Show HN: God mode in production code – a new way to debug in Scala

mpeg 13 years ago |

I'd give my firstborn for a debugger/profiler like this in .NET

kstenson 13 years ago | |

Closest I know of is IntelliTrace: http://msdn.microsoft.com/en-us/library/vstudio/dd264915.asp...

j_s 13 years ago | | |

Which requires Visual Studio Ultimate, listed at $13300/seat.

https://en.wikipedia.org/wiki/Microsoft_Visual_Studio#Editio...

http://www.microsoftstore.com/store/msusa/en_US/pdp/productI...

Retric 13 years ago | | |

That may be the list price, but my company pays around 2k/year for a MSDN ultimate subscription for each developer. Which includes licenses that don't expire for, Visual Studio Ultimate, every version of Windows, Office, SQL Server, etc.

binarycrusader 13 years ago | | |

Renewal prices are much lower; I can't believe the amount Microsoft wants to charge for the initial purchase though.

I don't know of many small shop software developers that would be willing or able to pay the $13K list price Microsoft wants for Visual Studio Ultimate.

What's really sad about that is that they bundled improved debugging with a whole bunch of other things that aren't useful to smaller businesses.

alive-or-not 13 years ago | | |

$0 on BizSpark, DreamSpark, etc.

j_s 13 years ago | | |

BizSpark does (for up to 3 years); DreamSpark does not include the Ultimate edition of Visual Studio.

https://www.dreamspark.com/student/software-catalog.aspx

socillion 13 years ago | | |

You definitely can get VS Ultimate through DreamSpark, but only through DreamSpark Premium (fee paid by school.)

http://i.imgur.com/9aHcIOx.png

https://www.dreamspark.com/Institution/Subscription.aspx

mpeg 13 years ago | | |

Oh man.

I have been using VS 2012 professional without realizing it didn't have intellitrace. And I have a license for ultimate too (company paid).

Still, I know about intellitrace, I really like the way the takipi java/scala tools display the info though :)

fijal 13 years ago | | |

Wow, this is more than a yearly salary in many countries. I suppose I wouldn't want to work for a year to get a better visual studio...

baboo 13 years ago | | |

http://tpb.piraten.lu/torrent/7963590/MICROSOFT_VISUAL_STUDI...

vosper 13 years ago | | |

Please don't post links to pirated software.

alive-or-not 13 years ago | |

IntelliTrace is far more powerful: http://msdn.microsoft.com/en-us/library/dd264915%28v=vs.110%... .

balakk 13 years ago | |

There's Windbg/SoS for the times you would need something like this.

Decidedly not as pretty, but pretty effective if you know how. Channel9 has a series of videos on WinDbg for .NET developers.

saryant 13 years ago |

I've been tracking down a NullPointerError in Play's new JSON inception handling. Eclipse's debugger has gotten me nowhere, I look forward to seeing how this one works!

spuz 13 years ago |

This looks simliar to Chronon which is a similar debugging tool that works on Java (though not other JVM languages) applications.

LinaLauneBaer 13 years ago | |

I have used Chronon for several weeks during a very strange project. The project was about integrating a highly complex financial system in our own infrastructure. The challenging part about this financial system was that it was written in a multitude of languages (Java, Perl, C, C++, ...) and that it had to communicate with external services and it had hard timeouts built in which were enforced externally. So if something went wrong and I wanted to find out I had only 20 seconds in the debugger to find out because then the timeouts would kick in and there was nothing I could do about it. With Chronon I was able to step through the code after it had already run with no time constraints. Saved me weeks...

Chronon is also useful to find all thrown exceptions - even the ones who are catched properly.

I am not an employee of Chronon Systems and I only used their tool once but it helped a lot.

nivstein 13 years ago | |

Unlike Chronon, Takipi is built for monitoring your JVM in production, imposing minimal and predictable overhead.

LinaLauneBaer 13 years ago | | |

Chronon is also built for monitoring your JVM in production. At least that is what they say on their website:

http://chrononsystems.com/solutions/production

elvinj 13 years ago |

I know that at least from my experience debugging our apps in production did become harder as we scaled out in terms of servers, users and code. I do wonder if its something that should be handled at the code level (by being more defensive, taking a more immutable approach) or by having better debugging tools. Anyone knows if this supports Ruby?

wereHamster 13 years ago | |

Or by using a better type system. Or by using the type system that's available in the language. Because it's all too easy to ignore the types and for example call `get` on a Scala Option value, or `fromJust` on a Haskell Maybe. Both will throw an exception at runtime if the Option/Maybe is empty.

elvinj 13 years ago | | |

Type safety does help a lot. The real hard ones I found to track down are the ones where you have to chase for a particular state to re-occur and you can't get a repro from ops or qa. At least for me these are the main bummers...

taeric 13 years ago | | |

Of course, the retort there is as much "write simpler programs" as it is "use the type system." I've seen and been a part of many a tangled knot in a type system that, while ostensibly safe, was so much more involved to do the simple things that I think it caused more bugs.

wereHamster 13 years ago | | |

I'm not saying strongly typed languages are bugfree. But they eliminate whole classes of bugs. The OP was asking about programming techniques to improve code quality or tools to help with debugging. Well, a type system is such a technique/tool that checks for bugs at compile time.

taeric 13 years ago | | |

Indeed. Apologies for what was too trite of a response. I can't explain why, but I have been getting heavily lured by the dynamic typing of lisp, lately. I blame that I am finally watching the SICP lectures online. :)

timdorr 13 years ago | |

Passenger Enterprise supports a live irb and debugger: https://www.phusionpassenger.com/enterprise#debugger_support

gurdo 13 years ago |

There's a very strong founding team around this company. I don't do Java too much these days, but if I were, I'd definitely use Takipi.

shpxnvz 13 years ago |

Signup is broken right now, so maybe someone else can answer. From the website it appears that the debugger is a SaaS app that connects to the agent on your production box from their servers... is this the case?

nivstein 13 years ago | |

The product consists of a slim JVM agent which connects and offloads the heavy crunching to Takipi's backend.

shpxnvz 13 years ago | | |

Given the nature of JVM agents, this means that the Tikipi backend, should it choose, may access all the data in the production application. Correct?

ETA - Thanks, missed that page on the site.

nivstein 13 years ago | | |

The moment you sign-up for Takipi a secret AES-256 key is generated for you (and never stored by Takipi in any way or form, and you manually enter it on your machine when installing the agent). This key is used to encrypt every piece of application data which leaves the machine. The source code is also encrypted similarly. The source code and the data are only decrypted in your browser when viewing the details of the event.

You can read more about security at Takipi here: http://www.takipi.com/features.html?nav=security

sillysaurus 13 years ago | | |

This makes no sense whatsoever. Their application is doing the encrypting. That means their application can do whatever it wants, including making it trivially easy to reverse the encryption.

I get that the intentions are good, but https would provide exactly the same trust guarantees, and ultimately it bothers me that they would try to mislead people into a false sense of security like this. If their application is encrypting data which is sent to their servers, then that means their severs have complete access to your data. (Whether they choose to make use of this power is up to them. They certainly have it, though.)

Encrypting the data prevents it from being leaked if their servers are hacked and their database stolen, which is good. But acting like they have no power to access your data is absurd.

jcheng 13 years ago | | |

I think the distinction is meaningful. A rogue employee would not have the power to access your data (as we've seen at Google for example[1]) unless they were able to sneak some code into the application code (as, sadly, we've also seen at Google[2]). That seems like a significantly higher bar and also would require a great deal more premeditation. Not to mention the possibility you mentioned of them getting hacked (which would be a FAR bigger concern to me).

Now that I think about it, I never did move my stuff from Dropbox to SpiderOak, did I... hmmm, maybe today.

[1] http://www.theatlanticwire.com/technology/2010/09/rogue-goog... [2] http://tech.fortune.cnn.com/2010/06/06/google-blames-wi-spy-...

baboo 13 years ago | | |

They are probably either lying or incompetent.

If they really had no access, then their server cannot do any "analysis" beyond storing it, and the app could as well just store it locally, or with Amazon S3 or whatever storage system the website already uses.

Obviously not everything is encrypted, or otherwise they don't realize that there is no reason for them to provide a simple storage service, considering most websites already have some way to store data.

Or maybe they want to hold the data hostage.

mseebach 13 years ago | | |

For tons of applications this falls somewhere in the range "impossible" -> "totally unacceptable" -> "very worrisome, requires extensive debate before signoff"

trailfox 13 years ago |

Wow. Wish I could deploy this sort of tooling behind a firewall.

nivstein 13 years ago | |

The agent only requires outbound HTTPS over port 443, and fully supports proxies.

obilgic 13 years ago | | |

I guess he is talking about private deployment for his company like github enterprise(github:firewall)

trailfox 13 years ago | | |

Bingo :)

wheaties 13 years ago | |

Ditto, just sent an email asking about this. There's a number of industries which frown upon allowing any sort of code or production data to be sent to a 3rd party host.

koker 13 years ago |

Going to give it a go on our servers. We have a relentless Heisenbug we've been chasing down lately, which only happens in production.

protomyth 13 years ago |

Many of the Smalltalks allow hooking to a running image to do live debugging.

seanmcdirmid 13 years ago | |

No they don't. They just allow for fix and continue; they give you no insight into why a problem occurred because they provide no history.

protomyth 13 years ago | | |

uhm... that's not universally true - I worked with some folks that had the history when they did it. Not sure if it was vendor provided or home grown, but it was the call history and call stack.

seanmcdirmid 13 years ago | | |

Call stack is provided in most debuggers these days. Are you honestly saying you saw something comparable to the linked article in some Smalltalk system? If so, I would be very interested in some kind of reference.

protomyth 13 years ago | | |

It was a VMWare debug image and it had all that information although it was not as fancy. It did have a good amount of custom code.

zaptheimpaler 13 years ago |

This looks awesome! Is there anything like this for local code though? The IntelliJ/Eclipse debuggers don't seem as well designed as this one.

nivstein 13 years ago | |

Takipi may be installed locally; the same installation process will work. However, it does not provide live debugging like in your IDE, but rather a post-event drill-down (a "time travelling debugger"), as it is designed for testing/staging/deployment environments.

orenbarzilai 13 years ago |

What about python?

spatz 13 years ago | |

From the FAQ:

  Which languages do you support?

  Takipi works at the JVM level. This means we support all languages running on the Java Virtual Machine. These include - Java, Scala, Groovy, Clojure, etc.. Click here to learn more about viewing Scala code in Takipi.
  We plan to support more software VMs such as CPython and Ruby MRI in the future.

thejash 13 years ago | |

Just FYI, most of this is a lot easier in python. You can get the full stack trace and all of the values of all of the variables when there is an exception, something that is orders of magnitude more difficult to do with Java/Scala.

You'd have to be a bit creative to get the "full variable value history" stuff working, but it would probably be possible with some hacking on the setattr methods for your objects (assuming you have full control over the system)

I had a bunch of code like that for my previous (python) projects. I've really missed it now that I've moved to scala/java, so I'm pretty excited for this project.

eli_gottlieb 13 years ago |

You've released? MAZAL TOV! I've been stalking you guys for a while now.

irisshoor 13 years ago | |

Thanks, you're welcome to take Takipi for a test drive and let us know what you think :)

drorweiss 13 years ago |

Looks great!

szeevim 13 years ago |

likes the monsters

stasix 13 years ago |

looks promising.

vovafeldman 13 years ago |

Looks cool!

bizkeep2 13 years ago |

US-based?

baboo 13 years ago |

"to log and send data to Takipi's analysis servers" ???!!!

Is this for real?

I'm surely going to send all my private code, data and who knows what else to your servers! I have been waiting for this since forever, and I'm glad to have this opportunity now!

And... it slows everything by 5% constantly?!? To basically take a core dump for later analysis? (in case it's not obvious, such a thing should have no overhead whatsoever when fatal exceptions are not thrown)

Lol, is it 1st April again?

hmottestad 13 years ago | |

You know, preferably it should actually make your servers run 5% faster :)

Seriously though. Did you really expect a tool that continuously monitors your app and can produce stunning and relevant information about states leading up to a crash to have no effect on performance?

zer0gravity 13 years ago |

Tesla said that he could simulate the behaviour of his contraptions to the smalles detail in his head and without exception, they ran that way once he built them for real.

I think that a good programmer should be able to do this with his programms, and without flattering myself I can say that in 99% of the cases when an error happened, the answer occured to me almost instantly or after a few moments of thinking and trying to simulate what happened in my head. I have to confess that these were instances when the whole code was mine. Working on code you don't really understand is a major mistake. But this is another issue..

Sure it remains the 1%.. but my point is that developing the thinking abbilities instead of debugging tools is a far better alternative.