Introduce cursor tracking akin to jdbc input (logstash-plugins#205)

Provide field value tracking, persisted to disk on each search_after page.

Adds `:last_value` and `:present` placeholders, allowing the plugin to inject the cursor value and now-30 seconds, respectively, in the query string.

Useful to track new data being written to an index or series of indices.

Works best with nano second precision timestamps added by Elasticsearch's Ingest Pipelines.

---------

Co-authored-by: Joel Andritsch <[email protected]>
Co-authored-by: Rob Bavey <[email protected]>
Co-authored-by: Karen Metts <[email protected]>

Author: 4 people

CHANGELOG.md

@@ -1,3 +1,6 @@
+## 4.22.0
+  - Add "cursor"-like index tracking [#205](https://github.com/logstash-plugins/logstash-input-elasticsearch/pull/205)
+
 ## 4.21.2
   - Add elastic-transport client support used in elasticsearch-ruby 8.x [#225](https://github.com/logstash-plugins/logstash-input-elasticsearch/pull/225)
 

docs/index.asciidoc

@@ -48,7 +48,7 @@ This would create an Elasticsearch query with the following format:
       "sort": [ "_doc" ]
     }'
 
-
+[id="plugins-{type}s-{plugin}-scheduling"]
 ==== Scheduling
 
 Input from this plugin can be scheduled to run periodically according to a specific
@@ -103,6 +103,133 @@ Common causes are:
  - When the hit result contains top-level fields that are {logstash-ref}/processing.html#reserved-fields[reserved in Logstash] but do not have the expected shape. Use the <<plugins-{type}s-{plugin}-target>> directive to avoid conflicts with the top-level namespace.
  - When <<plugins-{type}s-{plugin}-docinfo>> is enabled and the docinfo fields cannot be merged into the hit result. Combine <<plugins-{type}s-{plugin}-target>> and <<plugins-{type}s-{plugin}-docinfo_target>> to avoid conflict.
 
+[id="plugins-{type}s-{plugin}-cursor"]
+==== Tracking a field's value across runs
+
+.Technical Preview: Tracking a field's value
+****
+The feature that allows tracking a field's value across runs is in _Technical Preview_.
+Configuration options and implementation details are subject to change in minor releases without being preceded by deprecation warnings.
+****
+
+Some uses cases require tracking the value of a particular field between two jobs.
+Examples include:
+
+* avoiding the need to re-process the entire result set of a long query after an unplanned restart
+* grabbing only new data from an index instead of processing the entire set on each job.
+
+The Elasticsearch input plugin provides the <<plugins-{type}s-{plugin}-tracking_field>> and <<plugins-{type}s-{plugin}-tracking_field_seed>> options.
+When <<plugins-{type}s-{plugin}-tracking_field>> is set, the plugin records the value of that field for the last document retrieved in a run into
+a file. 
+(The file location defaults to <<plugins-{type}s-{plugin}-last_run_metadata_path>>.)
+
+You can then inject this value in the query using the placeholder `:last_value`. 
+The value will be injected into the query before execution, and then updated after the query completes if new data was found.
+
+This feature works best when:
+
+* the query sorts by the tracking field,
+* the timestamp field is added by {es}, and
+* the field type has enough resolution so that two events are unlikely to have the same value.
+
+Consider using a tracking field whose type is https://www.elastic.co/guide/en/elasticsearch/reference/current/date_nanos.html[date nanoseconds].
+If the tracking field is of this data type, you can use an extra placeholder called `:present` to inject the nano-second based value of "now-30s".
+This placeholder is useful as the right-hand side of a range filter, allowing the collection of
+new data but leaving partially-searchable bulk request data to the next scheduled job.
+
+[id="plugins-{type}s-{plugin}-tracking-sample"]
+===== Sample configuration: Track field value across runs
+
+This section contains a series of steps to help you set up the "tailing" of data being written to a set of indices, using a date nanosecond field added by an Elasticsearch ingest pipeline and the `tracking_field` capability of this plugin.
+
+. Create ingest pipeline that adds Elasticsearch's `_ingest.timestamp` field to the documents as `event.ingested`:
++
+[source, json]
+    PUT _ingest/pipeline/my-pipeline
+    {
+      "processors": [
+              {
+            "script": {
+              "lang": "painless",
+              "source": "ctx.putIfAbsent(\"event\", [:]); ctx.event.ingested = metadata().now.format(DateTimeFormatter.ISO_INSTANT);"
+            }
+          }
+      ]
+    }
+
+[start=2]
+. Create an index mapping where the tracking field is of date nanosecond type and invokes the defined pipeline:
++
+[source, json]
+    PUT /_template/my_template
+    {
+      "index_patterns": ["test-*"],
+      "settings": {
+        "index.default_pipeline": "my-pipeline",
+      },
+      "mappings": {
+        "properties": {
+          "event": {
+            "properties": {
+              "ingested": {
+                "type": "date_nanos",
+                "format": "strict_date_optional_time_nanos"
+              }
+            }
+          }
+        }
+      }
+    }
+
+[start=3]
+. Define a query that looks at all data of the indices, sorted by the tracking field, and with a range filter since the last value seen until present:
++
+[source,json]
+{
+  "query": {
+    "range": {
+      "event.ingested": {
+        "gt": ":last_value",
+        "lt": ":present"
+      }
+    }
+  },
+  "sort": [
+    {
+      "event.ingested": {
+        "order": "asc",
+        "format": "strict_date_optional_time_nanos",
+        "numeric_type": "date_nanos"
+      }
+    }
+  ]
+}
+
+[start=4]
+. Configure the Elasticsearch input to query the indices with the query defined above, every minute, and track the `event.ingested` field:
++
+[source, ruby]
+    input {
+      elasticsearch {
+        id => tail_test_index
+        hosts => [ 'https://..']
+        api_key => '....'
+        index => 'test-*'
+        query => '{ "query": { "range": { "event.ingested": { "gt": ":last_value", "lt": ":present"}}}, "sort": [ { "event.ingested": {"order": "asc", "format": "strict_date_optional_time_nanos", "numeric_type" : "date_nanos" } } ] }'
+        tracking_field => "[event][ingested]"
+        slices => 5 # optional use of slices to speed data processing, should be equal to or less than number of primary shards
+        schedule => '* * * * *' # every minute
+        schedule_overlap => false # don't accumulate jobs if one takes longer than 1 minute
+      }
+    }
+
+With this sample setup, new documents are indexed into a `test-*` index.
+The next scheduled run:
+
+* selects all new documents since the last observed value of the tracking field,
+* uses {ref}/point-in-time-api.html#point-in-time-api[Point in time (PIT)] + {ref}/paginate-search-results.html#search-after[Search after] to paginate through all the data, and
+* updates the value of the field at the end of the pagination.
+
 [id="plugins-{type}s-{plugin}-options"]
 ==== Elasticsearch Input configuration options
 
@@ -123,12 +250,14 @@ This plugin supports the following configuration options plus the <<plugins-{typ
 | <<plugins-{type}s-{plugin}-ecs_compatibility>> |<<string,string>>|No
 | <<plugins-{type}s-{plugin}-hosts>> |<<array,array>>|No
 | <<plugins-{type}s-{plugin}-index>> |<<string,string>>|No
+| <<plugins-{type}s-{plugin}-last_run_metadata_path>> |<<string,string>>|No
 | <<plugins-{type}s-{plugin}-password>> |<<password,password>>|No
 | <<plugins-{type}s-{plugin}-proxy>> |<<uri,uri>>|No
 | <<plugins-{type}s-{plugin}-query>> |<<string,string>>|No
 | <<plugins-{type}s-{plugin}-response_type>> |<<string,string>>, one of `["hits","aggregations"]`|No
 | <<plugins-{type}s-{plugin}-request_timeout_seconds>> | <<number,number>>|No
 | <<plugins-{type}s-{plugin}-schedule>> |<<string,string>>|No
+| <<plugins-{type}s-{plugin}-schedule_overlap>> |<<boolean,boolean>>|No
 | <<plugins-{type}s-{plugin}-scroll>> |<<string,string>>|No
 | <<plugins-{type}s-{plugin}-search_api>> |<<string,string>>, one of `["auto", "search_after", "scroll"]`|No
 | <<plugins-{type}s-{plugin}-size>> |<<number,number>>|No
@@ -148,6 +277,8 @@ This plugin supports the following configuration options plus the <<plugins-{typ
 | <<plugins-{type}s-{plugin}-ssl_verification_mode>> |<<string,string>>, one of `["full", "none"]`|No
 | <<plugins-{type}s-{plugin}-socket_timeout_seconds>> | <<number,number>>|No
 | <<plugins-{type}s-{plugin}-target>> | {logstash-ref}/field-references-deepdive.html[field reference] | No
+| <<plugins-{type}s-{plugin}-tracking_field>> |<<string,string>>|No
+| <<plugins-{type}s-{plugin}-tracking_field_seed>> |<<string,string>>|No
 | <<plugins-{type}s-{plugin}-retries>> | <<number,number>>|No
 | <<plugins-{type}s-{plugin}-user>> |<<string,string>>|No
 |=======================================================================
@@ -327,6 +458,17 @@ Check out {ref}/api-conventions.html#api-multi-index[Multi Indices
 documentation] in the Elasticsearch documentation for info on
 referencing multiple indices.
 
+[id="plugins-{type}s-{plugin}-last_run_metadata_path"]
+===== `last_run_metadata_path`
+
+  * Value type is <<string,string>>
+  * There is no default value for this setting.
+
+The path to store the last observed value of the tracking field, when used.
+By default this file is stored as `<path.data>/plugins/inputs/elasticsearch/<pipeline_id>/last_run_value`.
+
+This setting should point to file, not a directory, and Logstash must have read+write access to this file.
+
 [id="plugins-{type}s-{plugin}-password"]
 ===== `password`
 
@@ -407,6 +549,19 @@ for example: "* * * * *" (execute query every minute, on the minute)
 There is no schedule by default. If no schedule is given, then the statement is run
 exactly once.
 
+[id="plugins-{type}s-{plugin}-schedule_overlap"]
+===== `schedule_overlap`
+
+  * Value type is <<boolean,boolean>>
+  * Default value is `true`
+
+Whether to allow queuing of a scheduled run if a run is occurring.
+While this is ideal for ensuring a new run happens immediately after the previous on finishes if there
+is a lot of work to do, but given the queue is unbounded it may lead to an out of memory over long periods of time
+if the queue grows continuously.
+
+When in doubt, set `schedule_overlap` to false (it may become the default value in the future).
+
 [id="plugins-{type}s-{plugin}-scroll"]
 ===== `scroll`
 
@@ -617,6 +772,28 @@ When the `target` is set to a field reference, the `_source` of the hit is place
 This option can be useful to avoid populating unknown fields when a downstream schema such as ECS is enforced.
 It is also possible to target an entry in the event's metadata, which will be available during event processing but not exported to your outputs (e.g., `target \=> "[@metadata][_source]"`).
 
+[id="plugins-{type}s-{plugin}-tracking_field"]
+===== `tracking_field`
+
+* Value type is <<string,string>>
+* There is no default value for this setting.
+
+Which field from the last event of a previous run will be used a cursor value for the following run.
+The value of this field is injected into each query if the query uses the placeholder `:last_value`.
+For the first query after a pipeline is started, the value used is either read from <<plugins-{type}s-{plugin}-last_run_metadata_path>> file,
+or taken from <<plugins-{type}s-{plugin}-tracking_field_seed>> setting.
+
+Note: The tracking value is updated after each page is read and at the end of each Point in Time. In case of a crash the last saved value will be used so some duplication of data can occur. For this reason the use of unique document IDs for each event is recommended in the downstream destination.
+
+[id="plugins-{type}s-{plugin}-tracking_field_seed"]
+===== `tracking_field_seed`
+
+* Value type is <<string,string>>
+* Default value is `"1970-01-01T00:00:00.000000000Z"`
+
+The starting value for the <<plugins-{type}s-{plugin}-tracking_field>> if there is no <<plugins-{type}s-{plugin}-last_run_metadata_path>> already.
+This field defaults to the nanosecond precision ISO8601 representation of `epoch`, or "1970-01-01T00:00:00.000000000Z", given nano-second precision timestamps are the
+most reliable data format to use for this feature.
 
 [id="plugins-{type}s-{plugin}-user"]
 ===== `user`

lib/logstash/inputs/elasticsearch.rb

@@ -73,6 +73,7 @@ class LogStash::Inputs::Elasticsearch < LogStash::Inputs::Base
 
   require 'logstash/inputs/elasticsearch/paginated_search'
   require 'logstash/inputs/elasticsearch/aggregation'
+  require 'logstash/inputs/elasticsearch/cursor_tracker'
 
   include LogStash::PluginMixins::ECSCompatibilitySupport(:disabled, :v1, :v8 => :v1)
   include LogStash::PluginMixins::ECSCompatibilitySupport::TargetCheck
@@ -124,6 +125,20 @@ class LogStash::Inputs::Elasticsearch < LogStash::Inputs::Base
   # by this pipeline input.
   config :slices, :validate => :number
 
+  # Enable tracking the value of a given field to be used as a cursor
+  # Main concerns:
+  #       * using anything other than _event.timestamp easily leads to data loss
+  #       * the first "synchronization run can take a long time"
+  config :tracking_field, :validate => :string
+
+  # Define the initial seed value of the tracking_field
+  config :tracking_field_seed, :validate => :string, :default => "1970-01-01T00:00:00.000000000Z"
+
+  # The location of where the tracking field value will be stored
+  # The value is persisted after each scheduled run (and not per result)
+  # If it's not set it defaults to '${path.data}/plugins/inputs/elasticsearch/<pipeline_id>/last_run_value'
+  config :last_run_metadata_path, :validate => :string
+
   # If set, include Elasticsearch document information such as index, type, and
   # the id in the event.
   #
@@ -262,6 +277,10 @@ class LogStash::Inputs::Elasticsearch < LogStash::Inputs::Base
   # exactly once.
   config :schedule, :validate => :string
 
+  # Allow scheduled runs to overlap (enabled by default). Setting to false will
+  # only start a new scheduled run after the previous one completes.
+  config :schedule_overlap, :validate => :boolean
+
   # If set, the _source of each hit will be added nested under the target instead of at the top-level
   config :target, :validate => :field_reference
 
@@ -335,16 +354,30 @@ def register
 
     setup_query_executor
 
+    setup_cursor_tracker
+
     @client
   end
 
   def run(output_queue)
     if @schedule
-      scheduler.cron(@schedule) { @query_executor.do_run(output_queue) }
+      scheduler.cron(@schedule, :overlap => @schedule_overlap) do
+        @query_executor.do_run(output_queue, get_query_object())
+      end
       scheduler.join
     else
-      @query_executor.do_run(output_queue)
+      @query_executor.do_run(output_queue, get_query_object())
+    end
+  end
+
+  def get_query_object
+    if @cursor_tracker
+      query = @cursor_tracker.inject_cursor(@query)
+      @logger.debug("new query is #{query}")
+    else
+      query = @query
     end
+    LogStash::Json.load(query)
   end
 
   ##
@@ -354,6 +387,11 @@ def push_hit(hit, output_queue, root_field = '_source')
     event = event_from_hit(hit, root_field)
     decorate(event)
     output_queue << event
+    record_last_value(event)
+  end
+
+  def record_last_value(event)
+    @cursor_tracker.record_last_value(event) if @tracking_field
   end
 
   def event_from_hit(hit, root_field)
@@ -676,6 +714,28 @@ def setup_query_executor
                       end
   end
 
+  def setup_cursor_tracker
+    return unless @tracking_field
+    return unless @query_executor.is_a?(LogStash::Inputs::Elasticsearch::SearchAfter)
+
+    if @resolved_search_api != "search_after" || @response_type != "hits"
+      raise ConfigurationError.new("The `tracking_field` feature can only be used with `search_after` non-aggregation queries")
+    end
+
+    @cursor_tracker = CursorTracker.new(last_run_metadata_path: last_run_metadata_path,
+                                        tracking_field: @tracking_field,
+                                        tracking_field_seed: @tracking_field_seed)
+    @query_executor.cursor_tracker = @cursor_tracker
+  end
+
+  def last_run_metadata_path
+    return @last_run_metadata_path if @last_run_metadata_path
+
+    last_run_metadata_path = ::File.join(LogStash::SETTINGS.get_value("path.data"), "plugins", "inputs", "elasticsearch", pipeline_id, "last_run_value")
+    FileUtils.mkdir_p ::File.dirname(last_run_metadata_path)
+    last_run_metadata_path
+  end
+
   def get_transport_client_class
     # LS-core includes `elasticsearch` gem. The gem is composed of two separate gems: `elasticsearch-api` and `elasticsearch-transport`
     # And now `elasticsearch-transport` is old, instead we have `elastic-transport`.

lib/logstash/inputs/elasticsearch/aggregation.rb

@@ -12,14 +12,9 @@ def initialize(client, plugin)
           @client = client
           @plugin_params = plugin.params
 
+          @index = @plugin_params["index"]
           @size = @plugin_params["size"]
-          @query = @plugin_params["query"]
           @retries = @plugin_params["retries"]
-          @agg_options = {
-            :index => @plugin_params["index"],
-            :size  => 0
-          }.merge(:body => @query)
-
           @plugin = plugin
         end
 
@@ -33,10 +28,18 @@ def retryable(job_name, &block)
           false
         end
 
-        def do_run(output_queue)
+        def aggregation_options(query_object)
+          {
+            :index => @index,
+            :size => 0,
+            :body => query_object
+          }
+        end
+
+        def do_run(output_queue, query_object)
           logger.info("Aggregation starting")
           r = retryable(AGGREGATION_JOB) do
-            @client.search(@agg_options)
+            @client.search(aggregation_options(query_object))
           end
           @plugin.push_hit(r, output_queue, 'aggregations') if r
         end