Layer5 Meshery Development, 21 Aug 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Incorporating wrk2 into Meshery

Description

Demo of proof of concept of incorporating @giltene wrk2 to add to the existing support of @LaurentDemailly fortio for load generation.

A

So basically, this war, I'm gonna, show you what kids do to have it so yeah. Let me see if that's one? No, so if you can see the screen, yeah rains. Okay, perfect. So the statistics current the when we do is doublet. If you basically I just spend some time to see the difference, what are their key to adding what Hilton and what killer was added yeah.

B

A

The difference was, in the lower part, with BG was multi severed, a a which this one was up from the kabocha that was clear and then trying to bring toward the JSON I. Think I can do here.

A

Yep show you so basically that's adjacent war in the latency. That is one thing, is total con, which cannot be get real. Why we need to do some C code changes.

A

For the Christian, but the output we were getting already, we have is in JSON and I was checking the forty output. That's what I put here, basically right, okay, so basically the main goal is to confirm the format and the parity difference with is a 40 and W key to what they're getting cheap wine white.

C

A

So I just put some references and still going the kind of related world and then here pretty much, that's the part, a brief response and we can simulate similar personnel and value and actually to be winning the Chris count at this point for this, because in them I haven't seen in the fortieth at we are because it's super mention of yet one is percentile and Obama's value each.

A

If we move to wlk to it's going to be similar or something like that, you can just ingest on.

B

Part you're saying hey wanted to get. Was it Total Request count that would take a Seco change, yeah.

A

So, basically, right now, that's, for example. Okay, let me do like that.

A

So basically, you see here now: okay,.

A

That is a table coming, okay and in order right now we are taking from this table this percentile and the value did this latency, but we can also take the total count to take this top account and put in JSON in vastly. We need some changes in C code. That's what I was discussing. Yes, it is title to the so that that's basically I can show you quickly what we discussed yesterday. That's I were asking him some question, and so he was pointing that.

A

Currently it's this wall, that we can't get the total at least, but my question is: do we really need it because in 40, I haven't seen that the latency percentile and topic on a correlation I'll be getting this? We have 40 all right now.

C

You know, let's go with the, so if you look at here are trusting for count mm-hmm there was this duration, histogram section you know in done, I'm. Actually looking at the data you shared later in the document, but low generation integration mm-hmm, oh there, if you actually scroll down the first few blocks, you'll actually see a pure asian histogram account and there was a maniac, some average yeah. So that's sort of like nowhere, yeah I, think that's kind of where the value is used. Looks like the rest is all like standard deviation.

C

Okay, not so, which means you know, it's all percentiles not like. It's like.

A

That's you see in here yeah, that's a biggie. We have. We can get these STD deaths from marksman. Let me sleep in this output but like if you see many 1s today, you can get this. We can change the diamond to be exactly same, but yeah hold on. Something is.

C

Just something is off: like I mean your data is, is this I don't think it's? This is complete. It complete. This probably is not complete, so the duration histogram is one huge block. The other one is a person toilets. Percentiles is actually small. The duration histogram contains a start, stop start and start end. You know, so it actually contains all those time. Sequences I mean again like that again historical, so I, don't know you gotta be smart. You me yeah, like you see this.

A

Part here it's kind of what I work for into the data part. That's the one I got from the forty as a reference and I was checking from YouTube, but yeah. That's wanted to confirm holy.

A

So let's start enter it's whether you start the performance testing and then it when you end it. Yet it's not very first starting its just take a look.

C

At Florida, no I'm have something running more, but we can go to Hawaii over.

C

You can show you if you just give me a second look, maybe to actually find the best part. Yet no brochure.

A

Yes, there was learning some war scripting.

A

B

Sounds like it's possible, you might dip your toe and to see as well yeah.

A

That's a similar, this law pretty much event, although there is a debug, for example, you can print a big place response, but that's not exactly we put in it. Oh yeah. We can try to bring that and all right we can basically customize all these and when I like some of the requests errors. I'm, all you dance. My screen, sorry did I said.

C

Okay, I'm just finding on sharing I. Remember whenever you guys are done. Okay, brief, aside yeah, you can yeah.

C

Yes, can you raise your screen yeah.

A

C

Are you seeing the JSON data yeah? Here's a sample like you know, I'm calling it for this yeah? Oh, so you can't all the labels start I'm, sorry like no. We can actually manipulate these with less radiation, etcetera, like the students, a configuration actual from here on, like it also the actual QPS number of threads versions. I think like no probably I presumed on your t2, is giving us this data. Now the remaining is actually.

C

This is a duration, histogram part like I'm in this histogram part, you can see, there's a start and the percentage value, and so this was this was I mean I guess was closest to the data points I which were even before. So there is a percent. There is a count now this is kind of like an old one, for dial is actually using to chart this data.

C

So there are two parts to this that you know one is accumulated graph and yet one it's like the histogram so but the data for both are actually coming from here. So these are the data points where sometimes are not the date upon the person plants are just used for printing here at the top. You know you see this line here. Scream yeah see this line. If it he yeah, you see the 50 p90 chart right.

C

What are you seeing on the chart is actually the data point you can correlate it 48 50 60, M I, didn't you should be able to find them yeah. So that's like the percentage like in a percentage percentage percentage yeah. So that's kind of what know what it is.

C

You see the pointer that this is what this is. The data points and I was working to it's.

A

Done the total count against total count against the value, the percentile value, yeah yep right.

C

Here so one is actually the I think if I understand is part, one is actually discharge value. At the same time, the histograms are I think it's a combination of the two.

C

This is literally not used to draw the chart, but you know there is a you know. So there is a chance for function. It actually takes this chart. Data horses and create data is rendered not short, but pretty much. The main attraction here or drawing the chart is turning this data here. This whole thing you can see like that. Many data points there and what you know so the rest is all like min max, etc.

C

Like you know some metadata, but this is time of the point you know which was working as the data points you.

A

See the point see the main limited, the data that part or this charger anything. So then we can do the I will revisit and oh maybe PR fall because I asked him if you should do against his, but he mentioned to do.

A

Not talk about the mutant little against he just for the photo calm, but there is no start and date that should be mono is start getting time stand for it is it start end date? Is it used also tree the graph out to be.

C

No, the Detroit is actually a summary, so it is just a summarization, so it doesn't, it doesn't like I said it doesn't literally use the data, it actually converts it to. You know a summarized data and then that's what used to be charted here. This is the one I'm.

A

Just I may be in this case that the cuca me confused. It's three dimensional yeah, that's okay, I will show my skin yeah.

A

Let's put this son so here.

A

Okay now start- and we don't take this to you- it mainly percent and count correlation between.

C

Yes, but just mean something.

A

Because yeah, let me to the debug.

A

That's if you do the in the devout no- and we are getting this pants- is there a date Wednesday charity, but it's not the millisecond timing. Doing.

A

You may get the top timing and maybe add that fun.

A

Because yeah, that's what I was saying you can get some.

A

B

So I guess I'm Irish. What you're saying is no the start and stop time for the test isn't used in the graphs that we showed today. That's.

C

Correct yeah, it's actually computed. What is actually shown is like it also based on the total I mean, so they have this any 9% I'll try Mike a notice in milliseconds and the charges against it. So don't let me.

C

Duration, yeah.

A

That's kind of 50 percentile and the value of it. What's the value and then total number of flip twists, that's silly things which we may use because the the risk could be like you know, key names and smoke can be added from one strip once similar way. We can get the total count, but in India that's so.

B

So being between the two output formats, water, we identifying as being missing from the vrk to the Total Request can percenter.

A

So if you see here, there is a table like okay.

B

You're saying about request basis for the histogram data, so.

A

Yeah, because you know the to do to bring it to suit Jason, it is giving the out, but it's coming from a see. It's not so you cannot pull this data, putting the education law that's, but once we make changes, we can get this and then I asked also till about two years now about one to one: percenter I'm, not sure if that's is sink.

A

We need because I ask me also I, guess he didn't the player, he wasn't against not sure I can reach Tina.

A

And there was other stuff I asked him. He said it's not implemented yet here, if you see like uncollected, latency statistics latency all this, but there is like decreased date and Arceus. That's Vienna passes argument a little critical, pretty much me now in a better state to have handcraft Italy the way we need.

B

Rate, that's computed for for IO computer as well correct, which would be the total throughput.

B

So I guess I'll just reiterate it sounds like. Are we in fact saying we've got it? Is it one piece of data or we're missing? Total Request count for WR k, yeah and we're also missing the count per.

B

Histogram data point no.

A

It's total nakusoo countly, it's for history, a histogram point. We have the total number of wreckfish. Oh okay,.

A

It's here, if you like start and it keeps increasing okay at this point of time and that's and then we have total certainty, certainly.

B

So good, this is good progress. Not only did you do the something of a kind of the initial park you got with T, though we've identified the missing piece of data Girish of what you're seeing it looks like the json formatted output here sounds like it looks like that's the you can validate that. That's probably the only thing that we're missing is a per histogram computation request, count such that we can show in the UI how many I fell into that bucket and.

A

I felt actually most of the stuff OTO guys yeah. They also I, think the most since I see it's. They kind of taken from the tt-to at all, so I had a feeling it so that that there are some similarities there to using.

B

The place see that you might need to modify. Have you what does that look like so.

A

No, it was like, oh it like four or five that you're talking and I just spent quite enough time to do it via slaw and I was something to better understand. But then I saw okay, that's kind of he hit hit the rock and then I contacted me. He confirmed, and then they say yes, then probably you can do this way and yeah I will I haven't tried, but to play this apart. I just was mainly focus on one.

B

At your fingertips, or, like you not too difficult to rather not a hard computation to do and and the data is right there, but yeah identifying, which, which area and wlk to to modify, might be take a little bit. If you are considering contributing this to guild, teens done you're. Okay, too, it could be I mean you could. As you look for, that, you could also open up an issue or a PR and ask him yes.

A

So basically I guess to or said it could be nice but I haven't conducted is built in and maybe just to confirm the same. If it's good idea and then otherwise yeah we may need to do our own fork or maybe to push the t01 and then just pull.

B

From there sure yeah, it would seem to me that, like I think the most likely outcome, considering the time is of the essence, is that you'd probably start from Telos fork. You would simultaneously reach out to guillotine asking if you'd like this contribution, asking you know for jury instruction as to where to go, I went to code and then at the same time create a fork and begin attempting it in that fork and then that way, yeah later, if he wants it, murder just great if he doesn't fine sure.

A

Yeah I will spend some time and then should be good and once compiled it then test it.

A

Yeah, so basically, then we are good, did only start in just percentile, and value should be good to go and I will put mainly inside the data like that.

B

C

Sorry, just even said she knows frozen present out: yeah cool, okay, we'll be dead. Sorry.

A

Yeah sure I will put inside the big data.

C

Like yeah, absolutely that'll be that'll, be a great term, yeah I know, but there is no value eg.

C

For me know the the value is I mean there are two value so he's saying for this start and end, that's the percentage and that's the count.

C

So it's like two things like we're the same stored in period. Oh, there is no percentile which person.

A

Oh, it's just the value all right sure. A percentile is like similar to each guy, like they're great.

C

What is it in that bottom section then percentiles, the percentiles have a value, yeah I think. If it's well done we're gonna, you see in the percentile section there is a value.

C

Are you referring to the percentiles here yeah? Are you working on it centers our best percentage here, okay,.

A

C

You know the one about is actually I. Think, like a cumulative percentage, you can see, the percentage being rolled up can be scroll to the top.

C

You know I see you can t started where the percentage is very low and then it actually starts building up pulling out building out and then it actually goes out onto like hundred percent. So that's like a total percentage is a.

A

Cumulative percentage of requests yeah a to a big check, time and education. That I was like you or these sounds good, sounds good. Yeah I mean.

C

This would be an interesting thing to actually talk more fun. Like I mean what I mean honestly I, don't have like the best of understanding here as well, which is actually like. We actually leverage before iOS JavaScript library to actually help with charting so yeah it'll be it'll, be good for us to find out what take each other's brains and see if we can.

A

Actually, you know yeah I, think this hold understanding kind of be a kind of news must, in this case, I've just been some common codes and description and asked already tailored stuff and put all together in the document and and jump to an implementation.

A

What sounds good yeah sounds good, yeah yeah. Thank you so much the English so I like that. We once you're.

B

Often racist man, it's Jerry,.

C

Yeah, this is great yeah.

B

Thank you guys. Okay, yeah talk to you since echo cardiogram.