Node.js Diagnostics Working Group, 31 Jul 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Diagnostics WG meeting - July 31 2019

Description

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

A

So welcome to the diagnostic working group meeting for July, 30 29, sorry, 31st, 2019 Peter can't make it today because he hasn't a conflicting appointment. So he asked me to chair will go through the standard agenda, which is on the issue that was posted. It is issue number 321 and so first off we'll see. If there's anybody who wants to share any announcements.

A

Okay, if not we'll move on to the into items, the first one is they sent. Cooks, do not recognize execution contexts, 2, 2, 3, 6, 0, I, don't think, there's any update unless somebody else on the meeting has an update.

A

Ok, if not we'll move on to proposal inspected or flag number 298, this one I think Peter was trying to get Alexi to our meeting and I guess: I don't seem today. Summer is probably a hard time, so we'll skip that one for this week again, unless somebody else has had I can.

B

Talk with four likes it see if he can attend the next meeting. Okay,.

A

That would be great because I think that's the next step on that one was we needed some more input or discussion with him. Okay,.

B

A

Our next step, okay,.

B

A

Good, the next two, which are number 295 and number 211, are related to best practices which are still high on our list. Unfortunately, it's a little slow going lately summer and whatever and also Jewish has been temporarily pulled to some other work within IBM. So that's gonna mean that we're not gonna, really at least to reach, won't be able to loop back to that till pushing that forward until late August. So that's all the updates I have unless anybody else has I'd add.

A

Okay and then the last regularly scheduled item was perfetto in nodejs and we did for an update and you know Kevin who's working on that or Kelvin Kelvin Kelvin who's working on that is basically just waiting for some changes on the v8 side, because perfetto is changing in response to some of the things he discovered while looking at integrating into OGS- and you know, he's actively sort of waiting for that working on it and that's the update on that.

A

Okay, so that's the last regularly scheduled agenda things. The one thing we did at the agenda, though, was a demo of the tool that Chris Chris Keller had we're sorry Hillier had given us, you know, give us a bit of a demo a while back, and we wanted to take a few minutes to give us an updated devil. So let's do that at this point.

B

Already yeah so uh last time, I was at this meeting. I ran through some slides, so didn't have time to get to a demo, and so let me try to do that.

B

A terminal is the whoops. Is the yes fontsize readable all right so feel free to butt in with any questions as I'm doing this, and so to kind of recap. What I'm working on is a is some tooling around diagnostic reports. If you generate diagnostic reports, you can use this tool to modify it filter it. You can inspect it for potential problems, there's a few more things on the roadmap, such as like guided usage or some sort of wizard. That would assist people in diagnosing these.

B

These reports after, for example, a an exception. Maybe the process Crashers with an exception so yeah, so it's got kind of two two components, one as a CLI and so I'm, just gonna. Look at the CLI right now. So it's previously when I presented this tool is called Gnostic. But if I want to name it that I have to go through some enterprise process here, so they suggested I name it something very boring for now. So right now it is report toolkit, and so you can see here that we have several so commands.

B

And hopefully this isn't constant, like you know, constant development, so hopefully haven't broken anything. But there are there's some tools here that you can do so. Here's the first one is diff, and so, if I want to take a dip of two reports now you could you could do this with a visual diff tool or command line deaf or something but what's cool about this? Is that it? It ignores stuff that doesn't matter I plan to add the ability to actually choose what doesn't matter to you.

B

But when we take a diff here and we can see that the default output will show this tabular thing where it's got the the packages or the reports side by side, it shows you know the the path that, like the key path and the report JSON object. It shows whether something has been modified, deleted or added, and it just shows stuff side by side. Now you have control over whether or not you want to truncate things.

B

If you don't want this tabular output, you should be able to just get JSON output, or maybe you want pretty JSON output. You can do that too, or maybe you want CSV I, don't know if that works. Allen seems to work. Okay, it's got some anzi in it, then you, but so that's kind of the diffidence II. What change between two to two report files.

B

The next one I want to look at is redact and so redact is.

B

It allows you to take a a JC report and then it it is, tries to be smart and it read acts fields that might have sensitive information because, of course the diagnostic report contains the environment and your environment might have AWS keys or I mean cloud keys or Google Cloud keys or whatever in it, and we don't want to transmit those, and so, by default. The first thing that this tool does, when it loads any Jason report, is it will redact it, and you have to be explicit about turning that off.

B

If you want to turn it off so, for example, if I went into the help Ford, if we could see that there is a flag, show secrets unsafe, and so it will not redact. If you do that, and so all the commands, otherwise by default will redact secrets, and so the next one I wanted to look at was redact, and so, if I wanted to redact a report, it's just we see. Hopefully, I have something here: una redactor report, whoops yeah.

B

You have to fix that but I, but at any rate you can see that it pulls it red X this information, if it is not just something it doesn't like these may be false positives in shared objects. But if we go up to like here so it looks for a session ID, maybe the respuesta IDs sockets.

B

It wants to hide that stuff. So here's some tokens and stuff things that shouldn't be in there, and so it takes them out. You know it's just kind of its redact is, is kind of a sugar for for another, more powerful command, which is transform and so transform.

B

Essentially, if I go to transform help, so there are various transforms that you can do to to a report, and so you could take report and output in CSV format, or maybe you want new line, delimited JSON. You can actually chain these transforms together.

B

So if, for example, you want to get the stack hash, and so this is something that was mentioned when I presented diagnosis, I can't remember who suggested it, but it was that they wanted to get a like a hash of a an exception, and so what they do is they look at the they want to look at these reports and after after the process crash, doesn't see if that uncut exception that crashed the process is unique, and so every every exception will have a unique hash of it, and so there is always a a stack of some sort.

B

When you create a report, even if it's not an exception and so I should be able to say, I want the stack hash you see here.

B

I think it's confused about that.

B

Okay, and maybe we make this- we do that okay, so here this is an example of chaining transforms, and so we can see that there was one stack in there in this report. We generated an sha key and that should probably not be truncated by default, and this is what you can see there, and so, if I want to actually just pull that out and see well, what exactly is that thing I can say include I think it's JavaScript, okay,.

B

Okay, so this is actually the spec that that was in there, and so it takes this information and in this information and it basically concatenates the whole thing and generates this unique hash of your of your stack trace. And so you can take that and put it in your database or whatever and run it through your analytics and metrics and see. What's going on.

B

There's also in the inspection stuff, and so what that wants to do question go ahead. Where the stacks is it possible to do it on all the files in the directory, yeah, there's I mean I. Think in theory. Let me see so. If I wanted to go in here and just run it against everything, so yeah yeah do it in the tabular form.

B

Up tea, what I want is stack hash and then in one table, yeah, okay, so there you go there. Of course, all this kind, because the ones I'm working with for these fixtures, but we can see that some of them are actually just are the exactly the same. Writing.

A

Some are different right.

B

So yeah and so generally everything except the diff tool which expects to no more no less reports. Everything all the other commands should be able to work with and number of report files.

B

Inspect and so inspect is like a thing that runs these heuristics against your report file and it tries to detect problems in there, and so there are rules and you can make your own rules, but there are some built-in rules, and so this is what we have so we can assert.

B

Maybe we can pop an error if there's CPU usage is too high or memory uses are too high. If, for example, you have a library version, maybe you've compiled your own custom node and your library version mismatches. What what core expects to be in there?

B

That can be a source of problems, and so it will detect that it will also detect things like timeouts that are basically things that are are sitting there waiting to do stuff and they probably shouldn't be waiting that long and there was a article Girish wrote about about see these use cases that I wouldn't implement rules. For so we can see that and by default it will just run all of them, but you can configure this big file, so maybe I want to look at.

B

So this one should be clean, so no issues great. This one should have a library mismatch problem and so looks like we need to.

B

To do is to make this larger, but I think we can say like no drinkie or something, and maybe there we go okay, so yeah here we can see that the custody shared library, Lib : when one is conflicting, was opened. Ssl, 1b, and so it expect that a file was a different name essentially, and this uses some regular expressions. That may not be perfect, but it it works for me so another one we have using. This kind of you know test vector files.

B

Is this this particular one has a long time out, and so it says that it there's a time out. Somebody made like a set time out or what have you and it's a timer with a future exploration in three hours, so there's a timer in your code. It's going to execute three hours now, which is probably not what you intended to happen now. Maybe you did. If you did, then you can disable this rule and your configuration yeah.

A

B

This is just another test, so if there's a long time out, but that timeout is unwrapped by default, it ignores it, and so that means, if it's the only thing in that is in the event loop we're expecting the process to end, if that, if that timeout is in reffed, and so we ignore it, that may not be correct, but that seemed okay to me.

B

Grits I, think that was testing the redaction stuff. So this is another long timeout test. Here's some.

B

B

May have actually disabled this rule. Yeah I think I disabled that that rule by default, but what it would do is it would look in there and say uh well. The the CPU usage is is too high and what's neat about those particular rules, the CPU use is the memory usage one. Is you didn't give them like ten different reports, and you can tell it to take this average across those ten different reports? Now you can, you can do like a mean and a median you can or not a median I'm.

B

Sorry, you can do a mean you can do a Max and min or you can do like each and so each would check if the CPU is over. Some amount on every report that you give it or you could use mean to average all the reports or max to make sure whatever the maximum CPU is across. All those reports make sure it's not higher than X just kind of details, but that's what I've got going on right now. You know this is nearly ready, as so far as being being like an MVP I.

B

Think it's nearly there there's just a couple more things: I need to hammer out and then, after that, I've got a lot of documentation. Work like that I need to write so I'm hoping to release this within the next month or two. You know it just kind of the documentation. It's gonna be a bit of a swag, I, think so yeah and then in the future. I want to do this guided usage stuff that somebody was suggesting to kind of help. People debug these diagnostic reports. So that's what I got so.

A

I'm, just by one questions, are you holding off releasing the repo until you get the docs, or is it that you can release the repo, so people who look at the code while you work.

B

I I kind of wanted to like just have everything ready for for use. You know I'm I'm, happy to let people actually just like use it before then, um but I kind of wanted to do like a okay. Here's, this new thing, here's this website and and kind of make the rounds in social media and see if I can get a few more eyes on it than I normally would if, if it kind of trickled out there, okay.

A

Okay, so you're waiting, you're waiting till you can get the baseline doc and then you offer yeah.

B

It's just kind of yeah baseline, okay. This is a project that that is, is ready for you to use and if it's undocumented I, don't consider it really ready to use at that point. So.

A

So I'm just taking those so like, if you want to try it out in the meantime, you can ask you directly for a copy right, yeah, absolutely.

A

Okay, any questions a lot.

A

Okay, if not I, guess some, that's it there. Any other topics that people want to talk about this week.

A

Sounds like not so we'll call the meeting for today, thanks for everybody attending and we'll talk to everybody next time. Thank.

B

You thank you, but I.