Elasticsearch first query is slow, rest of them are fast - elasticsearch

Question

I'm using this kind of mapping (well, it's a shortener version in order to make the question easier) on a children-parent relationship where item is the parent and user_items is the children.
curl -XPUT 'localhost:9200/myindex?pretty=true' -d '{
"mappings": {
"items": {
"dynamic": "strict",
"properties" : {
"title" : { "type": "string" },
"body" : { "type": "string" },
}},
"user_items": {
"dynamic": "strict",
"_parent": {"type": "items" },
"properties" : {
"user_id" : { "type": "integer" },
"source_id" : { "type": "integer" },
}}}}'
And the type of query I usually make:
curl -XGET 'localhost:9200/myindex/items/_search?pretty=true' -d '{
"query": {
"bool": {
"must": [
{
"query_string": {
"fields": ["title", "body"],
"query": "mercado"
}
},
{
"has_child": {
"type": "user_items",
"query": {
"term": {
"user_id": 655
}}}}]}}}'
On this query it has to search on the fields title and body the string mercado on a given user_id, in this case 655.
I read that the reason of being so slow the first query is that it gets cacheed and then the rest queries are fast because it works with the cached content.
I read I can make the first query faster using eager to preload my data (using "loading" : "eager", right?) but I dont know what do I've to preload. Do I've to use the earger on title and body as follows?
{
"mappings": {
"items": {
"dynamic": "strict",
"properties" : {
"title" : { "type": "string" ,
"fielddata": {
"loading" : "eager"}},
"body" : { "type": "string",
"fielddata": {
"loading" : "eager"}},
}},
"user_items": {
"dynamic": "strict",
"_parent": {"type": "items" },
"properties" : {
"user_id" : { "type": "integer" },
"source_id" : { "type": "integer" },
}}}}'
Any other recommendation fot boosting/cacheeing the first query is welcome. Thanks in advance
PS: I'm using ES 2.3.2 under a Linux machine and I've a total of 25.396.369 documents.

Answer 1

Just fixed the same issue. Field Data preloading was the key. Index warming was deprecated and doc_values is on by default. My application searched a couple fields on a large index (100G+) and was slow I had to rebuild the index with loading=eager for all of the fields that I searched on. This preloads it and causes a pretty long startup but after that search went from initial of 10s (<400ms afterwards) to <900ms initial search (<400ms afterwards). Make the mapping and reimport the data
PUT localhost:9200/newindex/
{
"mappings": {
"items": {
"properties": {
"title": {
"type": "string",
"fielddata": {
"loading" : "eager"
}
},
"body": {
"type": "string",
"fielddata": {
"loading" : "eager"
}
}
}
}
}
}

Answer 2

There are three things you can do.
Use field data preloading
To preload field data use following snippet in mapping
"fielddata": {
"loading" : "eager"
}
More details here
Use index warmer
Index warmers are certain queries that you can configure which will run automatically whenever a index is refreshed.
This link contains details on how to set up a warmer.
Use doc_values
Doc values are the on-disk data structure, built at document index time, which makes data access pattern possible for aggregation and sorting possible.
Find more details here