Use POST method for search requests with body

[gleb-svitelskiy]

Summary of problem or feature request

GET requests with a body are wrong for some proxies. POST request type works. GET requests can't contain the body, that is the reason.

Code snippet of problem

use Elasticsearch\ClientBuilder;
...
$client = ClientBuilder::create()->setHosts(['http://elastic:password@localhost/'])->build();
            $query = [
                "body" => [
                    [
                        'index' => 'index_name',
                        'type' => 'type_name',
                        'timeout' => '3s',
                    ],
                    [
                        'query' => [
                            'match_all' => []
                        ],
                        "size" => 1
                    ]
                ]
            ];
$client->msearch($query);

-->

System details

• Operating System CentOS 6.9
• PHP Version 7.1.3
• ES-PHP client version v6.0.1
• Elasticsearch version 6.2

[gleb-svitelskiy]

[gleb.svitelskiy@dev.localhost ~]$ curl -XGET -H 'Content-Type: application/json' -o /dev/null -sSL -D - http://elasticsearch.localhost:9200/test_index/_search?pretty -d '
> {
>   "size": 1,
>   "_source": false
> }
> '
HTTP/1.1 411 Length Required
Server: squid
Mime-Version: 1.0
Date: Tue, 27 Mar 2018 06:23:52 GMT
Content-Type: text/html;charset=utf-8
Content-Length: 4547
X-Squid-Error: ERR_INVALID_REQ 0
Vary: Accept-Language
Content-Language: en
X-Cache: MISS from localhost.localdomain
X-Cache-Lookup: NONE from localhost.localdomain:3128
Via: 1.1 localhost.localdomain (squid)
Connection: close

[gleb.svitelskiy@dev.localhost ~]$ curl -XPOST -H 'Content-Type: application/json' -o /dev/null -sSL -D - http://elasticsearch.localhost:9200/test_index/_search?pretty -d '
> {
>   "size": 1,
>   "_source": false
> }
> '
HTTP/1.1 200 OK
Date: Tue, 27 Mar 2018 06:23:58 GMT
Content-Type: application/json; charset=UTF-8
Content-Length: 354
X-Cache: MISS from localhost.localdomain
X-Cache-Lookup: MISS from localhost.localdomain:3128
Via: 1.1 localhost.localdomain (squid)
Connection: keep-alive

[gleb.svitelskiy@dev.localhost ~]$ curl -XGET -H 'Content-Type: application/json' -o /dev/null -sSL -D - 'http://elasticsearch.localhost:9200/test_index/_search?pretty=pretty&size=1'
HTTP/1.1 200 OK
Date: Tue, 27 Mar 2018 06:41:25 GMT
Content-Type: application/json; charset=UTF-8
Content-Length: 2593
X-Cache: MISS from localhost.localdomain
X-Cache-Lookup: MISS from localhost.localdomain:3128
Via: 1.1 localhost.localdomain (squid)
Connection: keep-alive

[polyfractal]

Both are valid. GET requests can use a body; it isn't forbidden by the HTTP spec. See the callout titled "A GET Request with a Body?" at: https://www.elastic.co/guide/en/elasticsearch/guide/current/_empty_search.html for more details.

That said, if you need to change the HTTP method to better suit your environment, you can override it:

$client = Elasticsearch\ClientBuilder::create()->build();
$params = [
    'index' => 'test',
    'type' => 'test',
    'body' => [
        'query' => [
            'match_all' => []
        ]
    ],
    'client' => [
        'curl' => [
            CURLOPT_CUSTOMREQUEST => 'POST'
        ]
    ]
];
$client->search($params);

As an aside, running search requests through a caching proxy is probably a bad idea. They could easily cache a search request that is no longer valid because the underlying data has changed. Caching other API endpoints may make sense, but caching search rarely does. Elasticsearch has several built-in caches at various levels, so a top-level caching layer is not usually needed (and can lead to data inconsistency or confusing situations).

E.g. imagine the results are cached, user clicks on one of the hits for more information then the following GetDocument API fails because the document was actually deleted (but the search hits were cached).