'm, actually hopeful that this memory is not necessarily connected to this, but that we have like memory leak come that we just have a memory leak. That's you know related to something.

B

Now, there's a there's a huge win here: either way: yeah I'd, like I'd like to try it out master I, can try it out on my computer. Can you share with me your your sweet.

A

Yes, it's actually pretty just memory leak. You just tests are leaking memories, so this is a really good. Where is our thread? This is like a really good. Oh man, it's over.

B

It's almost fifteen hundred megabytes now yeah.

A

I mean another.

D

A

We can take a snapshot and can compare them. That's again, good.

C

A

And actually you know this is maybe the final thing that I would do and then hop off. How long has suggested a session an hour and maybe maybe you could get go back and you know use the last half hour to look into a test or something sure right yeah, but.

C

Now we are very curious about this, except.

D

For you, you know, I he's.

B

Like hey, you know why you guys are all migrating these specs. Well, it turns out, like they're, all unfocused.

C

About their before oh.

E

A

Yeah snapshot, I need snapshotting. The problem is with one and a half gigabyte. It also takes some time and zoom share on everything right, it's a little more than that right, cuz, it's gonna, be.

B

Because that's just the raw data, it's 1.5, but it's can't do a bunch of metadata around it too and stuff I guess. So what do we have here? Arey? Oh, my god, let's look at the numbers were raised. We had, but it's like clearly something's not being cleaned up like why.

E

Is that, why is so much I can't wait to see what it is? It's.

D

All those people who aren't saying their rapid or null.

B

Oh man, if it is that we're not Auto destroying you know a later version of utah's details, Winnie wrote a helper method and published it upstream. So we can enable auto, destroy for components and view ties details, but we're kind of Tokyo.

A

To be honest, like the earlier snapshots that I had actually had like circular watching component shenanigans, yeah, yeah and and I mean, if you, if you search for like view test you Till's memory, leak, you'll find like issues right ah possible memory, leak right and it's like you know it wasn't this version and could it be related to, and just like you know, it's like hey view. Router could be responsible right and.

B

I think I think one thing we could do, though, is is be a little systematic with isolating this I mean what can I snapchat be super helpful, but we like, if we think you testicles, might be offending.

B

Let's, let's you know, remove all of you test you till scripts and then review test, details, facts and then rerun it and see. Are we so leaking and.

A

We only see me.

B

A

Mean that's the thing that you would see in the snapshots right, that's right! Okay, if we compare snapshot to which snapshot one and we see hey, you know if it's if it's like the strings or whatever that are leaking. If they tell us the story like hey by the way, my name is exe Azad mock request, then it's likely related to the excess mock requests right, yeah or if they, because it was like objects that contain watch ah and depths. And it's like this looks like you.

B

This this whole process the script and opening up this Chrome, in spite of like this, is really valuable. So thanks for your thanks for bringing this up, I'm excited to.

C

Scale, this is make my label to look like a two three test and see how the memory grows there, because it's faster to integrate on top of that right. Well,.

A

The problem is that you don't, then don't necessarily see what's leaking I mean if you have like you know, if you, if you, if you see here that, like you, have an increase, it could also just be because it's not garbage collected right I mean we.

B

Bring up you bring up at the point Nico and I. We used to do this for like capybara aspects that were flaky when our like our specs that are flaky you its unfeasible to run the entire r-spec suite locally. So what we do is like select half or like we'd slight like a quarter and say: hey did this cause the issue and if it did, then we try to slice it down even more to like, let's try to find the.

A

B

Set of tests that reveal the issue and now, okay, now I, have something I can play because yeah I think looking at this, we can see you know in just one page. We went from one hundred to eight hundred, and so I could just run just these specs. Yes, definitely I.

A

Think I mean the problem: I, don't know how the garbage collection yeah it.

D

A

Internally right, so it could be that you know note similar to like Java or whatever that the node program is now just you know, taking like old Joseph.

E

A

The memory it can and then, if you hon, add that to gigabyte whatever so well another another thing that you could do is right. You have this max memory, setting that we've actually changed in the tests oops. What am I doing?

A

What? Oh god, my computer, so there's like this mix memory test a mixed memory settings, so you could, you know, turn it down as well right. So, instead of saying, hey here are two gigabytes or usually on my computer I. Have it set to four gigabytes, so this is like three-and-a-half. You could be like hey, you know, just you get half a gigabyte I mean you should be happy with that right. Yeah.

B

Yeah I can't I can't wait for us to resolve this problem.

D

Yeah, did you upload the snapshots to the issue that is tracking this problem? The memory.

A

Snapshots yeah, it might be too late. Can you save them should be no problem to upload them because, generally especially, if you have a memory leak, it's very similar object. So I had it in the past, buried memory leaks and like the snapshot was like two gigabytes, and if you ran zip on it, it was like hey, ten megabytes, nice, that's awesome, cool yeah I mean you can upload it to drive anyway, for example, right scary.

A

You know, you know what folks I have my guests over it's nearly the weekend you can share the is that is this: the script that.

B

A

Look at the thing at literally this described here and just a command. It's just I just copy pasted that command I'm, not that smart I, just copy paste things yeah.

B

Okay, the leak yeah locate the leak. Okay.

A

The leak it's cool, usually usually in the ground next to the potatoes, but.

C

E

C

Date: joke your specialist. huh Yes,.

A

Yes, so they describe what happened: yeah, I'm stoked.

B

About it, we'll hop on it, thanks IP for being so cool yeah.

A

Cool I mean check it out on masta and see if the. If the thing increases.

B

I think well: yeah I. Think. Can you share this, mr and this article on slack and then yeah.

A

Already done wait, let me just ping, you Paul someone.

B

Wanna does someone wanna pick this up? You know where.

C

Someone wanna pick and if you don't want to drive otherwise I will give it to you. If you are particularly interested.

B

Yeah I'm interested I, don't mind driving, but it's also my name.

C

B

That my I haven't been on fire these this morning's these this week so I. If you're feeling fresh. Maybe you want to pick it up nico.

C

My only thing is that I have are to stop at five exactly and money then I may I may.

B

Do it cuz I may get really into it. Alright,.

A

D

A

Is almost done, I think.

D

I'm, even earlier than you are okay, I know.

B

Come on, oh, it's gonna, take some time IP it. Why?

B

B

A

Okay, all objects give me those objects, those sweets. You.

C

Gotta wait. No! Oh.

A

A

Anyway, maybe not the smartest thing to do it on zoom I as well I mean I also killed a just process. Maybe I killed it too early.

C

Do you think it's really TV should I think we should work on a smaller scale, otherwise,.

A

We'll do that alright.

C

A

C

A

Fun I have a good one, have a nice weekend right. Thank you. You too, bye right. Alright,.

B

Here I'm doing something I, don't think I'm doing anything important, so I'm, just gonna.

C

They work or you're doing yesterday's trash now, yeah.

E

So I uh and here's my slag.

B

Here we go okay,.

B

You see my dear.

C

I'm, sorry, that's the trumpet bother you Oh does what bother me the trumpet okay? This means that you can't hear it good. No I can't I have a musician neighbor that rocks No.

B

So is it, is he or she early in their trumpet career or yes,.

B

Well, I'm sure that the sound of progress is inspiring headphones.

C

Noise-Cancelling those musics so.

B

I'm not really sure about what did you hear other than best accommodating too right? Yes, does the commune so we're running this thing and then, let's just identify a subset of specs to run. Does that sound like a good idea?

B

Okay, so I'm, not even gonna inspect this I'm, just gonna run it and then I just see it grow.

C

B

C

Chrome no, but he didn't block. So we miss a comment: okay right.

C

D

After node in that command, you have to type this. What I am about to put into the chat here or this one there, this full command.

C

B

Same one right, you.

C

Have some respect P expect BRK.

B

Since the same train ran after nerd, oh, this is a oh.

B

I clearly didn't copy.

C

Then you're saying I need to open Chrome and then do you need to go inspect element.

D

I, you can just take Chrome colon forward. Slash forward, slash, inspect II you, mr. e, oh gosh. This is what I was talking about. Okay, okay, you warned us. You know here inspector.

C

You can also like right-click inspect, and then there is devices. Okay, the console as well type words were debugged.

B

And we're just gonna hit play yeah.

C

B

We're doing some things and.

D

This is exciting yeah. It's.

C

The point it's.

D

Great, how, as you were, saying similar to the trumpet player like how much room for improvement, we have yeah.

C

I don't know this is not my first ride with memory leaks, usually the the motion actor is like frustration. Frustration, frustration, a penis rage rage is when, when you're identified, what happens? What caused it look.

E

B

At this look at this.

E

C

That bound right.

B

It's oh my gosh, but we always have.

D

Maybe good says you know which ones are bumping the hardest. Yeah I think that's a good idea. I'm gonna! Well, I! Guess we're recording this. So, oh! Well, let's find.

B

You recorded right yeah.

D

We can scroll through and just be like. Oh.

B

What's this team doing yeah I'm clearly missing some pictures, maybe.

C

Turns out that mound is memory leaking oh gosh. I won't be surprised.

B

But this one clearly well, this is running well, we're already only got a thousand.

C

Okay, like actually looking at those longer education desk, can be a nice deep like those red on the number yeah, so we're already out of thousand.

B

I'm gonna take I'm gonna, do something. What should I need to go memory memory, take a snapshot, yeah.

B

So I think you and I talked about an ego but like I'm, really inspired by seen how we talked about demo seeing Alexander. So these are like amazing.

E

Feats of software demo scene yeah will.

B

Create all these, like beautiful demos, these very graphical demos, but you can only do it in like four kilobytes people have done like amazing stuff in small amounts of space, and so then, when I see like you know,.

B

It hurts but I know it's also, not us it's because the whole JavaScript ecosystem is and I'm dancing now, but it's like who knows what code is running like anyways, I'm, Superman Singh, so I'm gonna also I mean for.

D

The most part, who cares it's only it's only when this happens, that we're like you know what? Maybe we should just use all the memory I mean like dicen, when your chrome is using 16.

C

B

Yeah bring up some good points and I thought I think. But what I think happens then, is that since we don't have any visibility, it doesn't become a problem. So it's really too late because it's hard to fix these things. After the fact I mean I'm a I'm, a huge fan of like we should try to make sure things run on like fairly constrained machines.

C

Can you okay, just out of curiosity, can you check the import of that spec from ten minor components, four spec, because it did a bump of 200 megabytes when, under the matey, this ten seconds.

B

One yeah that oh, this is happy the each hearts thing maybe because I saw that show up because here's the thing like it's gonna load, not just this stuff, but also the all the things I could load it here.

B

C

B

Never clears like that's really weird and.

C

Scary, so he just goes up means that those my friend stays there forever right. I.

B

Have no idea it doesn't feel good.

B

Yeah we could, we could run this one and try to isolate.

B

Nicely what's happening, I kind of wonder if anything's gonna be so that's another interesting point to that. What Lucas was talking about that this might not just be related to the mr, the fact that yeah we have some specs bumped more than other specs, signifies that it's probably not related to that. Oh, it's probably a lying issue for sure, but I I wonder what it is.

C

You know what they were thinking. Sometimes you can console.log in a test like a component body and it will crash the test because it goes in infinite loop like it consider liza an endless cycle right.

C

Could they be related.

B

So view- and that's that's interesting point, like the view object, has a lot of those references like you have a reference to the parents and the view object and the parent has reference to the child. Yeah.

C

But that's I think stops of body right.

B

Well, I mean the parent, though the parents view object, has a reference to the child and a child via logic has a reference. So you yeah, there's.

C

A few instances by Nature this also means that if you forget to destroy one, you will actually hold up a full three yeah.

B

C

Nothing is going to be garbage collected until there is at least one reference holding up there.

B

Yeah yeah: let's see oh here we go alright. So let's see we want to look at so.

C

What I don't Oh, what I say? What's that seeks me in there this.

B

E

B

Shallow sighs: this is starting to get a little bit beyond my oh, my gosh wait.

D

But the holes like it with the shallow sighs it says four percent. Next to it. We should we should know. If we're gonna look at any of these sizes, we should sort them from the.

B

D

Mean like does that mean I, don't know what the difference between this. Do they well I. Think the a good question is: do they correlate like if you sort by shallow, if you fill or sort by shallow size? Are you similarly sorting by retain size? Do they just happen to work out together? This.

C

Is a really shallow size? Is the object itself right and the retain is the with all the air katachi yeah. Let's do a quick, oh yeah. Someone should be Google.

D

A shallow size versus a second there you go.

B

Memory terminology.

C

B

By the object itself or implicitly by holding references to other objects preventing those objects, okay,.

A

B

Of the memory, those home.

E

B

Get it self o retain size reference.

E

B

Once logical self is the lead it along, along with its dependent objects that were made unreachable.

B

There are lots of internal which are not interesting, I'm skimming this I know.

C

So basically, a big retain size. It means that this object is connected with a lot of heavy stuff yeah.

D

C

Shallow strides, it means the days the guys be going itself. Yeah.

B

And that kind of makes sense, arrays I'd expect to be pretty big. This is interesting that there's this huge gap between retain size and shadow size, meaning from what I hear, is that the actual data of this individual object is small, but it has references to things a lot of things that other things don't have references to. So, if I delete to this this object, this is how much win I would get.

B

C

Not just related I think in simple term use on it's.

B

Like that, but that I I don't.

C

Really know what I'm talking about I mean. It also makes sense right because all our components, all our stuff, are all objects at the end of the day. Right, oh and so look at this! Oh no! This is oh.

B

C

My gosh I wonder if that mid stuff is causing issues. Well, it's just just road bike gate before we introduced meat and just run a quick. That's a really good idea. Just around 13 13 0 yeah. Can we check.

B

C

B

Gonna write down that idea of rollback before med and see what happens, but I want to double. Let's do a little bit more exploration while I'm here is that ok, sure.

D

B

D

B

Window up a little bit more yeah, but but but that's so here's an interesting thing and yeah these memory leaks are so real.

B

What can happen is because I create all these closures and these events, but I never I, never unlisted to the events. I've now have these objects enclosure and they never get cleaned up, and so that's my my size of these event. Listeners are so small, but they have objects. They have references all to lots. Lots of objects that aren't are no longer even directly accessible but I. Never unless it's to the event.

C

It can be, it can be, but don't we always call the like off meet, for example,.

B

Maybe we do but I know.

C

And you know, actually this doesn't even matter in actual code itself, because guess is loading everything all at once. Well for us, it's all web pages right, so this may not even happen this much in the production.

B

World but but because jazz is it's running it on every spec, every spec we have a brand new kind of environment so.

C

That's why I think it could be very problematic, but if me Jesus is important like the normal meat and meat is a singleton, it will keep the reference in memory of all the cold of all the tests. Yeah. Let's see.

B

And maybe we could use add, maybe there's an easy fix, maybe there's like we could do or on our test setup. We can do after each just get rid of all listeners or something like that, but I I.

D

Don't really know I'm talking about. Can we look at a few more of those objects we so we've looked at two that have met. Are they all you.

B

Know I want to look at a few more of these cuz. That's top one is mid, but you see that doesn't totally explain this huge 600,000 I'm trying to see what what this thing is and has some exports and oh gosh. Oh it just keeps going. Oh gosh, some self.

D

What visibility is the library we use for polling, so it basically tells you if a tab is in focus, I.

B

Wonder if this is our I wonder if this is our actually our polar or maybe some I'd, no idea.

B

I'm not entirely sure what I'm looking at or what I should be looking for it, except they see that something's yeah.

C

You could open it also inside up inside the tree in the constructor viewing which one up, if there is arrow to open up there as well double.

E

C

Yeah, if you click the arrow, you can open up that. Oh.

B

That's so much more helpful because.

C

We can see what is the source right, so this is visibility. This.

B

Is the yeah? This is actually visibility. We am for it. Okay, sorry, so use this quite a bit.

B

And yeah we, where we're, adding a listener here, we're not cleaning it up and we have I know this is like a singleton store to like so not super surprised, but this is EVB DJ s, right, yeah, I, guess I'm guessing. So that makes sense. This is a nice-looking array of like one of the best.

B

This is a pipeline thing. This is an army.

C

But you know the sites already went down a lot. This is the third one. I know no yeah apologize and.

B

Then here's our here's, our poll or I- guess I'm guessing. Oh, it's like no. This is for the MOR widget by itself, because we I guess we're not cleaning up after that. I have no idea what this is. Oh, this must be a jest thing yeah or like babble or the Jets motor or something it's got. Your name in.

D

It Hall, oh man,.

B

Good thing I don't have you're holding too many references turns out pause. The source of our memory.

B

Yeah, these look like just just stuff I.

C

Mean the shallow sighs. Now it's nothing. We shouldn't even be looking at this right, yep I agree. Then here's.

B

A array of interesting things: gosh I, don't even know what I'm looking at each arts, but it's each hearts and it's a few component. But this each one says sorry.

C

Like a, we have just mean I seen kings. It's all there.

B

Okay, for the sake of time, because I have not entirely sure I'm looking at I'm gonna save this I'm, not gonna, put on my desktop because I know I'll delete it and forget about it. I'm gonna put it in here: okay, but I like the idea of let's run a previous one and see what happens and see if there's something introduced within versions and so I think I'm just gonna run because we did seem. It was like.

D

Yeah, you know, you know what library right here doesn't have memory leaks, karma.

E

No you'd be surprised.

E

C

B

B

Mean can you just uh well, I was gonna just get the get to commit before this one yeah.

C

B

That out directly but I wanted to did I already copy over I. Don't know what I'm doing here, what am I doing here, I just wanted to copy over the files that we ran this run, but I can also just run out everything. So a.

C

B

C

Here we're doing the same mistake as ap. You know so.

B

I'm committing before.

B

But what's probably gonna happen is now it's gonna, be view, that's complaining, but potentially, if you doesn't complain because we were still we cuz, here's the thing: it's not like a library as a memory leaks, it's our code has no real accent. No matter what the library is. Probably problems. Noise.

C

How we use the library- maybe that's the reason, and if we and can find a pattern, maybe we can apply like a catch-all solution, yeah and I I think that'll be a good like after each we global taught me tickle another something I, don't know like clear, not destroy. Whatever its method. The library has as.

B

You suggested I think that's a good idea. Civil straightest out without made see what happens. Oh.

C

I gotta go back to here, I'm afraid you need to cool here, I'm afraid that it's gonna be we're. Gonna, have a series of offenders, not only one I'm, afraid, I I, think you're right geez, but.

B

This is kind of thing. It would be really nice if there is a way for us to get some visibility on the amount of memory taken by just.

C

Well, you know performance easily, seen our career metrics past seniors. So it's important to solve this and.

B

Now it's documented that you.

C

B

Is now at shallow and.

D

You're never gonna make staff that way.

B

All these are phonies friend and pair programming. Videos is like plots like he's, really posing a lot.

C

Mr widgets, thank you. Take some life. Can we open just out of curiosity? What is that guy doing there? It's a huge one, damn our widget to 85 seconds yeah, it's it's.

B

C

One, how does it it doesn't come out for you, because that's.

B

Nick we've said I think we've mainly over it.

C

And the area I'm making a Martha speed stuff to not have anything above six seconds. That's good! Do it.

B

C

I have capacity, don't let this discourage. You me go I'll, try and see, but.

A

C

Done with under light done, oh.

C

Because now we are leading against it, so this was before probably right. They called before yourself I'm, noticing less increase right yeah. It seems one megabyte no well, it depends, but by that time in we were right here. I already had a gig before now. If.

B

We could still get to a gig something so.

B

Wow these these big ones, they sure, do jump up a lot and it just stays around ah that hurts Wow yeah. So.

A

B

I think when we I think when we do it, we're gonna see instead of MIT I. Think we'll see that it's, because we were using view instances to do this. I think we're gonna see that it's a good view instances so bear with us what might be a nice fix and I like it is an after each cleanup of MIT events and view events, and we should probably be good. Okay, we've already, we've already hit it. I'm gonna take a snapshot, I guess, but.

C

Could you stop it do I need this I didn't! Stop it? Okay, okay, good! That's what I was worried so.

B

The weight was.

D

Just on master so.

B

This was on mass. This was on an old master.

D

Pre-Med master yep.

B

So and we still we're still growing some questions, I have- and this is me being paranoid like- is it possible that any of my commands here, like cause memory, leaks like the way I'm calling it no I, don't think so? Okay, because I, don't I, also don't really know what these mean either.

C

Right I see: okay, yeah.

B

That makes sense now that yeah, that makes sense that's what running band means is. It removes the parameterization okay. That makes sense.

B

So what I want to see is we go to our our mid thing? Okay, if there do you all know if the MIT's library, NPM.

B

Has like a like you.

C

Want to try to compile, search all this or something. How do they invited Milwaukee service.

B

So remove all the centers as a thing.

B

I'm gonna, instead of look at this issue, I'm like yeah.

C

Check the API API babies read a book.

B

The API is o is right here, yeah, thank you, yeah, um remove an event for the given type type of event under faster.

C

B

Yeah we can just doom it off star, so what I'm gonna do this I'm gonna have I think in our Jess setup we'll have an adapter for emit that.

B

We capture the instance that we created it and then we just remove all of it.

C

And then you can compare the EEP size snapshot and see if it actually did the difference, but I think this. This will require an epic at some investigation. I think you did a little experiment here. You want try to do just a little experiment. Yes, but just have five minutes before the next meeting. Okay,.

B

Saturday I'm gonna really try to do a little experiment.

B

Okay, so here we are in the new branch, something something died over here, but it's still got a nice naturally think do we want to look at that snapshot? Oh No? Maybe my snapshot doesn't finished yet, oh no, it was.

D

Doesn't really sound like a hundred.

C

Percent, you are loading and the percent is actually like 20% and then the rest is good. You're gonna wait for it. There.

B

Yeah that might not even be working right now: okay, I'm gonna, try to get you out of design. You go so I want to import. We have a. Where do we do the important it now we have this thing? Yes, yes, yes, okay, so I can I.

B

C

B

C

You should import that one right, no Dave library, one okay,.

B

Sounds good, do we don't yeah? We we do everything from this event hub Factory, yeah,.

C

B

Right so this isn't job this isn't helpers and then factory so I'm gonna.

B

C

B

Want to replace, because I have to I have to capture this instance. Yeah.

C

So your imported.

B

So every time this is called I need to OEM.

C

B

A reference to this instance yeah and what I'm realizing now is I might need to actually, as I was trying to like I guess mark this I guess I can just do that so I'm just gonna go, can I do like this. Am I doing something wrong.

C

You can actually pass a second argument with the function that does the return of that yeah.

B

But it's I can't one weird quirk about these callbacks. They don't have access to like previous variables and stuff. Maybe I do.

A

B

Each year or something I think if I do this, it will just give me a mock implementation. I think I can then do then have factory mark implementation. I think I could do this, but if you also want to require the actual, do you know how to do that? That is.

C

A like a require require the chassis with my actual yeah, something like that ice in the API yeah.

B

C

B

To call this thing get the.

E

Faster, why is this having so many problems, but.

B

What is your problem? I'll use, constants that come after each.

B

C

B

Ex-Officer, better.

D

Right yeah, is it the string star or is it just the star I guess it's.

C

Obvious string.

B

All right, thanks Nico for hopping on, are you okay,.

D

Oh I will stick around for the end of the experiment. All right, you know, I just wanna, see I want to see if this helps so.

B

This does it look like I'm missing anything, it's very messy I know, but.

B

Yeah I just want to see if it helps see what happens.

D

Yeah, if I dropped off now, just be you recording yourself like.

B

Man, you know Paul Paul host. These paper give you things he's the only one that shows up.

D

Yeah their needs that we see a pair at all time like.

E

I think you're kind of missing the point here. Oh.

B

It's stuck, oh my gosh yeah. This one feels what well so you see like up here. It did this. You know it did it had a native failure, yeah fail trying to save something. That's disappointing!

B

Oh so here I am gonna inspect this thing. It's unfortunate.

D

That we've we fixed the leak erase. Well, we won't have another opportunity, I guess snapshot. That's true! Oh.

B

B

It's so knowing I'm glad that there's not all like. That's like blowing up I, don't know if oh gosh, oh look did you sit.

C

B

It hasn't ever done that a lot of these are failing because I don't oh, no, this one failed because they don't have. The real factory is not functioning well.

B

D

What does the was the memory tab? Look like heads up in in a second here, my partner is gonna, be walking our cat or on the the apartment. Oh yeah,.

B

Oh you're good. Do you want me to like stop recording green, no.

D

I just want giving you a heads up, there's.

E

Someone behind you and they're being followed by a cab.

B

Yeah, so this is still blowing up and I'm, but I think part.

E

B

Think part of that is that we don't actually use the event this mid one as much as we think it is as much as we think we are. If I look at the important.

B

We're still also using the view thing alive. Oh, we.

D

How are we using mint so much that thing just got introduced like a month ago or something I think we had a hackathon to start introducing.

B

It but some of these are not a lot of these looks like they're. It looks like it wasn't working for me. Well, okay, yeah. It looks like it's not necessarily working for me. Oh yeah, real might.

A

B

To be a default that might be related to it all right, the require import thing. I could try that out, because because the situation is I do bite import here, but I've mocked it I'm, gonna get the mock here. I won't have a reference to the real one, ideally I'm, not sure, even see doing all this mess in this file probably, but it is encouraging for me to know that this is at least is getting our little helper.

B

Things actually is actually happy here and quite a bit too yeah, so I'm just gonna run it and see if this aerosol but I kind of think we need the same thing for view. Instances like on our old event objects.

B

So if I do something like forgive me all files named events, but that are importing view.

D

There's only like one of them yeah, that's Wow, really.

B

D

D

That that's see, what's me, that is what MIT is.

B

D

Supposed to be replacing yeah yeah.

B

Yeah I'm surprised that there's not a lot of them, but these are like commonly used ones. This is notes and boards likes for that. Mr widgets, like this is gonna, be a commonly used. One yeah.

A

B

Just replace it with MIT now: oh, it's not running cuz I'd, okay, I'm! Actually not gonna, inspect break cuz I! Don't I! Don't actually want to do that right now, totally.

D

B

Wonder if I can replace these with men, and now we have our mid clean up and see if that helps yeah.

D

Yeah, that makes sense I.

B

Don't really know it's our pattern here for.

B

Say we just that's it sorry that my friend and fixtures I mean look at this see. You know yeah.

E

We weren't seeing that before yeah.

D

Yeah and I believe Mr widgets, like like Nico, pointed out, was sort of blowing up before and now it's reasonable look.

B

At that, this might be it look at that yeah. It said we're still hanging out in 100 that.

D

Is you know, I made the joke? Good I was like oh I P. You can leave we're. Just gonna fix this for you here we are that's awesome.

B

Something's having problems, oh gosh,.

B

Yeah there might be some situations, there might be some issues with yeah. Look at this. That's exactly this is a hundred percent. While we need to do okay, well, I feel good that we have a. We have something. This is gonna make our test faster. This is exciting. Yeah.

A

B

Those are gone yeah, especially the ones that failed. They just fail right away.

B

Yeah I'm excited about this and I kind of wonder. Then we just need to replace all of all our other than tubby things, and now we have a way of doing these. Cleanups across test seems to be working pretty somewhat good. um So, okay to marthis we'll see what CI says.

B

Do you have any capacity to maybe take it across the finish line? I.

D

Do not know okay, I could all right also.

D

Not a hundred percent sure, what's what we're doing here so I, don't think I would be there right here. I.

B

Pull there this.

D

B

To me too so, but the main takeaway I guess so. To summarize this video then I'll make it my.

B

The main takeaway is previously what we saw in the snapshot. We had an object's actual size of like 20 bytes, but it had references to like 600 plus megabytes, so something ridiculously and we dive into it, and it was this mid stuff and what was then clearly happening was that we only had a very. We only have like on hand of actual real pointers that are being used by minutes. But MIT has all of these closure references to huge objects, they're being referenced, and these callback functions.

B

So we need to we needed to clean up net and I guess we can't totally trust our component destroy to do that. We need to do it here. We need to do at the component as well, because that helps things in production. But this was the thing: that's cleaning it up. It looks like it's not that bad, but clearly we can't just mess up his test setup files.

B

We should figure out a better place for this to live, but then the missing bit is if we do import view from these kind of in hubs, they're not being cleaned up. So, let's also do the event hug thing here and then we should be golden and yeah look at this. It's looking a lot better. It's still growing, which I'm sure that there's other issues like with the visibility thing and with our polar callbacks. We need a way cleaning them at the application level is gonna, be really difficult.

B

It would be nice to just have a catch-all of like let's just clean everything, every reference that could have ever registered anything that's cleaned up, but yeah cool, hey thanks for thanks for helping out and hopping on out and previously neat going at an IP yeah having Genki. Thank you for saying these up. These are great cool, all right! Well, catch! You later see you bye friends, bye, I, don't know how to stop recording.