Is performance affected by indexing naming scheme?

[manbearpig]

Supposed I have multiple identical sets of hardware, and the same index settings (same shard, same replica). They are both ingesting the same type of data.

With set A, I only have 1 index and 1 type, e.g. /myIndex/myType

With set B, I have a scheme where the index is separated by dates e.g. /indexDate_xxxx/myType where xxxx is month and year for example /indexDate_0117 for January 2017, /index_Date_0217 for February 2017

My question is would there be any performance difference between set A and set B?

[tanguy]

Hi,

As always, it depends. With a small set of documents, you should not notice any performance difference between solution A and B. But both solution won't scale in the same way and both have pro and cons.

I strongly encourage you to read the "Designing for Scale" chapter of the Definitive Guide (see https://www.elastic.co/guide/en/elasticsearch/guide/current/scale.html) which is short and have all the answers.

Also, I'm curious to know the lifecycle of the documents: are they stored and queried forever? Are they deleted after a given period of time? Are they updated?

[manbearpig]

Wow thank for you for the prompt reply!

Thank you for the recommendation. I will be dealing with very large set of documents. In particular emails and chats.

They will be stored and queried forever and they are seldom updated.

Thank you again