esbulk

Fast parallel bulk loading utility for elasticsearch. Data is read from a newline delimited JSON file or stdin and indexed into elasticsearch in bulk and in parallel. The shortest command would be:

$ esbulk -index my-index-name < file.ldj

Caveat: If indexing pressure on the bulk API is too high (dozens or hundreds of parallel workers, large batch sizes, depending on you setup), esbulk will halt and report an error:

$ esbulk -index my-index-name -w 100 file.ldj
2017/01/02 16:25:25 error during bulk operation, try less workers (lower -w value) or
                    increase thread_pool.bulk.queue_size in your nodes

Please note that, in such a case, some documents are indexed and some are not. Your index will be in an inconsistent state, since there is no transactional bracket around the indexing process.

However, using defaults (parallism: number of cores) on a single node setup will just work. For larger clusters, increase the number of workers until you see full CPU utilization. After that, more workers won't buy any more speed.

Installation

$ go get github.com/miku/esbulk/cmd/esbulk

For deb or rpm packages, see: https://github.com/miku/esbulk/releases

Usage

$ esbulk -h
  Usage of esbulk:
  -0    set the number of replicas to 0 during indexing
  -cpuprofile string
          write cpu profile to file
  -host string
          elasticsearch host (deprecated: use -server instead) (default "localhost")
  -id string
          name of field to use as id field, by default ids are autogenerated
  -index string
          index name
  -mapping string
          mapping string or filename to apply before indexing
  -memprofile string
          write heap profile to file
  -port int
          elasticsearch port (deprecated: use -server instead) (default 9200)
  -purge
          purge any existing index before indexing
  -server string
          elasticsearch server, this works with https as well (default "http://localhost:9200")
  -size int
          bulk batch size (default 1000)
  -type string
          elasticsearch doc type (default "default")
  -u string
          http basic auth username:password, like curl -u
  -v    prints current program version
  -verbose
          output basic progress
  -w int
          number of workers to use (default 4)
  -z    unzip gz'd file on the fly

To index a JSON file, that contains one document per line, just run:

$ esbulk -index example file.ldj

Where file.ldj is line delimited JSON, like:

{"name": "esbulk", "version": "0.2.4"}
{"name": "estab", "version": "0.1.3"}
...

By default esbulk will use as many parallel workers, as there are cores. To tweak the indexing process, adjust the -size and -w parameters.

You can index from gzipped files as well, using the -z flag:

$ esbulk -z -index example file.ldj.gz

Starting with 0.3.7 the preferred method to set a non-default server hostport is via -server, e.g.

$ esbulk -server https://0.0.0.0:9201

This way, you can use https as well, which was not possible before. Options -host and -port are kept for backwards compatibility.

Reusing IDs

Since version 0.3.8: If you want to reuse IDs from your documents in elasticsearch, you can specify the ID field via -id flag:

$ cat file.json
{"x": "doc-1", "db": "mysql"}
{"x": "doc-2", "db": "mongo"}

Here, we would like to reuse the ID from field x.

$ esbulk -id x -index throwaway -verbose file.json
...

$ curl -s http://localhost:9200/throwaway/_search | jq
{
  "took": 2,
  "timed_out": false,
  "_shards": {
    "total": 5,
    "successful": 5,
    "failed": 0
  },
  "hits": {
    "total": 2,
    "max_score": 1,
    "hits": [
      {
        "_index": "throwaway",
        "_type": "default",
        "_id": "doc-2",
        "_score": 1,
        "_source": {
          "x": "doc-2",
          "db": "mongo"
        }
      },
      {
        "_index": "throwaway",
        "_type": "default",
        "_id": "doc-1",
        "_score": 1,
        "_source": {
          "x": "doc-1",
          "db": "mysql"
        }
      }
    ]
  }
}

Nested ID fields

Version 0.4.3 adds support for nested ID fields:

$ cat fixtures/pr-8-1.json
{"a": {"b": 1}}
{"a": {"b": 2}}
{"a": {"b": 3}}

$ esbulk -index throwaway -id a.b < fixtures/pr-8-1.json
...

Concatenated ID

Version 0.4.3 adds support for IDs that are the concatenation of multiple fields:

$ cat fixtures/pr-8-2.json
{"a": {"b": 1}, "c": "a"}
{"a": {"b": 2}, "c": "b"}
{"a": {"b": 3}, "c": "c"}

$ esbulk -index throwaway -id a.b,c < fixtures/pr-8-1.json
...

      {
        "_index": "xxx",
        "_type": "default",
        "_id": "1a",
        "_score": 1,
        "_source": {
          "a": {
            "b": 1
          },
          "c": "a"
        }
      },

Using X-Pack

Since 0.4.2: support for secured elasticsearch nodes:

$ esbulk -u elastic:changeme -index myindex file.ldj

---

A similar project has been started for solr, called solrbulk.

Contributors

• Klaubert Herr
• sakshambathla
• KUOKA Yusuke

Measurements

$ csvlook -I measurements.csv
| es    | esbulk | docs      | avg_b | nodes | total_cores | total_heap_gb | t_s   | docs_per_s | repl |
|-------|--------|-----------|-------|-------|-------------|---------------|-------|------------|------|
| 6.1.2 | 0.4.8  | 138000000 | 2000  | 1     | 32          |  64           |  6420 |  22100     | 1    |
| 6.1.2 | 0.4.8  | 138000000 | 2000  | 1     |  8          |  30           | 27360 |   5100     | 1    |
| 6.1.2 | 0.4.8  |   1000000 | 2000  | 1     |  4          |   1           |   300 |   3300     | 1    |
| 6.1.2 | 0.4.8  |  10000000 |   26  | 1     |  4          |   8           |   122 |  81000     | 1    |
| 6.1.2 | 0.4.8  |  10000000 |   26  | 1     | 32          |  64           |    32 | 307000     | 1    |
| 6.2.3 | 0.4.10 | 142944530 | 2000  | 2     | 64          | 128           | 26253 |   5444     | 1    |
| 6.2.3 | 0.4.10 | 142944530 | 2000  | 2     | 64          | 128           | 11113 |  12831     | 0    |
| 6.2.3 | 0.4.13 |  15000000 | 6000  | 2     | 64          | 128           |  2460 |   6400     | 0    |

Why not add a row?