spandex

Elasticsearch new low level rest-client wrapper

Why?

To quote "State of the official Elasticsearch Java clients"

The Java REST client is the future for Java users of Elasticsearch.

Because the legacy native client is a bit of a nightmare to deal with (for many reasons) and the new REST client is quite capable and fast too, see "Benchmarking REST client and transport client"

Not to mention it supports some interesting features:

• compatibility with any Elasticsearch version

• load balancing across all available nodes

• failover in case of node failures and upon specific response codes

• failed connection penalization

• persistent connections

• trace logging of requests and responses

• optional automatic discovery of cluster nodes (also known as sniffing)

Goals

• Be minimal & performant

• RING inspired

• All "exotic" features should be optional

• Not a giant DSL over another DSL, just maps everywhere. Read ElasticSearch doc -> done, not another layer of indirection

• Provide minimal (and optional) utils to do the boring stuff (bulk, scroll queries, compose urls)

• Can do async via simple callbacks based api or core.async

• Provide specs

Setup

(require '[qbits.spandex :as s])

(def c (s/client {:hosts ["http://127.0.0.1:9200" "https://foo2:3838"]}))

;; add optional sniffer
(def s (s/sniffer c {... options ...}))

Constructing URLs

Most of spandex request functions take a request map as parameter. The :url key differs a bit from the original RING spec, it allows to pass a raw string but also a sequence (potentially 2d) of encodable things, keywords, .toString'able objects that make sense or nil (which could be caused by a missing :url key).

(s/request c {:url [:foo :bar :_search] ...})
(s/request c {:url [:foo [:bar :something "more"] :_search] ...})
(s/request c {:url :_search ...})
(s/request c {:url "/index/_search" ...})
(s/request c {:url (java.util.UUID/randomUUID) ...})
(s/request c {...}) ;; defaults to "/"

Blocking requests

(s/request c {:url [:entries :entry :_search]
              :method :get
              :body {:query {:match_all {}}}})

>> {:body {:_index "entries", :_type "entry", :_id "AVkDDJvdkd2OsNWu4oYk", :_version 1, :_shards {:total 2, :successful 1, :failed 0}, :created true}, :status 201, :headers {"Content-Type" "application/json; charset=UTF-8", "Content-Length" "141"}, :host #object[org.apache.http.HttpHost 0x62b90fad "http://127.0.0.1:9200"]}

Async requests (callbacks)

(s/request-async c {:url "/urls/url/"
                    :method :get
                    :body {:query {:match {:message "this is a test"}}}
                    :success (fn [response-as-clj] ... )
                    :error (fn [ex] :boom)})

Async requests: core.async/promise-chan

(async/<!! (s/request-chan c {:url "/urls/url/"
                              :method :get
                              :body {:query {:match {:message "this is a test"}}}}))

Scrolling

Scrolling via core.async (fully NIO internally), interruptible if you async/close! the returned chan.

(async/go
  (let [ch (s/scroll-chan client {:url "/foo/_search" :body {:query {:match_all {}}}})]
    (loop []
      (when-let [page (async/<! ch)]
        (do-something-with-page page)
        (recur)))))

Bulk requests scheduling

"Faux streaming" of _bulk requests (flushes bulk request after configurable interval or threshold. Uses request-chan internally, so it's quite cheap.

(let [{:keys [input-ch output-ch]} (bulk-chan client {:flush-threshold 100
                                                      :flush-interval 5000
                                                      :max-concurrent-requests 3})]
  ;; happily takes a sequence of actions or single fragments
  (async/put! input-ch [{:delete {:_index "foo" :_id "1234"}} {:_index :bar} {:create {...}}])
  (async/put! input-ch {"delete" {"_index" "website" "_type" "blog" "_id" "123"}}))

;; setup an response consumer (we just want to make sure we don't clog this channel)
(future (loop [] (async/<!! (:output-ch c))))

Installation

API Docs

Or the clj.specs if that's your thing:

Patreon

If you wish to support the work on this project you can do this via my patreon page.

License

Copyright © 2018 Max Penet

Distributed under the Eclipse Public License, the same as Clojure.