[00:00.000 --> 00:13.160]  Hello, everyone.
[00:13.160 --> 00:23.400]  And now we have a talk by Johannes Bergberg, I hope you pronounce that right, is a JVM
[00:23.400 --> 00:29.720]  developer working on profiles and the underlying technology, it currently works on the GAP,
[00:29.720 --> 00:34.440]  modeling documentation, smaller utilities, and the Firefox profiler.
[00:34.440 --> 00:39.880]  It's going to have a talk on Firefox profilers beyond the web, using Firefox profiler to
[00:39.880 --> 00:45.160]  view Java profiling data, and yes.
[00:45.160 --> 00:51.200]  Yeah, thanks for the kind introduction, yes, I'm Johannes Bergberg, I'm working at the
[00:51.200 --> 00:59.320]  sub machine team at SAP, we create the best JDK, so just download it, but I'm here because
[00:59.320 --> 01:08.520]  I worked on the Firefox profiler in the recent months, and yeah, I'm going to start now,
[01:08.520 --> 01:13.960]  because when I'm telling people that I'm like working at SAP at Firefox profilers,
[01:13.960 --> 01:19.120]  they first start me two questions, like first, wait, SAP does open source?
[01:19.120 --> 01:25.480]  Yes, SAP does open source, and quite a lot, for example, here at sub machine, my team
[01:25.480 --> 01:32.560]  we're working on the open JDK, for example, if you've ever used JDK 17, we have the JDK
[01:32.560 --> 01:41.760]  17 maintainer in our team, so we do many nice projects in this field, SAP also works with
[01:41.760 --> 01:48.520]  the Eclipse Foundation or other parts, so yeah, we're doing quite a bit of open source,
[01:48.520 --> 01:56.360]  but how did I end up here talking to you in this Mozilla death room as a JDK developer,
[01:56.360 --> 02:02.840]  which normally doesn't happen, so I had a problem, I had a project on debugging last
[02:02.840 --> 02:08.720]  year, so what I wanted to do essentially is that I had on the one hand my IDE, and on
[02:08.720 --> 02:16.480]  the other hand the JVM, they wanted to improve the protocol in between, so I had then some
[02:16.480 --> 02:21.880]  test cases and some of these, some integration testing, for example, here they parsed some
[02:21.880 --> 02:28.720]  program and I did something with it, and I wanted to see why it was slower than I expected,
[02:28.720 --> 02:35.400]  and so what I wanted to do is just write plug on my IDE, tell it that, hey, profile it and
[02:35.400 --> 02:40.920]  show me the profile, and that's all, and I wanted to do it with open source source, because
[02:40.920 --> 02:47.760]  I like open source and that our team open source is really key, so I didn't find anything,
[02:47.760 --> 02:55.040]  but I found some tools that got into this direction, so essentially there were tools
[02:55.040 --> 03:02.960]  that produced some flame troughs, STUI, and Mario Fusco set a few days back, a lot of
[03:02.960 --> 03:08.960]  frame graphs, when you do something stupid it punches you in your face and it's impossible
[03:08.960 --> 03:15.680]  not to see it, so that's great, you can use flame troughs for a lot of things, but sometimes
[03:15.680 --> 03:20.840]  they are not enough, so when you regard visualizations you have those easy tools that are easy to
[03:20.840 --> 03:25.000]  use, but don't have that many features, and on the other hand you have the big tool, it's
[03:25.000 --> 03:32.400]  called JMC, there's lots of features, but has a quite steep learning curve, so I had
[03:32.400 --> 03:36.640]  to write my own, because I wanted to have something with more visualizations than just
[03:36.640 --> 03:43.400]  flame troughs, but that was usable, also usable to the end user, and not just the OpenJD
[03:43.400 --> 03:51.360]  developer like JMC is, quite frankly, and I wanted to just right click on a method,
[03:51.360 --> 03:58.080]  on the test method, and just tell my IDE, hey, run the thing, so I don't end up writing
[03:58.080 --> 04:06.920]  so much code and boilerplate, so writing your own intelligent plugin, and that's the IDE
[04:06.920 --> 04:12.000]  I'm targeting, but it's quite the same for all the IDEs too, like visual studio code,
[04:12.000 --> 04:19.040]  is quite simple, you can just download some templates and work on them, so this I did
[04:19.040 --> 04:25.680]  like in August in like half a week, but then it came to visualization options, so I looked
[04:25.680 --> 04:30.600]  around and thought, maybe creating my own visualizations like flame troughs and so on,
[04:30.600 --> 04:38.520]  but this turned out to be cumbersome, and really would take a lot of work, so I looked
[04:38.520 --> 04:45.920]  around for web-based visualizations, because you can just embed a web-based profiling view
[04:45.920 --> 04:51.480]  into your IDE, because for example in Telstra you have a chromium, an embedded chromium,
[04:51.480 --> 04:55.880]  and in VS code it's essentially a browser anyway, so that's no problem, iPhone speed
[04:55.880 --> 05:01.800]  scope, which is quite nice, but the problem is it doesn't show anything other than stack
[05:01.800 --> 05:07.040]  traces and some profile timing information, and I wanted to show more, there is kind of
[05:07.040 --> 05:11.920]  an existing plugin for IntelliJ, but that only shows flame grouse, and so we're back
[05:11.920 --> 05:19.000]  to step one, and then I found Firefox profile, and this is how I ended up here, so essentially
[05:19.000 --> 05:24.000]  Firefox profile is this large application that you've probably seen in some other talks
[05:24.000 --> 05:32.200]  today too, this is actually taken from a previous talk, because I was too lazy to run Firefox
[05:32.200 --> 05:40.080]  profile, directly you see here it's like it does everything, and it's quite cool, and
[05:40.080 --> 05:46.920]  one of the many advantages is that it's developed by a small team that answer even stupid questions
[05:46.920 --> 05:53.520]  in a matrix channel, so check it out if you have some questions, and they were open to
[05:53.520 --> 06:01.400]  working with me on my project, and also it's backed by cooperation, it's backed by Mozilla,
[06:01.400 --> 06:09.880]  which is quite cool, because it's not a one-man project like other tools are, and so what did
[06:09.880 --> 06:15.920]  I do to integrate it, and what might you do if you want to also use Firefox profile,
[06:15.920 --> 06:23.320]  it essentially points down to creating a converted to my data format, it's the JFR file format
[06:23.320 --> 06:28.760]  to the Firefox profile format, and then you put it in the server, because Firefox profile
[06:28.760 --> 06:36.560]  likes it when you can just say take this file from a URL, so you put it in the server, that's
[06:36.560 --> 06:42.680]  fine, for me I created a travel in server, then you can just wrap it in IntelliJ plugin
[06:42.680 --> 06:54.840]  or use code plugin or zone, and then I took the Firefox profile UI, you can use profile.firefox.com,
[06:54.840 --> 06:58.880]  but you would typically just host it your own, because Daniel Burke's in demos like
[06:58.880 --> 07:05.640]  Bolly today, when I have time, and you can control all the changes that come into your
[07:05.640 --> 07:12.920]  Firefox profile UI, and also you can use a modified version, which I did to put all
[07:12.920 --> 07:17.840]  my progress that is still out there and not merged into one version, so the things you
[07:17.840 --> 07:26.080]  later see, that the inmate will sometimes not yet in the mainstream Firefox profile,
[07:26.080 --> 07:33.560]  so shortly to the file format, so file format is quite simple, it's just a chasing file,
[07:33.560 --> 07:40.200]  and this is zipped, and it has some metadata information like the name, the interval, some
[07:40.200 --> 07:46.800]  settings, and then you have the stretch information, so for every thread you have a separate section
[07:46.800 --> 07:52.320]  in the profile file, and you have a list of samples there, and this list contains essentially
[07:52.320 --> 07:59.480]  the time of when the sample is taken, the stack, and the CPU delta that elapsed the
[07:59.480 --> 08:05.600]  CPU time since the last sample, so it can be used to show the CPU usage data, then the
[08:05.600 --> 08:13.080]  stack is not all the stack, but it's an index into the stack area, and this contains like
[08:13.080 --> 08:22.200]  the list of stacks, and where the stack is just a frame appointed to the previous stack,
[08:22.200 --> 08:26.160]  like the top frame, and then the previous stack, so it points back, and then the category,
[08:26.160 --> 08:31.640]  and of course the frame is an index into the frame area, and that contains the functional
[08:31.640 --> 08:38.240]  line, so what you need for a frame, and then of course function is not really the function,
[08:38.240 --> 08:43.880]  but it's an index into the functions area, and you get the point here, because name and
[08:43.880 --> 08:50.720]  file are of course not strings, but they are indexes into the string table, that's quite
[08:50.720 --> 08:57.440]  hard to debug sometimes, and I had many struggles with it, but it's quite performant, and it's
[08:57.440 --> 09:05.240]  easy for the frontend to see, so after I explain to you how I did it, here's the plugin, I
[09:05.240 --> 09:12.680]  call the Java JVR profile plugin, you can find it in the track branch marketplace on
[09:12.680 --> 09:21.040]  GitHub, it's open source, it's MIT license, I believe, and it works because JVR was open
[09:21.040 --> 09:30.240]  sourced with JDK 11, so it's all open source, just try it out, so how can you get it, just
[09:30.240 --> 09:39.640]  open your IDE, and go into the plugin install, view and type in Java JVR profile, and you
[09:39.640 --> 09:49.640]  get it, then you get some nice buttons, when you right click, and you can just click on
[09:49.640 --> 09:57.280]  the profile with JVR, then you profile it, and then here's a simple example application
[09:57.280 --> 10:02.480]  that just computes kind of complicated, a few nuchin number or something like that,
[10:02.480 --> 10:09.240]  and then you can see that it executes the program with some JVR-related options to
[10:09.240 --> 10:17.920]  profile it, and then it automatically the profile of JVR-file, you get the call tree,
[10:17.920 --> 10:24.880]  you can also look at how every frame is like executed, whether it's in interpreted mode
[10:24.880 --> 10:31.600]  or in our JIT compile, then you can double click and jump back to the IDE, so it has
[10:31.600 --> 10:39.200]  basic IDE integrations, and you can shift double click and see the source view that was presented
[10:39.200 --> 10:45.040]  in the previous talk shortly, so you can see here that in this code, the recursive call
[10:45.040 --> 10:51.480]  is called the most, it's found the most time in the static calls, and then you can have
[10:51.480 --> 10:58.880]  a function table which lists all the methods combined, so from all the stack traces and
[10:58.880 --> 11:06.680]  you see what method is used the most, that's not yet in the mainline Firefox Profiler,
[11:06.680 --> 11:16.880]  you can have some flame crafts, and you can have some nice tooltips, and you can get some
[11:16.880 --> 11:24.600]  information on the profile, you can even upload it like a normal Firefox profile, so you can
[11:24.600 --> 11:32.840]  share these profiles and view them in a normal profile at Firefox.com, just with some few
[11:32.840 --> 11:40.720]  features, and then we can also open any JVR-file, and JVR is like the default, like the de facto
[11:40.720 --> 11:46.760]  standard for profiling files in Java, and you can see here that we get also some crafts
[11:46.760 --> 11:54.840]  that show us the CPU load and give us some summary on GC, like how much memory it requested,
[11:54.840 --> 12:00.040]  and what's also cool, we can not only see timing, but we can also see call tree for
[12:00.040 --> 12:04.960]  other things, like for Java error, for thread starts, where did they happen, or for object
[12:04.960 --> 12:14.240]  allocation, and when you get to Java error, you see that this code uses error to create
[12:14.240 --> 12:22.760]  error every time, like probably the parsing fails, so we can also see, as the last part,
[12:22.760 --> 12:29.480]  the market chart, so we can get all the events that the JVM remits, so for example, we see
[12:29.480 --> 12:43.720]  here at the top, that we had a drop in the memory that the heap was large, and we investigated,
[12:43.720 --> 12:49.800]  we could zoom in and see, okay, that's probably because here GC happened, and this GC took
[12:49.800 --> 12:54.960]  like 10 milliseconds, and this is quite nice because you can then investigate, have the
[12:54.960 --> 13:02.560]  call tree as a simple thing, and then later drop in and go deep dive into the data, and
[13:02.560 --> 13:08.640]  if you know Async Profile, it also supports Async Profile, and also when you want to create
[13:08.640 --> 13:17.080]  the profile, you can decide where you want JVR to build in for the JDK or Async Profile,
[13:17.080 --> 13:26.240]  and that's all, but I think I still have some time, so I hope this works, because I can
[13:26.240 --> 13:31.720]  tell you a lot that it might work, and we can show you some screens, but here is the
[13:31.720 --> 13:37.080]  actual working, hopefully working prototype, so just right click on the main method, tell
[13:37.080 --> 13:46.440]  it Profile with JVR, and then it tells you, hey, I profiled it, then it opens the profile,
[13:46.440 --> 14:01.200]  and then you can just look at it, zoom around, and see that you have to select the main JVR,
[14:01.200 --> 14:06.480]  you can jump back to the source code, you can shift, double click, and everything, and
[14:06.480 --> 14:12.920]  the cool thing is you can even open, as I showed you, arbitrary JVR files, they shouldn't
[14:12.920 --> 14:20.120]  be too large, because then my program runs out of memory sometimes, it's still a prototype,
[14:20.120 --> 14:30.280]  so try it if you want, I would be happy to have any feedback, go to the issue pages,
[14:30.280 --> 14:36.680]  and eventually you find this plug-in, as I said, in the chat print marketplace, you
[14:36.680 --> 14:43.000]  find me on Mastodon, on Twitter, and on GitHub, but you also find the subprojects, so you
[14:43.000 --> 14:50.080]  can use parts of this project, like only the converter, and only the JVR to Firefox Profile
[14:50.080 --> 14:59.200]  server, you find more information on my tool, and also a background information to this
[14:59.200 --> 15:06.640]  talk, in two blog posts on my blog, and you can find the submachine team at Sweet Submachine
[15:06.640 --> 15:24.120]  at Twitter, and find out about our great projects at Submachine.io, that's for me, thanks.
[15:24.120 --> 15:34.720]  Thank you very much, and we have quite a bit of time for questions, so...
[15:34.720 --> 15:35.720]  Thank you.
[15:35.720 --> 15:41.840]  Have you received any feedback from users about this tool, like some colleagues or...?
[15:41.840 --> 15:48.000]  Yes, good questions, yeah, I received quite some feedback on Twitter, they were quite
[15:48.000 --> 15:55.440]  happy, I did some internal... I showed it internally at SAP, to some colleagues, and
[15:55.440 --> 16:01.920]  they were quite happy, I also showed it to friends of mine, and also it was quite great.
[16:01.920 --> 16:08.600]  It still has, of course, problems, because it's a prototype, not everything might work,
[16:08.600 --> 16:13.440]  but yeah, I'm looking forward, so just give it a try, it's free, it's open source, open
[16:13.440 --> 16:18.360]  some issues on GitHub, if you like, so yeah.
[16:18.360 --> 16:20.400]  I will steal the mic.
[16:20.400 --> 16:25.640]  Did you add to forward some patches upstream to make it work in the first place?
[16:25.640 --> 16:27.280]  Yes, a lot, so for example...
[16:27.280 --> 16:31.800]  Because you mean they have bugs, or you need to adjust it for the...
[16:31.800 --> 16:40.480]  There for example, this feature here is not yet in Master, but for example, what I added
[16:40.480 --> 16:47.200]  was, you saw this, this, you saw this here, this, the listing of the categories, this
[16:47.200 --> 16:53.760]  is something that I added, and I added pull requests for the function table, which is
[16:53.760 --> 17:01.320]  not yet in because it's far more work, I added some resizing and searching, that's yet, that's
[17:01.320 --> 17:07.360]  I think in, so I added some pull requests that are in, but not all of them, but I hope
[17:07.360 --> 17:10.640]  I get it in in the next few months.
[17:10.640 --> 17:14.640]  You can give beer to the upstream developers that are here to get the patch.
[17:14.640 --> 17:20.880]  Yeah, but I fixed all the bugs that they wanted to be fixed, so I think we're working
[17:20.880 --> 17:21.880]  more like colleagues.
[17:21.880 --> 17:23.360]  I will ask a follow-up question quickly.
[17:23.360 --> 17:30.040]  Did you have to do anything Java specific in that case?
[17:30.040 --> 17:37.320]  There's nothing that is real on Java specific, so the only the converters like Java specific
[17:37.320 --> 17:51.360]  because it takes Java data and processes it, but anything else in the UI is generic.
[17:51.360 --> 18:00.120]  I was wondering if this could be useful to profile Java applications that run native
[18:00.120 --> 18:04.280]  code and the other way around, and how would that work?
[18:04.280 --> 18:12.640]  Speaking of, for example, Firefox 400, I don't think we have a lot of insight into the performance
[18:12.640 --> 18:16.520]  of our Java code, and that would be useful to have.
[18:16.520 --> 18:22.000]  So, it's in some ways where they can work with chain I or something like that.
[18:22.000 --> 18:29.800]  I think asking profile is a quite nice profile that I also support, and this gives you information
[18:29.800 --> 18:38.480]  on native, on native phrases, and I'm working on getting a profiling API into OpenJK that
[18:38.480 --> 18:44.200]  improves this a lot, but if you want to know more about the struggles, come tomorrow to
[18:44.200 --> 18:52.280]  my talk in the Fuji room, they hear enough about this topic.
[18:52.280 --> 18:53.280]  Hi.
[18:53.280 --> 18:58.040]  Yeah, thanks for the presentation, very impressive.
[18:58.040 --> 18:59.960]  The question I had, actually, I have two questions.
[18:59.960 --> 19:07.480]  The first one is, what is about Firefox profiler that made you choose it, I suppose, and then
[19:07.480 --> 19:17.760]  the second question is, do you see the potential for maybe a tool like this to have a web-based
[19:17.760 --> 19:18.760]  editor?
[19:18.760 --> 19:23.520]  Yes, the first thing, why did I choose it?
[19:23.520 --> 19:28.240]  Because it's a nice project and I didn't want to write everything on my own, and it contained
[19:28.240 --> 19:33.160]  lots of the features that I already wanted, so, yeah, and they could ask questions, because
[19:33.160 --> 19:38.040]  when you're working alone on such a prototype is that you don't get any feedback.
[19:38.040 --> 19:45.160]  Here I got a lot of feedback on progress, and it was a quite good learning experience.
[19:45.160 --> 19:51.880]  The other thing is, yeah, as I said, the project consists only of a tiny intelligent
[19:51.880 --> 19:57.120]  wrapper around the other code, and this can be used to just integrate it into your Spring
[19:57.120 --> 20:02.200]  web app to directly show it in the web.
[20:02.200 --> 20:10.400]  I have some ideas on this, just follow my blog to know more about this, or my Twitter
[20:10.400 --> 20:19.320]  almost.
[20:19.320 --> 20:21.840]  So do you think about the upload local profile?
[20:21.840 --> 20:25.360]  Is this an important feature in your opinion?
[20:25.360 --> 20:31.680]  Is it something that you'll, is it part of the features that made you choose Firefox
[20:31.680 --> 20:32.680]  profile?
[20:32.680 --> 20:36.720]  Yeah, it's quite a great feature, because you can essentially, in open source, you could
[20:36.720 --> 20:43.960]  later than just tell other people, I have some performance issue here, for example,
[20:43.960 --> 20:44.960]  for issues.
[20:44.960 --> 20:52.440]  It's essentially the same thing Mozilla does, and I hope I can open it to use it in a corporate
[20:52.440 --> 21:02.200]  setting, for example, adding features to use Microsoft Thrive for it, so it's more safe,
[21:02.200 --> 21:06.240]  because currently you're uploading it into the web, and that's not that great when you're
[21:06.240 --> 21:08.840]  doing internal company stuff.
[21:08.840 --> 21:14.080]  But this feature then could really make it easy for people to just say, hey, here I have
[21:14.080 --> 21:21.400]  some problem, give me your stack trace, just upload it, and you're fine.
[21:21.400 --> 21:26.440]  So maybe next feature could be a server, and make it configure.
[21:26.440 --> 21:33.880]  The problem is, it's just me currently working on it, and so it's just a prototype, I have
[21:33.880 --> 21:39.320]  to see what could be implemented in the future.
[21:39.320 --> 21:41.720]  Any other questions?
[21:41.720 --> 21:47.000]  We still have three minutes, so if you have questions, we can take them.
[21:47.000 --> 21:53.920]  Thank you very much, do you want to add anything else?
[21:53.920 --> 22:00.040]  Just try it out, I would like to have some users, because I want to know what features
[22:00.040 --> 22:04.880]  are important to you, and also where you find some problems, for example, someone found
[22:04.880 --> 22:11.840]  that the Microsoft support wasn't there, so on Microsoft it just crashed, on Windows,
[22:11.840 --> 22:19.280]  so if you find any issues, please drop by, and come to the Matrix channel of the Firefox
[22:19.280 --> 22:26.240]  profile, there you can find me sitting around all day, and can also answer questions directly
[22:26.240 --> 22:27.240]  on it.
[22:27.240 --> 22:28.240]  Thank you very much.
[22:28.240 --> 22:35.240]  Thank you very much.