Why hasn't this repo been updated in the last 10 months?

[draco2003]

If i had to guess it's because their efforts are being put else where: www.andrewclegg.org/tech/KaleTalk.html

maybe @andrewclegg could give us some insight into Kale v2 or access to the code :)

[wyaeld]

Its not exactly a shining example of open source when you abandon the v1, ignore the community, and build a v2 in secret.

@MyNameIsMeerkat @draco2003 @jonlives as the 3 Etsy employees in the top 10 committers from what I can see, are any of you able to shed some light on the situation?

[jonlives]

Hi folks,

I can certainly give you a bit of an update here - a much lengthier blogpost on the subject is currently being worked on (by me) for release hopefully later this month or early next month, but here's the gist.

At the point Oculus and Skyline were released, they were still relatively experimental tools internally - we found fairly quickly that the "one size fits all" approach we were taking to anomaly detection wasn't actually proving all that useful to us. Earlier this year, our very own John Allspaw wrote about this very thing in http://www.kitchensoap.com/2015/05/01/openlettertomonitoringproducts/.

The fact that the tools weren't being heavily used internally, combined with an extremely busy year, brought development work to a crawl, despite the fact that the tools were already open sourced. Earlier this year we sat down to decide what to do about this as the core ideas around those tools still generated a lot of excitement and interest.

The approach we're taking this time around is based on learning from the mistakes we made the first time - specifically, we're not trying to treat "anomaly detection" in its entirety as a solved problem. The v2 that @andrewclegg is working on is actually the very first component of this new approach - Andy's working on a toolkit called Thyme which will make it easy to plug together different algorithms and approaches to both anomaly detection and correlation which we then plan to try using to solve discrete, smaller problems - for example, "can we alert to unusual behaviour on any of our top 5 site pages".

This is absolutely not being developed in secret - Andy's demoed and talked about his work at SRECon in Europe, and at Monitorama yesterday, and the aim is to release Thyme when it's functionally complete.

In more general terms of open sourcing what we've done, truthfully we're being a little more careful with what we open source and when this time, for the specific reason that we don't want to open source anything this time around we haven't actually found useful in production.

Going forward, Oculus and Skyline v1 are not going to be taken down, but will most likely be updated to state that they are presented "as-is" and will not be actively worked on. We realise this isn't an ideal solution, and in hindsight (which is always 20/20) we should have waited to release those tools until we'd seen them being directly useful to us in production.

Hope this clears things up a bit :)

Cheers,

Jon

[wyaeld]

Thanks Jon, really appreciate the clarification. Interested in seeing what you guys come up with next.

On Thu, Jun 18, 2015 at 9:19 PM, Jon Cowie notifications@github.com wrote:

Hi folks,

I can certainly give you a bit of an update here - a much lengthier blogpost on the subject is currently being worked on (by me) for release hopefully later this month or early next month, but here's the gist.

At the point Oculus and Skyline were released, they were still relatively experimental tools internally - we found fairly quickly that the "one size fits all" approach we were taking to anomaly detection wasn't actually proving all that useful to us. Earlier this year, our very own John Allspaw wrote about this very thing in http://www.kitchensoap.com/2015/05/01/openlettertomonitoringproducts/.

The fact that the tools weren't being heavily used internally, combined with an extremely busy year, brought development work to a crawl, despite the fact that the tools were already open sourced. Earlier this year we sat down to decide what to do about this as the core ideas around those tools still generated a lot of excitement and interest.

The approach we're taking this time around is based on learning from the mistakes we made the first time - specifically, we're not trying to treat "anomaly detection" in its entirety as a solved problem. The v2 that @andrewclegg https://github.com/andrewclegg is working on is actually the very first component of this new approach - Andy's working on a toolkit called Thyme which will make it easy to plug together different algorithms and approaches to both anomaly detection and correlation which we then plan to try using to solve discrete, smaller problems - for example, "can we alert to unusual behaviour on any of our top 5 site pages".

This is absolutely not being developed in secret - Andy's demoed and talked about his work at SRECon in Europe, and at Monitorama yesterday, and the aim is to release Thyme when it's functionally complete.

In more general terms of open sourcing what we've done, truthfully we're being a little more careful with what we open source and when this time, for the specific reason that we don't want to open source anything this time around we haven't actually found useful in production.

Going forward, Oculus and Skyline v1 are not going to be taken down, but will most likely be updated to state that they are presented "as-is" and will not be actively worked on. We realise this isn't an ideal solution, and in hindsight (which is always 20/20) we should have waited to release those tools until we'd seen them being directly useful to us in production.

Hope this clears things up a bit :)

Cheers,

Jon

— Reply to this email directly or view it on GitHub https://github.com/etsy/skyline/issues/119#issuecomment-113085556.

[srgrn]

First, this is an amazing tool and it seems that everywhere i look to learn about monitoring etsy comes up in the results.

now i know it is a couple of months too late, but does your response means that all the pull requests that are waiting will not be merged back?

[jeremyquinton]

@jonlives just curious as to if you every released a blog post. If you did can you post it here please :-)

[BoltzmannBrain]

I think you all would be interested in the Numenta Anomaly Benchmark (NAB), a benchmark we've created for evaluating online anomaly detection algorithms. It's a dataset of real-world data streams and a custom scoring methodology. We compare Etsy Skyline and several other algorithms, including our own HTM anomaly detector--available open source in NuPIC.