Dead letter queues (DLQ)

By default, when Logstash encounters an event that it cannot process because the data contains a mapping error or some other issue, the Logstash pipeline either hangs or drops the unsuccessful event. In order to protect against data loss in this situation, you can configure Logstash to write unsuccessful events to a dead letter queue instead of dropping them.

Each event written to the dead letter queue includes the original event, metadata that describes the reason the event could not be processed, information about the plugin that wrote the event, and the timestamp when the event entered the dead letter queue.

To process events in the dead letter queue, create a Logstash pipeline configuration that uses the dead_letter_queue input plugin to read from the queue. See Processing events in the dead letter queue for more information.

HTTP request failure. If the HTTP request fails (because Elasticsearch is unreachable or because it returned an HTTP error code), the Elasticsearch output retries the entire request indefinitely. In these scenarios, the dead letter queue has no opportunity to intercept.

HTTP request success. The Elasticsearch Bulk API can perform multiple actions using the same request. If the Bulk API request is successful, it returns 200 OK, even if some documents in the batch have failed. In this situation, the errors flag for the request will be true.

The response body can include metadata indicating that one or more specific actions in the bulk request could not be performed, along with an HTTP-style status code per entry to indicate why the action could not be performed. If the DLQ is configured, individual indexing failures are routed there.

Even if you regularly process events, events remain in the dead letter queue. The dead letter queue requires manual intervention to clear it.

Dead letter queues are disabled by default. To enable dead letter queues, set the dead_letter_queue_enable option in the logstash.yml settings file:

dead_letter_queue.enable: true

Dead letter queues are stored as files in the local directory of the Logstash instance. By default, the dead letter queue files are stored in path.data/dead_letter_queue. Each pipeline has a separate queue. For example, the dead letter queue for the main pipeline is stored in LOGSTASH_HOME/data/dead_letter_queue/main by default. The queue files are numbered sequentially: 1.log, 2.log, and so on.

You can set path.dead_letter_queue in the logstash.yml file to specify a different path for the files:

path.dead_letter_queue: "path/to/data/dead_letter_queue"

Dead letter queue entries are written to a temporary file, which is then renamed to a dead letter queue segment file, which is then eligible for ingestion. The rename happens either when this temporary file is considered full, or when a period of time has elapsed since the last dead letter queue eligible event was written to the temporary file.

This length of time can be set using the dead_letter_queue.flush_interval setting. This setting is in milliseconds, and defaults to 5000ms. A low value here will mean in the event of infrequent writes to the dead letter queue more, smaller, queue files may be written, while a larger value will introduce more latency between items being "written" to the dead letter queue, and being made available for reading by the dead_letter_queue input.

Note that this value cannot be set to lower than 1000ms.

dead_letter_queue.flush_interval: 5000

dead_letter_queue.retain.age: 2d

input {
  dead_letter_queue {
  	path => "/path/to/data/dead_letter_queue"
  	pipeline_id => "main"
    clean_consumed => true
  }
}

When you are ready to process events in the dead letter queue, you create a pipeline that uses the dead_letter_queue input plugin to read from the dead letter queue. The pipeline configuration that you use depends, of course, on what you need to do. For example, if the dead letter queue contains events that resulted from a mapping error in Elasticsearch, you can create a pipeline that reads the "dead" events, removes the field that caused the mapping issue, and re-indexes the clean events into Elasticsearch.

The following example shows a simple pipeline that reads events from the dead letter queue and writes the events, including metadata, to standard output:

input {
  dead_letter_queue {
    path => "/path/to/data/dead_letter_queue" 
    commit_offsets => true 
    pipeline_id => "main" 
  }
}

output {
  stdout {
    codec => rubydebug { metadata => true }
  }
}

For another example, see Example: Processing data that has mapping errors.

When the pipeline has finished processing all the events in the dead letter queue, it will continue to run and process new events as they stream into the queue. This means that you do not need to stop your production system to handle events in the dead letter queue.

When you read from the dead letter queue, you might not want to process all the events in the queue, especially if there are a lot of old events in the queue. You can start processing events at a specific point in the queue by using the start_timestamp option. This option configures the pipeline to start processing events based on the timestamp of when they entered the queue:

input {
  dead_letter_queue {
    path => "/path/to/data/dead_letter_queue"
    start_timestamp => "2017-06-06T23:40:37"
    pipeline_id => "main"
  }
}

For this example, the pipeline starts reading all events that were delivered to the dead letter queue on or after June 6, 2017, at 23:40:37.

In this example, the user attempts to index a document that includes geo_ip data, but the data cannot be processed because it contains a mapping error:

{"geoip":{"location":"home"}}

Indexing fails because the Logstash output plugin expects a geo_point object in the location field, but the value is a string. The failed event is written to the dead letter queue, along with metadata about the error that caused the failure:

{
   "@metadata" => {
    "dead_letter_queue" => {
       "entry_time" => #<Java::OrgLogstash::Timestamp:0x5b5dacd5>,
        "plugin_id" => "fb80f1925088497215b8d037e622dec5819b503e-4",
      "plugin_type" => "elasticsearch",
           "reason" => "Could not index event to Elasticsearch. status: 400, action: [\"index\", {:_id=>nil, :_index=>\"logstash-2017.06.22\", :_type=>\"doc\", :_routing=>nil}, 2017-06-22T01:29:29.804Z My-MacBook-Pro-2.local {\"geoip\":{\"location\":\"home\"}}], response: {\"index\"=>{\"_index\"=>\"logstash-2017.06.22\", \"_type\"=>\"doc\", \"_id\"=>\"AVzNayPze1iR9yDdI2MD\", \"status\"=>400, \"error\"=>{\"type\"=>\"mapper_parsing_exception\", \"reason\"=>\"failed to parse\", \"caused_by\"=>{\"type\"=>\"illegal_argument_exception\", \"reason\"=>\"illegal latitude value [266.30859375] for geoip.location\"}}}}"
    }
  },
  "@timestamp" => 2017-06-22T01:29:29.804Z,
    "@version" => "1",
       "geoip" => {
    "location" => "home"
  },
        "host" => "My-MacBook-Pro-2.local",
     "message" => "{\"geoip\":{\"location\":\"home\"}}"
}

To process the failed event, you create the following pipeline that reads from the dead letter queue and removes the mapping problem:

input {
  dead_letter_queue {
    path => "/path/to/data/dead_letter_queue/" 
  }
}
filter {
  mutate {
    remove_field => "[geoip][location]" 
  }
}
output {
  elasticsearch{
    hosts => [ "localhost:9200" ] 
  }
}

Monitor the size of the dead letter queue before it becomes a problem. By checking it periodically, you can determine the maximum queue size that makes sense for each pipeline.

The size of the DLQ for each pipeline is available in the node stats API.

pipelines.${pipeline_id}.dead_letter_queue.queue_size_in_bytes.

Where {pipeline_id} is the name of a pipeline with DLQ enabled.

The dead letter queue cannot be cleared with the upstream pipeline running.

The dead letter queue is a directory of pages. To clear it, stop the pipeline and delete location/<file-name>.

${path.data}/dead_letter_queue/${pipeline_id}

Where {pipeline_id} is the name of a pipeline with DLQ enabled.

The pipeline creates a new dead letter queue when it starts again.