Improve search and fix '*' in search box (#871)

* Improve ES search

The new performance is very good.
Some examples on my 1.5gb vm:
INFO[0153] Query 'shingeki' took 6 milliseconds.
INFO[0125] Query 'アニメ' took 17 milliseconds.
INFO[0102] Query 'shingeki -kyojin horrible ' took 12 milliseconds

Looking at the criteria we wanted here:
https://pad.riseup.net/p/i8DrilHDWRvf, it meets:

1. Fast: sub-100ms for a typical query, sub-50ms is good and sub-20ms is
optimal
2. Prefix match: "horrible" finds horriblesubs
3. Substring match? "アニメ" finds "TVアニメ"
4. Position-independent terms ("shingeki kyojin" finds the same as
"kyojin shingeki")
5. Works with short term lengths correctly and fast (no in "kyoukai no
kanata", 04 in "horrible shingeki 04" etc)
7. (nice to have) search negation: shingeki kyojin -horriblesubs

* Use match_all query instead of *, fix *

common/torrent.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"net/http"
 	"strconv"
+	"strings"
 
 	"github.com/gorilla/mux"
 	elastic "gopkg.in/olivere/elastic.v5"
@@ -38,10 +39,7 @@ type TorrentParam struct {
 func (p *TorrentParam) FromRequest(r *http.Request) {
 	var err error
 
-	nameLike := r.URL.Query().Get("q")
-	if nameLike == "" {
-		nameLike = "*"
-	}
+	nameLike := strings.TrimSpace(r.URL.Query().Get("q"))
 
 	page := mux.Vars(r)["page"]
 	pagenum, err := strconv.ParseUint(page, 10, 32)
@@ -143,10 +141,15 @@ func (p *TorrentParam) Find(client *elastic.Client) (int64, []model.Torrent, err
 	// TODO Why is it needed, what does it do ?
 	ctx := context.Background()
 
-	query := elastic.NewSimpleQueryStringQuery(p.NameLike).
+	var query elastic.Query
+	if p.NameLike == "" {
+		query = elastic.NewMatchAllQuery()
+	} else {
+		query = elastic.NewSimpleQueryStringQuery(p.NameLike).
 			Field("name").
 			Analyzer(config.Conf.Search.ElasticsearchAnalyzer).
 			DefaultOperator("AND")
+	}
 
 	fsc := elastic.NewFetchSourceContext(true).
 		Include("id")

config/default_config.yml

@@ -34,8 +34,8 @@ cache:
 # Size by default for the cache
   size: 1024
 search:
-# default analyzer for ES
-  es_analyze: nyaapantsu_analyzer
+# default search analyzer for ES
+  es_analyze: nyaapantsu_search_analyzer
 # default search index for ES
   es_index: nyaapantsu
 # Name of the type in the es mapping

deploy/ansible/group_vars/all

@@ -4,7 +4,6 @@ nyaapantsu_password: nyaapantsu
 nyaapantsu_pgpool_port: 9998
 nyaapantsu_directory: /nyaapantsu/
 nyaapantsu_gpg_passphrase_file: "{{ nyaapantsu_directory }}/passphrase"
-nyaapantsu_elasticsearch_index: nyaapantsu
 nyaapantsu_build_directory: go_nyaa/
 nyaapantsu_elasticsearch_alias: nyaapantsu
 # nyaapantsu_elasticsearch_alias: sukebei

deploy/ansible/roles/elasticsearch/files/elasticsearch_settings.yml

@@ -2,19 +2,37 @@
 settings:
   analysis:
     analyzer:
-      nyaapantsu_analyzer:
-        tokenizer: nyaapantsu_tokenizer
+      # Don't use ngram for search otherwise 'horribleexample' would match
+      # 'horriblesubs'
+      nyaapantsu_search_analyzer:
+        tokenizer: standard
         filter:
           - standard
           - lowercase
-    tokenizer:
-      nyaapantsu_tokenizer:
+        char_filter:
+          - dash_to_underscore
+
+      nyaapantsu_index_analyzer:
+        tokenizer: standard
+        filter:
+          - standard
+          - lowercase
+          - e_ngram_filter
+        char_filter:
+          - dash_to_underscore
+
+    filter:
+      e_ngram_filter:
         type: edge_ngram
         min_gram: 2
         max_gram: 15
-        token_chars:
-          - letter
-          - digit 
+
+    char_filter:
+      dash_to_underscore:
+        type: pattern_replace
+        pattern: "([^\\s]+)-(?=[^\\s]+)"
+        replacement: "$1_"
+
   index:
     number_of_shards: 1
     number_of_replicas: 0
@@ -28,8 +46,10 @@ mappings:
         type: long
       name:
         type: text
-        analyzer: nyaapantsu_analyzer
-        fielddata: true # Use to sort by id because it is currently a text field
+        analyzer: nyaapantsu_index_analyzer
+        fields:
+          raw:
+            type: keyword
       category:
         type: text
       sub_category: