Why index in 5.0 is bigger than that in 1.3?


(chenjinyuan87) #1

Hello:
I've upgraded my cluster to 5.0 these days.
And i found the index size is bigger than 1.3.
The following is the index metadata in 5.0:

"_all": {
    "search_analyzer": "default_search",
    "analyzer": "default",
    "enabled": false
    },
    "properties": {
    "heat": {
    "type": "integer"
    },
    "_nid": {
    "type": "keyword",
    "doc_values": false
    },
    "channel": {
    "type": "keyword",
    "doc_values": false
    },
    "_tid": {
    "type": "long"
    },
    "title": {
    "index": false,
    "type": "keyword"
    }
    }

And the metadata in 1.3:

 "_all": {
    "enabled": false
    },
    "properties": {
    "heat": {
    "type": "integer"
    },
    "_nid": {
    "index": "not_analyzed",
    "type": "string"
    },
    "channel": {
    "index": "not_analyzed",
    "type": "string"
    },
    "_tid": {
    "type": "long"
    },
    "title": {
    "index": "no",
    "type": "string"
    }
    }

Thanks


(Adrien Grand) #2

How much bigger?

One difference is that your 5.x index has doc values enabled on the heat, tid and title fields since the default was changed in 2.0.


(chenjinyuan87) #3

In 1.3, it's about 500M, and know 750M..
I thought it will be smaller in 5.0


(Adrien Grand) #4

High cardinality numeric fields can indeed be expected to compress a bit better, but here you have more features enabled compared to the 1.3 index so we are comparing apples to oranges. One good news is that the fact that you have doc values enabled, in spite of increasing disk usage, should significantly decrease memory usage when aggregating or sorting on those fields.


(chenjinyuan87) #5

doc values in title field is not needed.
I thought it was disabled in not indexed fields.
Thanks for your help


(system) #6

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