esqa 0.4.2

Tiny tool to checks the search qualities of the Elasticsearch indices.

Table of Contents

• Overview
• Install
• Behavior
• Functions

Overview

Esqa automates the checks the qualities of the Elasticsearch indices as the unit test frameworks such as RSpec or PyTests. Users add the test cases into the setting files and checks if the target indices is build as expected running the command esqa.

Install

$ pip install esqa

Behavior

When we run Esqa, the following steps are executed.

1. Submit Es query to an Elasticsearch cluster
2. Get the result ranking from Elasticsearch
3. Check if the rankings from Es cluster satisfy the conditions described in configuration file

The following is the image.

Functions

Specifically esqa provides two functions, assertion and compute distance between rankings from two index and query settings.

With assertion function, we can check if the results ranking satisfy the expectation for the specified queries. With distance function, we can see the queries which is much different from previous settings (index and query`).

The successive sections, we see the assertion and distance functions.

Assertion function

Esqa provides the esqa command which check if the queries gets the expected search rankings from Elasticsearch indices.

We run the esqa command specifying the configuration file and target index.

$ esqa assertion --config sample_config.json --index document-index

Configurations

Esqa has the settings file in which we add the test cases. The following is an example of the setting file of esqa. The setting file means that results from Elasticsearch clusters must satisfy the conditions defined in asserts block when we run the defined query (searching engineer to the message field) to the target index.

{
  "cases": [
    {
      "name": "match query",
      "request": {
        "query": {
          "match": {
            "message": {
              "query": "engineer"
            }
          }
        }
      },
      "asserts": [
        {
          "type": "equal",
          "rank": 0,
          "item": {
            "field": "document_id",
            "value": "24343"
          }
        }
      ]
    }
  ]
}

We add all the test cases into cases block. Each test cases have three elements name, request and asserts. name is the name of the test case. request is the target Es query which we want to validate. We add a set of expected behaviors to the asserts block.

The asserts block contains the conditions that search results from Elasticsearch cluster must satisfy. Each condition contains several elements type, rank and item.

Element	Summary
type	condition types (equal、higher、lower）
rank	rank of the specified item
item	item stored in Elasticsearch indices specified in rank element must satisfy

item element specifies the document in Es indices. The item is specified with the field value.

Element	Summary
field	field name
value	value of the field specified in field element

Templates

Sometimes queries in the test cases are almost the same. In such cases, esqa provides templates in the configuration files.

Template files are JSON file which contains an Elasticsearch query with variables.

The following is an example of template file. As we can see, query block contains a variable ${query_str}. The variables are injected from the Esqa configuration file.

{
  "query": {
    "match": {
      "message": {
        "query": "${query_str}"
      }
    }
  }
}

The following is a configuration file which specifies the template file. To uses template files in the configuration file, we add template element in query block. The variables in the specified template file need to be added in the query block. For example the configuration file added a variable query_str defined in template file.

{
  "templates": [
    {
      "name": "basic_query",
      "path": "tests/fixtures/default_template.json"
    }
  ],
  "cases": [
    {
      "name": "match identical",
      "request": {
        "template": "basic_query",
        "query_str": "engineer"
      },
      "asserts": [
        {
          "type": "equal",
          "rank": 0,
          "item": {
            "field": "id",
            "value": "2324"
          }
        }
      ]
    }
  ]
}

Distance function

When we tune the Es indices, we somtimes want to compare the rankings from the previous indices. Esqa computes the comparison between the rankings in the current settings and previous ones.

Before we run the command we prepare the configuration for the esqa distance function. The format is the almost the same as validation settings except that the settings for distance function does not have assert blocks.

{
  "templates": [{
    "name": "basic_query",
    "path": "sample/template.json"
  }],
  "cases": [
    {"request": {"template": "basic_query", "query_str":  "Windows PC"}, "name": "Windows PC"},
    {"request": {"template": "basic_query", "query_str": "Tablet"}, "name": "Tablet"}
  ]
}

Before changing the Es settings, we run the save command to preserve the current ranking.

esqa save --config sample/ranking.json --index sample > output/ranking_before_change.json

Then we change the Es index or query settings and run distance command specifying the ranking file.

esqa distance --config sample/compared_ranking.json --index sample --ranking output/ranking.json
[
  {
    "name": "Windows PC",
    "similarity": 0.5,
    "ranking_pair": [
      [
        "4",
        "6"
      ],
      [
        "5",
        "4"
      ],
      [
        "6",
        "5"
      ]
    ]
  },
  {
    "name": "Tablet",
    "similarity": 0.5416666666666666,
    "ranking_pair": [
      [
        "22",
        "21"
      ],
      [
        "23",
        "22"
      ],
      [
        "3",
        "23"
      ],
      [
        "21",
        "3"
      ]
    ]
  }
]

Or, we can compare between two preserved rankings by distance-rankings command.

esqa distance-rankings --ranking1 output/ranking1.json --ranking2 output/ranking2.json
[
  {
    "name": "Windows PC",
    "similarity": 0.5,
    "ranking_pair": [
      [
        "4",
        "6"
      ],
      [
        "5",
        "4"
      ],
      [
        "6",
        "5"
      ]
    ]
  },
  {
    "name": "Tablet",
    "similarity": 0.5416666666666666,
    "ranking_pair": [
      [
        "22",
        "21"
      ],
      [
        "23",
        "22"
      ],
      [
        "3",
        "23"
      ],
      [
        "21",
        "3"
      ]
    ]
  }
]

Finally, we get the query cases that have been changed significantly.