Hacker News Reader

Comment by beepbeepbeep1 | Hacker News Reader

[redacted by author request]

Atlassian employee here. JIRA's backend is not Node based. It's mostly Java. We do use Node for other products, including Trello, and it's more than fast enough.

I think of Atlassian as the Trump of software.

It’s slow, annoying UX, and very inconsistent.

But for some reason businesses like it, it is what it is and as an employee I have little choice so I have to bear with it.

> Jira is 10 years of technical debt

> Cloud was full buy in to AWS...

> everyone is using nodejs and your language is now banned...

> Add internal politics... hire as many people as possible and terminated anyone raising valid concerns...

> monolith service now becomes highly distributed...

As some comments mentioned before, there are perfectly good software written in NodeJS (trello) which are blazingly fast. Similarly, there are enough good, well performant multi-tenanted applications on AWS. Anything that has been used for a long time (linux, java, protocols, etc) is always going to carry tech debt and there is nothing wrong in that.

While there may be a grain of truth in some corner cases of what is written, it overall comes across as an emotionally negative let off, and in some places borderlining on indirect propaganda against generic stuff (sorry).

In this day and age, languages, frameworks, and libraries that deal poorly with parallelism deserve criticism for it.

Yes, competent engineers can work around these problem given time (e.g. pivoting from thread-level to process-level parallelism), so it wouldn't be fair to entirely blame nodejs for the problems, but that wasn't the vibe I got from GP.

Finally some sort of explanation. Thee feature set is great for complex use cases but the cloud version is unusably slow. Worse the support team seems trained for pretend like it's not a problem. The last web app I remember being as slow as Jira is Friendster.

The hyper distributed micro services are definitely one of the big reasons for unnecessary complexity and latency problems. It was surprising for me to see that the JVM based services were load tested for 4-10 minutes to call the test successful.

Yeah it's odd as some very clever talented people have passed through there doors, some very mixed feelings towards them.

The JVM is just getting warmed up load testing for that short a period.

Do you know of any sources that show it takes that long to warm up JVM?

It has nothing to do with time and everything to do with the number of times hot paths get executed.

On a server JVM instance a method will have to be executed (by default) 10 thousand times before it is compiled.

There is also a question of how stable the compilation profile is and how likely deoptimizations are but that's a different story.

> nodejs because everyone is using nodejs and your language is now banned it's not surprising there's noticeable performance differences, it was certainly predicated early on but they where to caught up trying to be google and using the popular language to hire as many people as possible

I wonder how typical is this nowadays in IT companies and departments all around the world.

At least somewhat.

I'm aware of a nodejs script that is basically packed up with the node runtime and run on Windows as a "native application," which interacts with the filesystem, windows registry, and more. The packer is years old and nobody knows how it works.

It is everything you expect of such a program. From both a user and developer perspective.

It is deployed on very expensive, very important equipment in pharma labs and hospitals.

This is less uncommon than people like to admit. I've seen, and in some cased directly involved with, code written by an inexperienced (giving benefit of doubt here) developer whose position was replaced twice over, that runs critical systems on big-money (but not life-threatening) equipment and business systems. No documentation, no tests, even one that relies on a known platform bug to never be fixed (not kidding). Nobody knows how it works, and there is no time go go through it and clean it up (business decision).

I've even stated in writing on one system that I refuse to receive "emergency" calls on holidays, weekends, or nights to fix a particular system if it ever goes down at that time. Management refusal to plan ahead or heed warnings does not constitute an emergency on my part.

> Add internal politics where some things which are heavily multi threaded built in staticly compiled languages had to be built in single threaded nodejs because everyone is using nodejs and your language is now banned it's not surprising there's noticeable performance differences, it was certainly predicated early on but they where to caught up trying to be google and using the popular language to hire as many people as possible and terminated anyone raising valid concerns i.e it's slow and node wasn't right.

What languages were banned? This is a very interesting data point.

No languages were banned. We use Java, Python, Node, and Golang at Atlassian, possibly others, too.

FYI, I have huge problems with your products at work. We use a variety of your products and competing products, on a project-by-project basis. I always advocate for your competitors.

One reason is that you made the business decision to attempt customer lock-in. You supported wikimedia-style text markup in your wiki (in addition to the GUI) so that people could migrate to your stuff, and then you took out that feature so that people would have trouble leaving. I'm sure that makes sense to an MBA, but I have been discouraging use of your products all throughout a large company.

The other reason is that yes, all your stuff is slow as fuck. OMG it is slow. The Java grows to consume gigabytes for no damn reason, and it munches CPU time, and generally it sucks pretty hard.

Golang might be right for you. I think it's the fastest choice available for development teams that can't handle stuff like pointers. On the other hand, I think you would still manage to sort-of-leak memory by hanging on to references that you really don't need.

I think you could fix the slowness problem by requiring all development and testing to be done on computers that are slower than the ones your customers use. Get an old Pentium II with 256 MiB of RAM... which is still overkill for the task at hand. Remember, back in the day we ran stuff like your products on computers with 8 MiB or less and a 486 or less. You can live with a Pentium II and 256 MiB of RAM, and the resulting performance of your software will delight your customers.

[redacted by author request]

I'm skeptical that there isn't another side of this story or at least less absolute terms used.

For example, maybe there's a push against everyone using all their own favourites everywhere. I can see a strong argument for, "please stop writing in Foo. We use Java, Go, Python, and C++. Pick the right one of those for the job at hand. We all benefit from using a common set of tools."

I'm not denying what you're saying. Just feeling a healthy skepticism that they "ban" languagues without a good faith objective.

I’d probably stop talking on this thread if I were you. If your management chain sees this, you’ll have at minimum a conversation about using judgement when speaking in a public forum, at worst, your comments might be used as justification for disciplinary action.

His management has already nudged him to find a new job.

Last comment

This was a bit back but not to far back.

Took a significant pay rise. elsewhere and got to work with some other ex atlassian colleagues.

Disparaging your former employer might also be against your employment agreement and doesn’t make you look good either. Just giving a friendly heads up!

Does Google even use nodejs or try to hire devs for 'popular languages'? Last I heard everything is still mostly C/C++, Python and a little bit of Go.

Google is one of the bigger Java shops out there. That, if anything, is their first-and-foremost language outside of search.

Reply on news.ycombinator.com