Elasticsearch 6.0 _id and size_in_bytes

Elastic Stack Elasticsearch
1

I upgraded to 6.0 because of sequental ids. I expected that after switching to sequental _id it will saves memory.
In my case I have per day index with lots of events with small amount of fields (like ip_src, ip_dst, port_dst). I noticed that _id cosumes lots of memory. Is it possible to optimese _id field or somehow disable it ? If id is sequental I expect that it some sort of memoty offset could be calculated per search request and it is not requred to store it in memory

This is part of statistics:
{
"description" : "field '_id' [BlockTreeTerms(seg=_32b terms=531218720,postings=531218720,positions=-1,docs=531218720)]",
"size_in_bytes" : 78714997,
"children" : [
{
"description" : "term index [FST(input=BYTE1,output=ByteSequenceOutputs]",
"size_in_bytes" : 78714837
}
]
},
{
"description" : "field 'ip_dst' [BlockTreeTerms(seg=_32b terms=255005,postings=519813549,positions=-1,docs=519813549)]",
"size_in_bytes" : 68845,
"children" : [
{
"description" : "term index [FST(input=BYTE1,output=ByteSequenceOutputs]",
"size_in_bytes" : 68685
}
]
},

2

In Elasticsearch 6.0 all operations get a sequence id, which can help speed up recovery. The logic for generating document ids is not affected by this (they are not sequential).

3

How do you think if _id were sequental were memory consumption lower? What is a reason why _id are not numeric?

4

Trying to assign sequential numeric ids automatically, does generally not scale or perform in a distributed, highly concurrent system. If you want to, you can however assign your own id at the application layer.