There's a lot of open issues
Opensearch output plugin for Fluent event collector


~> 2.0
~> 5.8
>= 0
~> 3.3.0
~> 1.7.0


 Project Readme

Fluent::Plugin::OpenSearch, a plugin for Fluentd

Gem Version Testing on Windows Testing on macOS Testing on Ubuntu Coverage Status

Send your logs to OpenSearch (and search them with OpenSearch Dashboards maybe?)

  • Installation
  • Usage
    • Index templates
  • Configuration
    • host
    • port
    • emit_error_for_missing_id
    • hosts
    • user, password, path, scheme, ssl_verify
    • logstash_format
    • logstash_prefix
    • logstash_prefix_separator
    • logstash_dateformat
    • pipeline
    • time_key_format
    • time_precision
    • time_key
    • time_key_exclude_timestamp
    • include_timestamp
    • utc_index
    • suppress_type_name
    • target_index_key
    • target_type_key
    • target_index_affinity
    • template_name
    • template_file
    • template_overwrite
    • customize_template
    • index_date_pattern
    • application_name
    • index_prefix
    • templates
    • max_retry_putting_template
    • fail_on_putting_template_retry_exceed
    • fail_on_detecting_os_version_retry_exceed
    • max_retry_get_os_version
    • request_timeout
    • reload_connections
    • reload_on_failure
    • resurrect_after
    • include_tag_key, tag_key
    • id_key
    • parent_key
    • routing_key
    • remove_keys
    • remove_keys_on_update
    • remove_keys_on_update_key
    • retry_tag
    • write_operation
    • time_parse_error_tag
    • reconnect_on_error
    • with_transporter_log
    • content_type
    • include_index_in_url
    • http_backend
    • http_backend_excon_nonblock
    • prefer_oj_serializer
    • compression_level
    • Client/host certificate options
    • Proxy Support
    • Buffer options
    • Hash flattening
    • Generate Hash ID
    • sniffer_class_name
    • selector_class_name
    • reload_after
    • validate_client_version
    • unrecoverable_error_types
    • unrecoverable_record_types
    • emit_error_label_event
    • verify os version at startup
    • default_opensearch_version
    • custom_headers
    • api_key
    • Not seeing a config you need?
    • Dynamic configuration
    • Placeholders
    • Multi workers
    • log_os_400_reason
    • suppress_doc_wrap
    • ignore_exceptions
    • exception_backup
    • bulk_message_request_threshold
    • truncate_caches_interval
    • use_legacy_template
    • metadata section
      • include_chunk_id
      • chunk_id_key
  • Configuration - OpenSearch Input
  • Configuration - OpenSearch Filter GenID
  • Configuration - OpenSearch Output Data Stream
  • Configuration - AWS OpenSearch Service
  • Troubleshooting
  • Contact
  • Contributing
  • Running tests


fluent-plugin-opensearch fluentd ruby
>= 1.0.0 >= v1.x >= 2.4

NOTE: Since fluent-plugin-opensearch 1.1.0, it requires faraday 2.0 or later.

NOTE: This documentation is for fluent-plugin-opensearch 1.x or later.


$ gem install fluent-plugin-opensearch


In your Fluentd configuration, use @type opensearch. Additional configuration is optional, default values would look like this:

<match my.logs>
  @type opensearch
  host localhost
  port 9200
  index_name fluentd

NOTE: type_name parameter is fixed value and cannot change and configure from _doc value for OpenSearch 1.

Index templates

This plugin creates OpenSearch indices by merely writing to them. Consider using Index Templates to gain control of what get indexed and how.



host user-custom-host.domain # default localhost

You can specify OpenSearch host by this parameter.

To use IPv6 address on host parameter, you can use the following styles:

string style

To use string style, you must quote IPv6 address due to prevent to be interpreted as JSON:

host "[2404:7a80:d440:3000:192a:a292:bd7f:ca10]"

raw style

You can also specify raw IPv6 address. This will be handled as [specified IPv6 address]:

host 2404:7a80:d440:3000:192a:a292:bd7f:ca10


port 9201 # defaults to 9200

You can specify OpenSearch port by this parameter.


emit_error_for_missing_id true

When write_operation is configured to anything other then index, setting this value to true will cause the plugin to emit_error_event of any records which do not include an _id field. The default (false) behavior is to silently drop the records.


hosts host1:port1,host2:port2,host3:port3

You can specify multiple OpenSearch hosts with separator ",".

If you specify multiple hosts, this plugin will load balance updates to OpenSearch. This is an opensearch-ruby feature, the default strategy is round-robin.

If you specify hosts option, host and port options are ignored.

host user-custom-host.domain # ignored
port 9200                    # ignored
hosts host1:port1,host2:port2,host3:port3

If you specify hosts option without port, port option is used.

port 9200
hosts host1:port1,host2:port2,host3 # port3 is 9200

Note: If you will use scheme https, do not include "https://" in your hosts ie. host "https://domain", this will cause ES cluster to be unreachable and you will receive an error "Can not reach OpenSearch cluster"

Note: Embedded the username/password in the URL syntax is not recommended to use because it was found to cause serious connection problems. Please do not use its style on your settings and use the user and password field (described below) instead.

IPv6 addresses

When you want to specify IPv6 addresses, you must specify schema together:

hosts http://[2404:7a80:d440:3000:de:7311:6329:2e6c]:port1,http://[2404:7a80:d440:3000:de:7311:6329:1e6c]:port2,http://[2404:7a80:d440:3000:de:6311:6329:2e6c]:port3

If you don't specify hosts with schema together, OpenSearch plugin complains Invalid URI for them.

user, password, path, scheme, ssl_verify

user demo
password secret
path /elastic_search/
scheme https

You can specify user and password for HTTP Basic authentication.

And this plugin will escape required URL encoded characters within %{} placeholders.

user %{demo+}
password %{@secret}

Specify ssl_verify false to skip ssl verification (defaults to true)


logstash_format true # defaults to false

This is meant to make writing data into OpenSearch indices compatible to what Logstash calls them. By doing this, one could take advantage of Kibana. See logstash_prefix and logstash_dateformat to customize this index name pattern. The index name will be #{logstash_prefix}-#{formatted_date}

⚠️ Setting this option to true will ignore the index_name setting. The default index name prefix is logstash-.


include_timestamp true # defaults to false

Adds a @timestamp field to the log, following all settings logstash_format does, except without the restrictions on index_name. This allows one to log to an alias in OpenSearch and utilize the rollover API.


logstash_prefix mylogs # defaults to "logstash"


logstash_prefix_separator _ # defaults to "-"


The strftime format to generate index target index name when logstash_format is set to true. By default, the records are inserted into index logstash-YYYY.MM.DD. This option, alongwith logstash_prefix lets us insert into specified index like mylogs-YYYYMM for a monthly index.

logstash_dateformat %Y.%m. # defaults to "%Y.%m.%d"


This param is to set a pipeline id of your opensearch to be added into the request, you can configure ingest node.

pipeline pipeline_id


The format of the time stamp field (@timestamp or what you specify with time_key). This parameter only has an effect when logstash_format is true as it only affects the name of the index we write to. Please see Time#strftime for information about the value of this format.

Setting this to a known format can vastly improve your log ingestion speed if all most of your logs are in the same format. If there is an error parsing this format the timestamp will default to the ingestion time. You can get a further performance improvement by installing the "strptime" gem: fluent-gem install strptime.

For example to parse ISO8601 times with sub-second precision:

time_key_format %Y-%m-%dT%H:%M:%S.%N%z


Should the record not include a time_key, define the degree of sub-second time precision to preserve from the time portion of the routed event.

For example, should your input plugin not include a time_key in the record but it able to pass a time to the router when emitting the event (AWS CloudWatch events are an example of this), then this setting will allow you to preserve the sub-second time resolution of those events. This is the case for: fluent-plugin-cloudwatch-ingest.


By default, when inserting records in Logstash format, @timestamp is dynamically created with the time at log ingestion. If you'd like to use a custom time, include an @timestamp with your record.

{"@timestamp": "2014-04-07T000:00:00-00:00"}

You can specify an option time_key (like the option described in tail Input Plugin) to replace @timestamp key.

Suppose you have settings

logstash_format true
time_key vtm

Your input is:

  "title": "developer",
  "vtm": "2014-12-19T08:01:03Z"

The output will be

  "title": "developer",
  "@timestamp": "2014-12-19T08:01:03Z",
  "vtm": "2014-12-19T08:01:03Z"

See time_key_exclude_timestamp to avoid adding @timestamp.


time_key_exclude_timestamp false

By default, setting time_key will copy the value to an additional field @timestamp. When setting time_key_exclude_timestamp true, no additional field will be added.


utc_index true

By default, the records inserted into index logstash-YYMMDD with UTC (Coordinated Universal Time). This option allows to use local time if you describe utc_index to false.


If OpenSearch cluster complains types removal warnings, this can be suppressed with:

suppress_type_name true


Tell this plugin to find the index name to write to in the record under this key in preference to other mechanisms. Key can be specified as path to nested record using dot ('.') as a separator.

If it is present in the record (and the value is non falsy) the value will be used as the index name to write to and then removed from the record before output; if it is not found then it will use logstash_format or index_name settings as configured.

Suppose you have the following settings

target_index_key @target_index
index_name fallback

If your input is:

  "title": "developer",
  "@timestamp": "2014-12-19T08:01:03Z",
  "@target_index": "logstash-2014.12.19"

The output would be

  "title": "developer",
  "@timestamp": "2014-12-19T08:01:03Z",

and this record will be written to the specified index (logstash-2014.12.19) rather than fallback.


Similar to target_index_key config, find the type name to write to in the record under this key (or nested record). If key not found in record - fallback to type_name (default "fluentd").


Enable plugin to dynamically select logstash time based target index in update/upsert operations based on already indexed data rather than current time of indexing.

target_index_affinity true # defaults to false

By default plugin writes data of logstash format index based on current time. For example daily based index after mignight data is written to newly created index. This is normally ok when data is coming from single source and not updated after indexing.

But if you have a use case where data is also updated after indexing and id_key is used to identify the document uniquely for updating. Logstash format is wanted to be used for easy data managing and retention. Updates are done right after indexing to complete the data (all data not available from single source) and no updates are done anymore later point on time. In this case problem happends at index rotation time where write to 2 indexes with same id_key value may happen.

This setting will search existing data by using elastic search's id query using id_key value (with logstash_prefix and logstash_prefix_separator index pattarn e.g. logstash-*). The index of found data is used for update/upsert. When no data is found, data is written to current logstash index as normally.

This setting requires following other settings:

logstash_format true
id_key myId  # Some field on your data to identify the data uniquely
write_operation upsert  # upsert or update

Suppose you have the following situation where you have 2 different match to consume data from 2 different Kafka topics independently but close in time with each other (order not known).

  <match data1>
    @type opensearch
    id_key myId
    write_operation upsert
    logstash_format true
    logstash_dateformat %Y.%m.%d
    logstash_prefix myindexprefix
    target_index_affinity true

  <match data2>
    @type opensearch
    id_key myId
    write_operation upsert
    logstash_format true
    logstash_dateformat %Y.%m.%d
    logstash_prefix myindexprefix
    target_index_affinity true

If your first (data1) input is:

  "myId": "myuniqueId1",
  "datafield1": "some value",

and your second (data2) input is:

  "myId": "myuniqueId1",
  "datafield99": "some important data from other source tightly related to id myuniqueId1 and wanted to be in same document.",

Date today is 10.05.2021 so data is written to index myindexprefix-2021.05.10 when both data1 and data2 is consumed during today. But when we are close to index rotation and data1 is consumed and indexed at 2021-05-10T23:59:55.59707672Z and data2 is consumed a bit later at 2021-05-11T00:00:58.222079Z i.e. logstash index has been rotated and normally data2 would have been written to index myindexprefix-2021.05.11. But with target_index_affinity setting as value true, data2 is now written to index myindexprefix-2021.05.10 into same document with data1 as wanted and duplicated document is avoided.


The name of the template to define. If a template by the name given is already present, it will be left unchanged, unless template_overwrite is set, in which case the template will be updated.

This parameter along with template_file allow the plugin to behave similarly to Logstash (it installs a template at creation time) so that raw records are available. See uken/fluent-plugin-elasticsearch#33.

template_file must also be specified.


The path to the file containing the template to install.

template_name must also be specified.


Specify index templates in form of hash. Can contain multiple templates.

templates { "template_name_1": "path_to_template_1_file", "template_name_2": "path_to_template_2_file"}

Note: Before ES plugin v4.1.2, if template_file and template_name are set, then this parameter will be ignored. In 4.1.3 or later, template_file and template_name can work with templates.


Specify the string and its value to be replaced in form of hash. Can contain multiple key value pair that would be replaced in the specified template_file.

customize_template {"string_1": "subs_value_1", "string_2": "subs_value_2"}

If template_file and template_name are set, then this parameter will be in effect otherwise ignored.


Specify this to override the index date pattern for creating a rollover index. The default is to use "now/d", for example: <logstash-default-{now/d}-000001>. Overriding this changes the rollover time period. Setting "now/w{xxxx.ww}" would create weekly rollover indexes instead of daily.

index_date_pattern "now/w{xxxx.ww}" # defaults to "now/d"

If empty string("") is specified in index_date_pattern, index date pattern is not used. OpenSearch plugin just creates <target_index-application_name-000001> rollover index instead of <target_index-application_name-{index_date_pattern}-000001>.

If customize_template is set, then this parameter will be in effect otherwise ignored.


This parameter is marked as obsoleted.


Specify the application name for the rollover index to be created.

application_name default # defaults to "default"


Always update the template, even if it already exists.

template_overwrite true # defaults to false

One of template_file or templates must also be specified if this is set.


You can specify times of retry putting template.

This is useful when OpenSearch plugin cannot connect OpenSearch to put template. Usually, booting up clustered OpenSearch containers are much slower than launching Fluentd container.

max_retry_putting_template 15 # defaults to 10


Indicates whether to fail when max_retry_putting_template is exceeded. If you have multiple output plugin, you could use this property to do not fail on fluentd statup.

fail_on_putting_template_retry_exceed false # defaults to true


Indicates whether to fail when max_retry_get_os_version is exceeded. If you want to use fallback mechanism for obtaining OpenSearch version, you could use this property to do not fail on fluentd statup.

fail_on_detecting_os_version_retry_exceed false

And the following parameters should be working with:

verify_os_version_at_startup true
max_retry_get_os_version 2 # greater than 0.
default_opensearch_version 1 # This version is used when occurring fallback.


You can specify times of retry obtaining OpenSearch version.

This is useful when OpenSearch plugin cannot connect OpenSearch to obtain OpenSearch version. Usually, booting up clustered OpenSearch containers are much slower than launching Fluentd container.

max_retry_get_os_version 17 # defaults to 15


You can specify HTTP request timeout.

This is useful when OpenSearch cannot return response for bulk request within the default of 5 seconds.

request_timeout 15s # defaults to 5s


You can tune how the opensearch-transport host reloading feature works. By default it will reload the host list from the server every 10,000th request to spread the load. This can be an issue if your OpenSearch cluster is behind a Reverse Proxy, as Fluentd process may not have direct network access to the OpenSearch nodes.

reload_connections false # defaults to true


Indicates that the opensearch-transport will try to reload the nodes addresses if there is a failure while making the request, this can be useful to quickly remove a dead node from the list of addresses.

reload_on_failure true # defaults to false


You can set in the opensearch-transport how often dead connections from the opensearch-transport's pool will be resurrected.

resurrect_after 5s # defaults to 60s

include_tag_key, tag_key

include_tag_key true # defaults to false
tag_key tag # defaults to tag

This will add the Fluentd tag in the JSON record. For instance, if you have a config like this:

<match my.logs>
  @type opensearch
  include_tag_key true
  tag_key _key

The record inserted into OpenSearch would be

{"_key": "my.logs", "name": "Johnny Doeie"}


id_key request_id # use "request_id" field as a record id in ES

By default, all records inserted into OpenSearch get a random _id. This option allows to use a field in the record as an identifier.

This following record {"name": "Johnny", "request_id": "87d89af7daffad6"} will trigger the following OpenSearch command

{ "index" : { "_index": "logstash-2013.01.01", "_type": "fluentd", "_id": "87d89af7daffad6" } }
{ "name": "Johnny", "request_id": "87d89af7daffad6" }

Fluentd re-emits events that failed to be indexed/ingested in OpenSearch with a new and unique _id value, this means that congested OpenSearch clusters that reject events (due to command queue overflow, for example) will cause Fluentd to re-emit the event with a new _id, however OpenSearch may actually process both (or more) attempts (with some delay) and create duplicate events in the index (since each have a unique _id value), one possible workaround is to use the fluent-plugin-genhashvalue plugin to generate a unique _hash key in the record of each event, this _hash record can be used as the id_key to prevent OpenSearch from creating duplicate events.

id_key _hash

Example configuration for fluent-plugin-genhashvalue (review the documentation of the plugin for more details)

<filter logs.**>
  @type genhashvalue
  keys session_id,request_id
  hash_type md5    # md5/sha1/sha256/sha512
  base64_enc true
  base91_enc false
  set_key _hash
  separator _
  inc_time_as_key true
  inc_tag_as_key true

⚠️ In order to avoid hash-collisions and loosing data careful consideration is required when choosing the keys in the event record that should be used to calculate the hash

Using nested key

Nested key specifying syntax is also supported.

With the following configuration

id_key $.nested.request_id

and the following nested record

{"nested":{"name": "Johnny", "request_id": "87d89af7daffad6"}}

will trigger the following OpenSearch command


⚠️ Note that Hash flattening may be conflict nested record feature.


parent_key a_parent # use "a_parent" field value to set _parent in opensearch command

If your input is

{ "name": "Johnny", "a_parent": "my_parent" }

OpenSearch command would be

{ "index" : { "_index": "****", "_type": "****", "_id": "****", "_parent": "my_parent" } }
{ "name": "Johnny", "a_parent": "my_parent" }

if parent_key is not configed or the parent_key is absent in input record, nothing will happen.

Using nested key

Nested key specifying syntax is also supported.

With the following configuration

parent_key $.nested.a_parent

and the following nested record

{"nested":{ "name": "Johnny", "a_parent": "my_parent" }}

will trigger the following OpenSearch command


⚠️ Note that Hash flattening may be conflict nested record feature.


Similar to parent_key config, will add _routing into opensearch command if routing_key is set and the field does exist in input event.


parent_key a_parent
routing_key a_routing
remove_keys a_parent, a_routing # a_parent and a_routing fields won't be sent to opensearch


Remove keys on update will not update the configured keys in opensearch when a record is being updated. This setting only has any effect if the write operation is update or upsert.

If the write setting is upsert then these keys are only removed if the record is being updated, if the record does not exist (by id) then all of the keys are indexed.

remove_keys_on_update foo,bar


This setting allows remove_keys_on_update to be configured with a key in each record, in much the same way as target_index_key works. The configured key is removed before indexing in opensearch. If both remove_keys_on_update and remove_keys_on_update_key is present in the record then the keys in record are used, if the remove_keys_on_update_key is not present then the value of remove_keys_on_update is used as a fallback.

remove_keys_on_update_key keys_to_skip


This setting allows custom routing of messages in response to bulk request failures. The default behavior is to emit failed records using the same tag that was provided. When set to a value other then nil, failed messages are emitted with the specified tag:

retry_tag 'retry_es'

NOTE: retry_tag is optional. If you would rather use labels to reroute retries, add a label (e.g '@label @SOMELABEL') to your fluent opensearch plugin configuration. Retry records are, by default, submitted for retry to the ROOT label, which means records will flow through your fluentd pipeline from the beginning. This may nor may not be a problem if the pipeline is idempotent - that is - you can process a record again with no changes. Use tagging or labeling to ensure your retry records are not processed again by your fluentd processing pipeline.


The write_operation can be any of:

Operation Description
index (default) new data is added while existing data (based on its id) is replaced (reindexed).
create adds new data - if the data already exists (based on its id), the op is skipped.
update updates existing data (based on its id). If no data is found, the op is skipped.
upsert known as merge or insert if the data does not exist, updates if the data exists (based on its id).

Please note, id is required in create, update, and upsert scenario. Without id, the message will be dropped.


With logstash_format true, opensearch plugin parses timestamp field for generating index name. If the record has invalid timestamp value, this plugin emits an error event to @ERROR label with time_parse_error_tag configured tag.

Default value is opensearch_plugin.output.time.error. Note that this default values is quite different from Elasticsearch plugin.


Indicates that the plugin should reset connection on any error (reconnect on next send). By default it will reconnect only on "host unreachable exceptions". We recommended to set this true in the presence of opensearch shield.

reconnect_on_error true # defaults to false


This is debugging purpose option to enable to obtain transporter layer log. Default value is false for backward compatibility.

We recommend to set this true if you start to debug this plugin.

with_transporter_log true


With content_type application/x-ndjson, opensearch plugin adds application/x-ndjson as Content-Type in payload.

Default value is application/json which is default Content-Type of OpenSearch requests. If you will not use template, it recommends to set content_type application/x-ndjson.

content_type application/x-ndjson


With this option set to true, Fluentd manifests the index name in the request URL (rather than in the request body). You can use this option to enforce an URL-based access control.

include_index_in_url true


With http_backend typhoeus, opensearch plugin uses typhoeus faraday http backend. Typhoeus can handle HTTP keepalive.

Default value is excon which is default http_backend of opensearch plugin.

http_backend typhoeus


With http_backend_excon_nonblock false, opensearch plugin use excon with nonblock=false. If you use opensearch plugin with jRuby for https, you may need to consider to set false to avoid follwoing problems.

But for all other case, it strongly reccomend to set true to avoid process hangin problem reported in uken/fluent-plugin-elasticsearch#732

Default value is true.

http_backend_excon_nonblock false


You can add gzip compression of output data. In this case default_compression, best_compression or best speed option should be chosen. By default there is no compression, default value for this option is no_compression

compression_level best_compression


With default behavior, OpenSearch client uses Yajl as JSON encoder/decoder. Oj is the alternative high performance JSON encoder/decoder. When this parameter sets as true, OpenSearch client uses Oj as JSON encoder/decoder.

Default value is false.

prefer_oj_serializer true

Client/host certificate options

Need to verify OpenSearch's certificate? You can use the following parameter to specify a CA instead of using an environment variable.

ca_file /path/to/your/ca/cert

Does your OpenSearch cluster want to verify client connections? You can specify the following parameters to use your client certificate, key, and key password for your connection.

client_cert /path/to/your/client/cert
client_key /path/to/your/private/key
client_key_pass password

If you want to configure SSL/TLS version, you can specify ssl_version parameter.

ssl_version TLSv1_2 # or [SSLv23, TLSv1, TLSv1_1]

⚠️ If SSL/TLS enabled, it might have to be required to set ssl_version.

In OpenSearch plugin v4.0.2 with Ruby 2.5 or later combination, OpenSearch plugin also support ssl_max_version and ssl_min_version.

ssl_max_version TLSv1_3
ssl_min_version TLSv1_2

OpenSearch plugin will use TLSv1.2 as minimum ssl version and TLSv1.3 as maximum ssl version on transportation with TLS. Note that when they are used in Elastissearch plugin configuration, ssl_version is not used to set up TLS version.

If they are not specified in the OpenSearch plugin configuration, ssl_max_version and ssl_min_version is set up with:

In OpenSearch plugin v1.0.0 or later with Ruby 2.5 or later environment, ssl_max_version should be TLSv1_3 and ssl_min_version should be TLSv1_2.

Proxy Support

Starting with version 0.8.0, this gem uses excon, which supports proxy with environment variables -

Buffer options

fluentd-plugin-opensearch extends Fluentd's builtin Output plugin and use compat_parameters plugin helper. It adds the following options:

buffer_type memory
flush_interval 60s
retry_limit 17
retry_wait 1.0
num_threads 1

The value for option buffer_chunk_limit should not exceed value http.max_content_length in your OpenSearch setup (by default it is 100mb).

Note: If you use or evaluate Fluentd v0.14, you can use <buffer> directive to specify buffer configuration, too. In more detail, please refer to the buffer configuration options for v0.14

Note: If you use disable_retry_limit in v0.12 or retry_forever in v0.14 or later, please be careful to consume memory inexhaustibly.

Hash flattening

OpenSearch will complain if you send object and concrete values to the same field. For example, you might have logs that look this, from different places:

{"people" => 100} {"people" => {"some" => "thing"}}

The second log line will be rejected by the OpenSearch parser because objects and concrete values can't live in the same field. To combat this, you can enable hash flattening.

flatten_hashes true
flatten_hashes_separator _

This will produce opensearch output that looks like this: {"people_some" => "thing"}

Note that the flattener does not deal with arrays at this time.

Generate Hash ID

By default, the fluentd opensearch plugin does not emit records with a _id field, leaving it to OpenSearch to generate a unique _id as the record is indexed. When an OpenSearch cluster is congested and begins to take longer to respond than the configured request_timeout, the fluentd opensearch plugin will re-send the same bulk request. Since OpenSearch can't tell its actually the same request, all documents in the request are indexed again resulting in duplicate data. In certain scenarios, this can result in essentially and infinite loop generating multiple copies of the same data.

The bundled opensearch_genid filter can generate a unique _hash key for each record, this key may be passed to the id_key parameter in the opensearch plugin to communicate to OpenSearch the uniqueness of the requests so that duplicates will be rejected or simply replace the existing records. Here is a sample config:

<filter **>
  @type opensearch_genid
  hash_id_key _hash    # storing generated hash id key (default is _hash)
<match **>
  @type opensearch
  id_key _hash # specify same key name which is specified in hash_id_key
  remove_keys _hash # OpenSearch doesn't like keys that start with _
  # other settings are omitted.

Sniffer Class Name

The default Sniffer used by the OpenSearch::Transport class works well when Fluentd has a direct connection to all of the OpenSearch servers and can make effective use of the _nodes API. This doesn't work well when Fluentd must connect through a load balancer or proxy. The parameter sniffer_class_name gives you the ability to provide your own Sniffer class to implement whatever connection reload logic you require. In addition, there is a new Fluent::Plugin::OpenSearchSimpleSniffer class which reuses the hosts given in the configuration, which is typically the hostname of the load balancer or proxy. For example, a configuration like this would cause connections to logging-os to reload every 100 operations:

host logging-os
port 9200
reload_connections true
sniffer_class_name Fluent::Plugin::OpenSearchSimpleSniffer
reload_after 100


The included sniffer class is not required out_opensearch. You should tell Fluentd where the sniffer class exists.

If you use td-agent, you must put the following lines into TD_AGENT_DEFAULT file:

sniffer=$(td-agent-gem contents fluent-plugin-opensearch|grep opensearch_simple_sniffer.rb)
TD_AGENT_OPTIONS="--use-v1-config -r $sniffer"

If you use Fluentd directly, you must pass the following lines as Fluentd command line option:

sniffer=$(td-agent-gem contents fluent-plugin-opensearch|grep opensearch_simple_sniffer.rb)
$ fluentd -r $sniffer [AND YOUR OTHER OPTIONS]

Selector Class Name

The default selector used by the OpenSearch::Transport class works well when Fluentd should behave round robin and random selector cases. This doesn't work well when Fluentd should behave fallbacking from exhausted ES cluster to normal ES cluster. The parameter selector_class_name gives you the ability to provide your own Selector class to implement whatever selection nodes logic you require.

The below configuration is using plugin built-in OpenSearchFallbackSelector:

hosts exhausted-host:9201,normal-host:9200
selector_class_name "Fluent::Plugin::OpenSeartchFallbackSelector"


The included selector class is required in out_opensearch by default. But, your custom selector class is not required in out_opensearch. You should tell Fluentd where the selector class exists.

If you use td-agent, you must put the following lines into TD_AGENT_DEFAULT file:

TD_AGENT_OPTIONS="--use-v1-config -r $selector"

If you use Fluentd directly, you must pass the following lines as Fluentd command line option:

$ fluentd -r $selector [AND YOUR OTHER OPTIONS]

Reload After

When reload_connections true, this is the integer number of operations after which the plugin will reload the connections. The default value is 10000.

Validate Client Version

When you use mismatched OpenSearch server and client libraries, fluent-plugin-opensearch cannot send data into OpenSearch. The default value is false.

validate_client_version true

Unrecoverable Error Types

Default unrecoverable_error_types parameter is set up strictly. Because rejected_execution_exception is caused by exceeding OpenSearch's thread pool capacity. Advanced users can increase its capacity, but normal users should follow default behavior.

If you want to increase it and forcibly retrying bulk request, please consider to change unrecoverable_error_types parameter from default value.

Change default value of thread_pool.bulk.queue_size in opensearch.yml: e.g.)

thread_pool.bulk.queue_size: 1000

Then, remove rejected_execution_exception from unrecoverable_error_types parameter:

unrecoverable_error_types ["out_of_memory_error"]

Unrecoverable Record Types

Default unrecoverable_record_types parameter is set up loosely. Because json_parse_exception is caused by invalid JSON record. Another possible unrecoverable record error should be included within this paramater.

If you want to handle security_exception as unrecoverable exceptions, please consider to change unrecoverable_record_types parameter from default value:

unrecoverable_record_types ["json_parse_exception", "security_exception"]

If this error type is included in OpenSearch response, an invalid record should be sent in @ERROR data pipeline.


Default emit_error_label_event value is true.

Emitting error label events is default behavior.

When using the followin configuration, OpenSearch plugin will cut error events on error handler:

emit_error_label_event false


Because OpenSearch plugin will ought to change behavior each of OpenSearch major versions.

For example, OpenSearch 1 requests to handle only _doc type_name in index.

If you want to disable to verify OpenSearch version at start up, set it as false.

When using the following configuration, OpenSearch plugin intends to communicate into OpenSearch 1.

verify_os_version_at_startup false
default_opensearch_version 1

The default value is true.


This parameter changes that OpenSearch plugin assumes the default OpenSearch version. The default value is 1.


This parameter adds additional headers to request. The default value is {}.

custom_headers {"token":"secret"}

Not seeing a config you need?

We try to keep the scope of this plugin small and not add too many configuration options. If you think an option would be useful to others, feel free to open an issue or contribute a Pull Request.

Alternatively, consider using fluent-plugin-forest. For example, to configure multiple tags to be sent to different OpenSearch indices:

<match my.logs.*>
  @type forest
  subtype opensearch
  remove_prefix my.logs
    logstash_prefix ${tag}
    # ...

And yet another option is described in Dynamic Configuration section.

Note: If you use or evaluate Fluentd v0.14, you can use builtin placeholders. In more detail, please refer to Placeholders section.


v0.14 placeholders can handle ${tag} for tag, %Y%m%d like strftime format, and custom record keys like as record["mykey"].

Note that custom chunk key is different notations for record_reformer and record_modifier. They uses record["some_key"] to specify placeholders, but this feature uses ${key1}, ${key2} notation. And tag, time, and some arbitrary keys must be included in buffer directive attributes.

They are used as below:


<match my.logs>
  @type opensearch
  index_name elastic.${tag} #=> replaced with each event's tag. e.g.) elastic.test.tag
  <buffer tag>
    @type memory
  # <snip>


<match my.logs>
  @type opensearch
  index_name elastic.%Y%m%d #=> e.g.) elastic.20170811
  <buffer tag, time>
    @type memory
    timekey 3600
  # <snip>

custom key

records = {key1: "value1", key2: "value2"}
<match my.logs>
  @type opensearch
  index_name elastic.${key1}.${key2} # => e.g.) elastic.value1.value2
  <buffer tag, key1, key2>
    @type memory
  # <snip>

Multi workers

Since Fluentd v0.14, multi workers feature had been implemented to increase throughput with multiple processes. This feature allows Fluentd processes to use one or more CPUs. This feature will be enabled by the following system configuration:

  workers N # where N is a natural number (N >= 1).


By default, the error logger won't record the reason for a 400 error from the OpenSearch API unless you set log_level to debug. However, this results in a lot of log spam, which isn't desirable if all you want is the 400 error reasons. You can set this true to capture the 400 error reasons without all the other debug logs.

Default value is false.


By default, record body is wrapped by 'doc'. This behavior can not handle update script requests. You can set this to suppress doc wrapping and allow record body to be untouched.

Default value is false.


A list of exception that will be ignored - when the exception occurs the chunk will be discarded and the buffer retry mechanism won't be called. It is possible also to specify classes at higher level in the hierarchy. For example

ignore_exceptions ["OpenSearch::Transport::Transport::ServerError"]

will match all subclasses of ServerError - OpenSearch::Transport::Transport::Errors::BadRequest, OpenSearch::Transport::Transport::Errors::ServiceUnavailable, etc.

Default value is empty list (no exception is ignored).


Indicates whether to backup chunk when ignore exception occurs.

Default value is true.


Configure bulk_message request splitting threshold size.

Default value is -1(unlimited).

If you specify this size as negative number, bulk_message request splitting feature will be disabled.


Specify truncating caches interval.

If it is set, timer for clearing alias_indexes and template_names caches will be launched and executed.

Default value is nil.


Use legacy template or not.

Composable template documentation is Put Index Template API | OpenSearch Reference. OpenSearch still supports legacy template but we recommend to migrate to use Composable template.

Please confirm that whether the using OpenSearch cluster(s) support the composable template feature or not when turn on the brand new feature with this parameter.

<metadata> section

Users can specify whether including chunk_id information into records or not:

<match your.awesome.routing.tag>
  @type opensearch
  # Other configurations.
    include_chunk_id true
    # chunk_id_key chunk_id # Default value is "chunk_id".


Whether including chunk_id for not. Default value is false.

<match your.awesome.routing.tag>
  @type opensearch
  # Other configurations.
    include_chunk_id true


Specify chunk_id_key to store chunk_id information into records. Default value is chunk_id.

<match your.awesome.routing.tag>
  @type opensearch
  # Other configurations.
    chunk_id_key chunk_hex

Configuration - OpenSearch Input

See OpenSearch Input plugin document

Configuration - OpenSearch Filter GenID

See OpenSearch Filter GenID document

Configuration - OpenSearch Output Data Stream

Since Elasticsearch 7.9 that is predessor software of OpenSearch, Data Streams was introduced.

NOTE: This feature is slated to official support. Currently, this fetaure is beta.

You can enable this feature by specifying @type opensearch_data_stream.

@type opensearch_data_stream
data_stream_name test

The matching index template will be also created automatically.


You can specify OpenSearch data stream name by this parameter. This parameter is mandatory for opensearch_data_stream.


You can specify an existing matching index template for the data stream. If not present, it creates a new matching index template.

Default value is data_stream_name.

Configuration - AWS OpenSearch Service

This settings are effective for AWS OpenSearch Service that is successor of AWS Elasticsearch service.

<endpoint> section

AWS OpenSearch Service related settings are placed in <endpoint> directive. This is because elasticsearch-ruby does not work with OpenSearch and opensearch-ruby does not work with Elasticsearch.

Configuration example is below:

<match es.**>
  type opensearch
  logstash_format true
  include_tag_key true
  flush_interval 1s

    url https://CLUSTER_ENDPOINT_URL
    region YOUR_AWS_REGION
    # access_key_id "secret"
    # secret_access_key "foo_secret"


Specify AWS region.

  region us-east-2 # e.g.) AWS Ohio region
  # other stuffs.


Specify AWS OpenSearch Service endpoint.

  # other stuffs.

NOTE: This plugin will remove trailing slashes automatically. You don't need to pay attention to the trailing slash characters.


Specify AWS access key.

  access_key_id YOUR_AWS_ACCESS_KEY
  # other stuffs.


Specify AWS secret access key.

  secret_access_key YOUR_AWS_SECRET_ACCESS_KEY
  # other stuffs.


If you don't want to use access_key_id and secret_access_key on your endpoint configuration, you should use IAM policies instead.

Assign an IAM instance role

The first step needs to assign an IAM instance role (ROLE) to your EC2 instances. You should change its name appropriately. The attaching role should not contain no policy: We're using the role as the authenticating factor and placing the policy into the OpenSearch cluster.

Then, you should configure a policy for the OpenSearch cluster policy with substitution s for the capitalized terms:

  "Version": "2012-10-17",
  "Statement": [
      "Effect": "Allow",
      "Principal": {
        "AWS": "arn:aws:iam::ACCOUNT:role/ROLE"
      "Action": "es:*",
      "Resource": "arn:aws:es:AWS_REGION:ACCOUNT:domain/OPENSEARCH_DOMAIN/*"
      "Effect": "Allow",
      "Principal": {
        "AWS": "*"
      "Action": "es:*",
      "Resource": "arn:aws:es:AWS_REGION:ACCOUNT:domain/OPENSEARCH_DOMAIN/*",
      "Condition": {
        "IpAddress": {
          "aws:SourceIp": [

The above policy will allow your Fluentd hosts and any traffic coming from the specified IP addresses (you querying OpenSearch Dashboard) to access the various endpoints. While not ideally secure it should allow you to get up and ingesting logs without anything

Use STS assumed role as the authenticating factor

Additionally, you can use a STS assumed role as the authenticating factor and instruct the plugin to assume this role. This is useful for cross-account access and when assigning a standard role is not possible. In this case, the endpoint configuration looks like:

  assume_role_arn arn:aws:sts::ACCOUNT:role/ROLE
  assume_role_session_name SESSION_ID # Defaults to fluentd if omitted
  sts_credentials_region YOUR_AWS_STS_REGION # Defaults to region if omitted

The policy attached into your OpenSearch cluster becomes something like:

  "Version": "2012-10-17",
  "Statement": [
      "Effect": "Allow",
      "Principal": {
        "AWS": "arn:aws:sts::ACCOUNT:assumed-role/ROLE/SESSION_ID"
      "Action": "es:*",
      "Resource": "arn:aws:es:AWS_REGION:ACCOUNT:domain/ES_DOMAIN/*"

Ensure the environment to assume roles

You'll need to ensure that the environment where the Fluentd plugin runs to have the capability to assume this role, by attaching a policy something like this to the instance profile:

    "Version": "2012-10-17",
    "Statement": {
        "Effect": "Allow",
        "Action": "sts:AssumeRole",
        "Resource": "arn:aws:iam::ACCOUNT:role/ROLE"


If you want to use IAM roles for service accounts on your Amazon EKS clusters, please refer to the official documentation and specify a Service Account for your fluentd Pod.

In this case, the endpoint configuration looks like:

  region eu-west-1
  assume_role_arn "#{ENV['AWS_ROLE_ARN']}"
  assume_role_web_identity_token_file "#{ENV['AWS_WEB_IDENTITY_TOKEN_FILE']}"

Expiring AWS credentials

If you want to expire AWS credentials in certain interval, you should specify refresh_credentials_interval parameter under endpoint section:

  region eu-west-1
  # ...
  refresh_credentials_interval 3h # default is 5h (five hours).

Use OpenSearch Serverless

If you want to use Serverless version of OpenSearch service, you have to specify aoss in aws_service_name under endpoint section:

  region us-east-2
  # ...
  aws_service_name aoss # default is es that is for AWS OpenSearch Service not Serverless.


See Troubleshooting document


If you have a question, open an Issue.


There are usually a few feature requests, tagged Easy, Normal and Hard. Feel free to work on any one of them.

Pull Requests are welcomed.

Before sending a pull request or reporting an issue, please read the contribution guideline.

Running tests

Install dev dependencies:

$ gem install bundler
$ bundle install
$ bundle exec rake test
# To just run the test you are working on:
$ bundle exec rake test TEST=test/plugin/test_out_opensearch.rb TESTOPTS='--verbose --name=test_bulk_error_retags_with_error_when_configured_and_fullfilled_buffer'