# Hot/Warm architecture and Machine-Learning

**URL:** <https://discuss.elastic.co/t/hot-warm-architecture-and-machine-learning/161707>\
**Category:** Elasticsearch\
**Tags:** elastic-stack-machine-learning\
**Created:** [December 20, 2018, 2:37pm UTC](https://discuss.elastic.co/t/hot-warm-architecture-and-machine-learning/161707 "2018-12-20T14:37:32Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![Rohlik](https://avatars.discourse-cdn.com/v4/letter/r/aeb1de/32.png) [@Rohlik](https://discuss.elastic.co/u/Rohlik)\
**Post date:** [December 20, 2018, 2:37pm UTC](https://discuss.elastic.co/t/hot-warm-architecture-and-machine-learning/161707/1 "2018-12-20T14:37:32Z")

</div>

Hi there,  
How ML node works in How/Warm data node architecture? We want store aprox. 6 months on Hot data nodes and aprox. 18 months on Warn nodes. Because all data will go through hot data nodes, is it necessary to have ML node also connected to warm nodes for right anomaly detection?

---

<div class="post-metadata">

**Author:** ![richcollier](https://sea2.discourse-cdn.com/elastic/user_avatar/discuss.elastic.co/richcollier/32/115035_2.png) [@richcollier](https://discuss.elastic.co/u/richcollier)\
**Post date:** [December 20, 2018, 4:08pm UTC](https://discuss.elastic.co/t/hot-warm-architecture-and-machine-learning/161707/2 "2018-12-20T16:08:55Z")

</div>

As long as all data that you want ML to model/analyze is searchable, it will be fine.

In most cases, once you've run ML on some historical data to help build the models, ML runs on "real-time" data (data in the last X minutes, for example). In this mode, ML will only be routinely looking at data in your hot nodes.

---

<div class="post-metadata">

**Author:** ![Rohlik](https://avatars.discourse-cdn.com/v4/letter/r/aeb1de/32.png) [@Rohlik](https://discuss.elastic.co/u/Rohlik)\
**Post date:** [December 27, 2018, 8:48am UTC](https://discuss.elastic.co/t/hot-warm-architecture-and-machine-learning/161707/3 "2018-12-27T08:48:31Z")

</div>

@richcollier Thank you for you answer. I have one additional question.  
Do we need license for warm data nodes If our solution will be fresh new cluster (ML, hot/warm data nodes, etc...) with no historical data for ML to be functional?

PS: Happy new year!

---

<div class="post-metadata">

**Author:** ![Christian\_Dahlqvist](https://sea2.discourse-cdn.com/elastic/user_avatar/discuss.elastic.co/christian_dahlqvist/32/4617_2.png) [@Christian\_Dahlqvist](https://discuss.elastic.co/u/Christian_Dahlqvist)\
**Post date:** [December 27, 2018, 9:20am UTC](https://discuss.elastic.co/t/hot-warm-architecture-and-machine-learning/161707/4 "2018-12-27T09:20:58Z")

</div>

The license need to cover all nodes in the cluster.

---

<div class="post-metadata">

**Author:** ![system](https://us1.discourse-cdn.com/elastic/original/3X/1/a/1ac57faf039f6b580b3f104ef42a2a89e41014de.png) [@system](https://discuss.elastic.co/u/system)\
**Post date:** [January 24, 2019, 9:21am UTC](https://discuss.elastic.co/t/hot-warm-architecture-and-machine-learning/161707/5 "2019-01-24T09:21:06Z")

</div>

Closed on January 24, 2019, 9:21am UTC

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.