Dots in Fieldnames => ES 2.X Upgrade will fail?

Elastic Stack Elasticsearch
1

Hi guys,

i apologize if this is 1242032nd thread to this particular upgrade topic.

After getting rid of the different mapping-types of fields in different ES-types, i now got the "problem" to have some fields with dots in it in almost every index, mostly completely nonsense fields that got produced by an not 100% working Logstashfilter.

Migration-Tool is telling me after the last Update that it is a big issue, however a long time it displayed the "dot-fields" as a yellow error - meaning it is not optimal and those fields wont continue to work in ES2.0.

When i upgrade my ES 1.7 to 2.1 or 2.3, will the upgrade process fails cause of this annoying fields or will there be a warning displayed?

If it fails, why? Seems like ES 2.0 can deal with dot fields, as it does e.g. with the ".raw" multi fields.

It would be a work of months to reindex all of the indices, so i hope i could get around it :slight_smile:

Thanks for your response.

2

You can not get around it. ES uses dots as internal field name separators.

ES 1 produced an inconsistency between

"a" : {
  "b" : "c"
}

and

"a.b" : "c"

and

"b" : "c"

when referencing field name a.b or short form b in a query.

ES2 resolves this to the former case and enforces full field names in the mapping.

3

Thanks for your detailed answer.

Too bad ES dont just ignore that inconsistent fields on upgrade/mapping.

But it wont help than, sigh

4

What I would LOVE is an option to replace the dots in fields upon snapshot restore/upgrade.

At the moment I'm having to write a program to go through and re-index all documents using fields without dots so that we can later upgrade.

This makes me sad, especially when: http://stackoverflow.com/questions/38615227/elasticsearch-allows-duplicate-id-with-different-body-data