Node.js Benchmarking Working Group, 28 May 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Node.js Benchmarking WG Meeting - May 28th 2019

Description

https://github.com/nodejs/bootstrap/issues/114

A

So I think, though we can take I will take a look at at the minimum. We should do is set up a job, so it which runs once a month well beginning on the end of the month, and we do I just run simple script, which runs everything we capture the data and we now think about how do we present it as a charge? Okay,.

B

But, given that the script will run for several days, what kind of a what kind of a load is that going to impose on the infrastructure right, because meanwhile, we still need to do testing and PRS and all that stuff right. We have. We have two machines: Intel dedicated for.

C

Benchmarking related stuff, oh I, see.

B

C

Elwood all else they're used for but I think that it's fair for us to use a fair amount of cycles on at least one of those for that kind of stuff. Oh I.

B

See, okay, okay, so we do have a machine for this. Okay, yeah, all right, yeah yeah! That's, let's see! If, if okay I don't know, I can try to look at it but I, it's a very, very, very weak promise. Just so you know well.

C

I think anything.

B

C

Was gonna try and do take a look at that one just running everything, right, yeah and then, let's once we have it running you can. We can think about well like how do you actually display the data right, yeah yeah, yeah yeah, because we need to record it yeah.

A

Yes, so going to from there running everything running every nightly, so it wasn't a discussion about what benchmarks they want to remove in QC. It.

B

Was I actually I actually pasted correct me if I'm wrong Michael, if I pasted the wrong PR accidently but I, think I got the right one, it's it should be in the chat, so so so so that there's yeah. So it's fairly extensively discussed as to which benchmarks to remove, because it looks like some of them may actually be like duplications.

B

So so so they they benchmark the same thing over again with like a slightly different twist or something like that, but but either way Reuben is is is proposing that some of them be removed because buffer takes a very long time to bench all right.

B

So so, if, if you Tom and Octavian, if you could have a look at those or at this PR and and see, maybe maybe make a comment or two that that would be really good for a PR, yeah, okay, so yeah, that's basically the one that that, if it lands should greatly reduce the buffer benchmark run time. But of course, correctness has to be preserved and coverage has to be preserved.

A

A

Okay, I think it's okay, I did a lot of calming so yep.

D

C

To update those at least one of them, unfortunately, I was a little more worried about. Let's see which one was that.

C

So there's some discussion like basically the one that changes at one of the dependencies completely I think at least needs to be tested. So I was trying to explain a little bit to rich how we might test that one out do.

A

You know, do you know.

C

Yeah, there's 273 I'll paste, one of them, but it's two: seventy three or four five and six.

A

C

Probably the first, the two that don't look too scary to me: I'll, probably just when I get back from the collaborator summit, land them and we'll, hopefully nothing breaks. The other one it sounded like rich might find some cycles to actually test, which I think is the.

C

A

Okay, so he objected the the package and my sequel version.

C

Yeah, like the thing is the phantom Jas is actually like a different package.

D

C

That one, like you know going to my sequel, it's like, hopefully it's ember miner. It doesn't make any difference, but I make things so.

A

Local machine will it be able to will I be able to generate the charts locally, not as.

C

Long as you have the data like, you need some data from the database as well. Yeah.

A

Or maybe I will just run on my account on our No on our infrastructure and see.

C

Why yeah yeah I mean even like rich, has cuz he's part of the build working group? He has access. I mean my my suggestion is the easiest way to do it is to log on to the machine run. Chart Jen everything works, then there's actually a job that will that updates like a cron top job.

C

If we just landed it ran that thing, so it updated and then Rancher can charge and again and made sure that it's still right. Okay, then it's probably fine right and if it doesn't well, we can back out that PR from the repo.

C

That's the probably the easiest most efficient way to test it. I mean oh, this would be like you could duplicate that whole directory pull in the changes into the directory. Right and again, you know validate that it works.

C

Okay, so it's kind of couple different ways, but I think that one at least we should have liked as opposed to just landing and hoping it works. We should actually have somebody run. It say: yeah, okay, it looks like it was: okay, okay,.

A

So let me give it a try, I um and if I come and me, if I run into some problem, I'll I'll, let you know sure.

C

Yeah, if you want to chime in there on that one and say hey I'll, take that you know I've got access to the Machine and that you'll give it a try. That'll avoid rich doing it in parallel.

B

Another thing another thing that might be good is: if we could, if the three of us with Octavian, could sit down at some point, and you could show us how how this machine works and how this repo generates the charts and stuff like that. Yep then, if no, it might be I'd, be easier for us to sort of pick up some of the some of the work yeah.

C

So I guess it's how much like what time is it it's 2:30 I'm, just thinking we could do some of that now, even okay.

C

Checking calendars still have.

C

Just check-in calendar trying to anyway yep okay, so we have till 3:00. So if you want I can just can.