Amazon Kinesis ingestion

When you enable the Kinesis indexing service, you can configure supervisors on the Overlord to manage the creation and lifetime of Kinesis indexing tasks. These indexing tasks read events using Kinesis’ own shard and sequence number mechanism to guarantee exactly-once ingestion. The supervisor oversees the state of the indexing tasks to coordinate handoffs, manage failures, and ensure that scalability and replication requirements are maintained.

This topic contains configuration reference information for the Kinesis indexing service supervisor for Apache Druid.

Setup

To use the Kinesis indexing service, you must first load the druid-kinesis-indexing-service core extension on both the Overlord and the Middle Manager. See Loading extensions for more information. Review the Kinesis known issues before deploying the druid-kinesis-indexing-service extension to production.

Supervisor spec

The following table outlines the high-level configuration options for the Kinesis supervisor object. See Supervisor API for more information.

Druid starts a new supervisor when you define a supervisor spec. To create a supervisor, send a POST request to the /druid/indexer/v1/supervisor endpoint. Once created, the supervisor persists in the configured metadata database. There can only be a single supervisor per datasource, and submitting a second spec for the same datasource overwrites the previous one.

When an Overlord gains leadership, either by being started or as a result of another Overlord failing, it spawns a supervisor for each supervisor spec in the metadata database. The supervisor then discovers running Kinesis indexing tasks and attempts to adopt them if they are compatible with the supervisor’s configuration. If they are not compatible because they have a different ingestion spec or shard allocation, the tasks are killed and the supervisor creates a new set of tasks. In this way, the supervisors persist across Overlord restarts and failovers.

The following example shows how to submit a supervisor spec for a stream with the name KinesisStream. In this example, http://SERVICE_IP:SERVICE_PORT is a placeholder for the server address of deployment and the service port.

curl -X POST "http://SERVICE_IP:SERVICE_PORT/druid/indexer/v1/supervisor" \
-H "Content-Type: application/json" \
-d '{
  "type": "kinesis",
  "spec": {
    "ioConfig": {
      "type": "kinesis",
      "stream": "KinesisStream",
      "inputFormat": {
        "type": "json"
      },
      "useEarliestSequenceNumber": true
    },
    "tuningConfig": {
      "type": "kinesis"
    },
    "dataSchema": {
      "dataSource": "KinesisStream",
      "timestampSpec": {
        "column": "timestamp",
        "format": "iso"
      },
      "dimensionsSpec": {
        "dimensions": [
          "isRobot",
          "channel",
          "flags",
          "isUnpatrolled",
          "page",
          "diffUrl",
          {
            "type": "long",
            "name": "added"
          },
          "comment",
          {
            "type": "long",
            "name": "commentLength"
          },
          "isNew",
          "isMinor",
          {
            "type": "long",
            "name": "delta"
          },
          "isAnonymous",
          "user",
          {
            "type": "long",
            "name": "deltaBucket"
          },
          {
            "type": "long",
            "name": "deleted"
          },
          "namespace",
          "cityName",
          "countryName",
          "regionIsoCode",
          "metroCode",
          "countryIsoCode",
          "regionName"
        ]
      },
      "granularitySpec": {
        "queryGranularity": "none",
        "rollup": false,
        "segmentGranularity": "hour"
      }
    }
  }
}'

POST /druid/indexer/v1/supervisor
HTTP/1.1
Host: http://SERVICE_IP:SERVICE_PORT
Content-Type: application/json
{
  "type": "kinesis",
  "spec": {
    "ioConfig": {
      "type": "kinesis",
      "stream": "KinesisStream",
      "inputFormat": {
        "type": "json"
      },
      "useEarliestSequenceNumber": true
    },
    "tuningConfig": {
      "type": "kinesis"
    },
    "dataSchema": {
      "dataSource": "KinesisStream",
      "timestampSpec": {
        "column": "timestamp",
        "format": "iso"
      },
      "dimensionsSpec": {
        "dimensions": [
          "isRobot",
          "channel",
          "flags",
          "isUnpatrolled",
          "page",
          "diffUrl",
          {
            "type": "long",
            "name": "added"
          },
          "comment",
          {
            "type": "long",
            "name": "commentLength"
          },
          "isNew",
          "isMinor",
          {
            "type": "long",
            "name": "delta"
          },
          "isAnonymous",
          "user",
          {
            "type": "long",
            "name": "deltaBucket"
          },
          {
            "type": "long",
            "name": "deleted"
          },
          "namespace",
          "cityName",
          "countryName",
          "regionIsoCode",
          "metroCode",
          "countryIsoCode",
          "regionName"
        ]
      },
      "granularitySpec": {
        "queryGranularity": "none",
        "rollup": false,
        "segmentGranularity": "hour"
      }
    }
  }
}

Supervisor I/O configuration

The following table outlines the configuration options for ioConfig:

Task autoscaler properties

The following table outlines the configuration options for autoScalerConfig:

Lag based autoscaler strategy related properties

Unlike the Kafka indexing service, Kinesis reports lag metrics measured in time difference in milliseconds between the current sequence number and latest sequence number, rather than message count.

The following table outlines the configuration options for autoScalerStrategy:

The following example shows a supervisor spec with lagBased auto scaler enabled.

Click to view the example

{
  "type": "kinesis",
  "dataSchema": {
    "dataSource": "metrics-kinesis",
    "timestampSpec": {
      "column": "timestamp",
      "format": "auto"
    },
    "dimensionsSpec": {
      "dimensions": [],
      "dimensionExclusions": [
        "timestamp",
        "value"
      ]
    },
    "metricsSpec": [
      {
        "name": "count",
        "type": "count"
      },
      {
        "name": "value_sum",
        "fieldName": "value",
        "type": "doubleSum"
      },
      {
        "name": "value_min",
        "fieldName": "value",
        "type": "doubleMin"
      },
      {
        "name": "value_max",
        "fieldName": "value",
        "type": "doubleMax"
      }
    ],
    "granularitySpec": {
      "type": "uniform",
      "segmentGranularity": "HOUR",
      "queryGranularity": "NONE"
    }
  },
  "ioConfig": {
    "stream": "metrics",
    "autoScalerConfig": {
      "enableTaskAutoScaler": true,
      "taskCountMax": 6,
      "taskCountMin": 2,
      "minTriggerScaleActionFrequencyMillis": 600000,
      "autoScalerStrategy": "lagBased",
      "lagCollectionIntervalMillis": 30000,
      "lagCollectionRangeMillis": 600000,
      "scaleOutThreshold": 600000,
      "triggerScaleOutFractionThreshold": 0.3,
      "scaleInThreshold": 100000,
      "triggerScaleInFractionThreshold": 0.9,
      "scaleActionStartDelayMillis": 300000,
      "scaleActionPeriodMillis": 60000,
      "scaleInStep": 1,
      "scaleOutStep": 2
    },
    "inputFormat": {
      "type": "json"
    },
    "endpoint": "kinesis.us-east-1.amazonaws.com",
    "taskCount": 1,
    "replicas": 1,
    "taskDuration": "PT1H"
  },
  "tuningConfig": {
    "type": "kinesis",
    "maxRowsPerSegment": 5000000
  }
}

Specify data format

The Kinesis indexing service supports both inputFormat and parser to specify the data format. Use the inputFormat to specify the data format for the Kinesis indexing service unless you need a format only supported by the legacy parser.

Supported values for inputFormat include:

• csv
• delimited
• json
• avro_stream
• avro_ocf
• protobuf

For more information, see Data formats. You can also read thrift formats using parser.

Supervisor tuning configuration

The tuningConfig object is optional. If you don’t specify the tuningConfig object, Druid uses the default configuration settings.

The following table outlines the configuration options for tuningConfig:

IndexSpec

The following table outlines the configuration options for indexSpec:

Operations

This section describes how to use the Supervisor API with the Kinesis indexing service.

AWS authentication

To authenticate with AWS, you must provide your AWS access key and AWS secret key using runtime.properties, for example:

druid.kinesis.accessKey=AKIAWxxxxxxxxxx4NCKS
druid.kinesis.secretKey=Jbytxxxxxxxxxxx2+555

Druid uses the AWS access key and AWS secret key to authenticate Kinesis API requests. If not provided, the service looks for credentials set in environment variables, via Web Identity Token, in the default profile configuration file, and from the EC2 instance profile provider (in this order).

To ingest data from Kinesis, ensure that the policy attached to your IAM role contains the necessary permissions. The required permissions depend on the value of useListShards.

If the useListShards flag is set to true, you need following permissions:

• ListStreams to list your data streams.
• Get* required for GetShardIterator.
• GetRecords to get data records from a data stream’s shard.
• ListShards to get the shards for a stream of interest.

The following is an example policy:

[
  {
    "Effect": "Allow",
    "Action": ["kinesis:List*"],
    "Resource": ["*"]
  },
  {
    "Effect": "Allow",
    "Action": ["kinesis:Get*"],
    "Resource": [<ARN for shards to be ingested>]
  }
]

If the useListShards flag is set to false, you need following permissions:

• ListStreams to list your data streams.
• Get* required for GetShardIterator.
• GetRecords to get data records from a data stream’s shard.
• DescribeStream to describe the specified data stream.

The following is an example policy:

[
  {
    "Effect": "Allow",
    "Action": ["kinesis:ListStreams"],
    "Resource": ["*"]
  },
  {
    "Effect": "Allow",
    "Action": ["kinesis:DescribeStreams"],
    "Resource": ["*"]
  },
  {
    "Effect": "Allow",
    "Action": ["kinesis:Get*"],
    "Resource": [<ARN for shards to be ingested>]
  }
]

Get supervisor status report

To retrieve the current status report for a single supervisor, send a GET request to the /druid/indexer/v1/supervisor/:supervisorId/status endpoint.

The report contains the state of the supervisor tasks, the latest sequence numbers, and an array of recently thrown exceptions reported as recentErrors. You can control the maximum size of the exceptions using the druid.supervisor.maxStoredExceptionEvents configuration.

The two properties related to the supervisor’s state are state and detailedState. The state property contains a small number of generic states that apply to any type of supervisor, while the detailedState property contains a more descriptive, implementation-specific state that may provide more insight into the supervisor’s activities.

Possible state values are PENDING, RUNNING, SUSPENDED, STOPPING, UNHEALTHY_SUPERVISOR, and UNHEALTHY_TASKS.

The following table lists detailedState values and their corresponding state mapping:

On each iteration of the supervisor’s run loop, the supervisor completes the following tasks in sequence:

1. Fetch the list of shards from Kinesis and determine the starting sequence number for each shard (either based on the last processed sequence number if continuing, or starting from the beginning or ending of the stream if this is a new stream).
2. Discover any running indexing tasks that are writing to the supervisor’s datasource and adopt them if they match the supervisor’s configuration, else signal them to stop.
3. Send a status request to each supervised task to update the view of the state of the tasks under supervision.
4. Handle tasks that have exceeded taskDuration and should transition from the reading to publishing state.
5. Handle tasks that have finished publishing and signal redundant replica tasks to stop.
6. Handle tasks that have failed and clean up the supervisor’s internal state.
7. Compare the list of healthy tasks to the requested taskCount and replicas configurations and create additional tasks if required.

The detailedState property shows additional values (marked with “first iteration only” in the preceding table) the first time the supervisor executes this run loop after startup or after resuming from a suspension. This is intended to surface initialization-type issues, where the supervisor is unable to reach a stable state. For example, if the supervisor cannot connect to Kinesis, if it’s unable to read from the stream, or cannot communicate with existing tasks. Once the supervisor is stable; that is, once it has completed a full execution without encountering any issues, detailedState will show a RUNNING state until it is stopped, suspended, or hits a failure threshold and transitions to an unhealthy state.

Update existing supervisors

To update an existing supervisor spec, send a POST request to the /druid/indexer/v1/supervisor endpoint.

When you call this endpoint on an existing supervisor for the same datasource, the running supervisor signals its tasks to stop reading and begin publishing their segments, exiting itself. Druid then uses the provided configuration from the request body to create a new supervisor with a new set of tasks that start reading from the sequence numbers, where the previous now-publishing tasks left off, but using the updated schema. In this way, configuration changes can be applied without requiring any pause in ingestion.

You can achieve seamless schema migrations by submitting the new schema using the /druid/indexer/v1/supervisor endpoint.

Suspend and resume a supervisor

To suspend a supervisor, send a POST request to the /druid/indexer/v1/supervisor/:supervisorId/suspend endpoint. Suspending a supervisor does not prevent it from operating and emitting logs and metrics. It ensures that no indexing tasks are running until the supervisor resumes.

To resume a supervisor, send a POST request to the /druid/indexer/v1/supervisor/:supervisorId/resume endpoint.

Reset a supervisor

The supervisor must be running for this endpoint to be available

To reset a supervisor, send a POST request to the /druid/indexer/v1/supervisor/:supervisorId/reset endpoint. This endpoint clears stored sequence numbers, prompting the supervisor to start reading from either the earliest or the latest sequence numbers in Kinesis (depending on the value of useEarliestSequenceNumber). After clearing stored sequence numbers, the supervisor kills and recreates active tasks, so that tasks begin reading from valid sequence numbers.

This endpoint is useful when you need to recover from a stopped state due to missing sequence numbers in Kinesis. Use this endpoint with caution as it may result in skipped messages, leading to data loss or duplicate data.

The indexing service keeps track of the latest persisted sequence number to provide exactly-once ingestion guarantees across tasks. Subsequent tasks must start reading from where the previous task completed for the generated segments to be accepted. If the messages at the expected starting sequence numbers are no longer available in Kinesis (typically because the message retention period has elapsed or the topic was removed and re-created) the supervisor will refuse to start and in-flight tasks will fail. This endpoint enables you to recover from this condition.

Terminate a supervisor

To terminate a supervisor and its associated indexing tasks, send a POST request to the /druid/indexer/v1/supervisor/:supervisorId/terminate endpoint. This places a tombstone marker in the database to prevent the supervisor from being reloaded on a restart and then gracefully shuts down the currently running supervisor. The tasks stop reading and begin publishing their segments immediately. The call returns after all tasks have been signaled to stop but before the tasks finish publishing their segments.

The terminated supervisor continues exists in the metadata store and its history can be retrieved. The only way to restart a terminated supervisor is by submitting a functioning supervisor spec to /druid/indexer/v1/supervisor.

Capacity planning

Kinesis indexing tasks run on Middle Managers and are limited by the resources available in the Middle Manager cluster. In particular, you should make sure that you have sufficient worker capacity, configured using the druid.worker.capacity property, to handle the configuration in the supervisor spec. Note that worker capacity is shared across all types of indexing tasks, so you should plan your worker capacity to handle your total indexing load, such as batch processing, streaming tasks, and merging tasks. If your workers run out of capacity, Kinesis indexing tasks queue and wait for the next available worker. This may cause queries to return partial results but will not result in data loss, assuming the tasks run before Kinesis purges those sequence numbers.

A running task can be in one of two states: reading or publishing. A task remains in reading state for the period defined in taskDuration, at which point it transitions to publishing state. A task remains in publishing state for as long as it takes to generate segments, push segments to deep storage, and have them loaded and served by a Historical process or until completionTimeout elapses.

The number of reading tasks is controlled by replicas and taskCount. In general, there are replicas * taskCount reading tasks. An exception occurs if taskCount > {numKinesisShards}, in which case Druid uses {numKinesisShards} tasks. When taskDuration elapses, these tasks transition to publishing state and replicas * taskCount new reading tasks are created. To allow for reading tasks and publishing tasks to run concurrently, there should be a minimum capacity of:

workerCapacity = 2 * replicas * taskCount

This value is for the ideal situation in which there is at most one set of tasks publishing while another set is reading. In some circumstances, it is possible to have multiple sets of tasks publishing simultaneously. This would happen if the time-to-publish (generate segment, push to deep storage, load on Historical) is greater than taskDuration. This is a valid and correct scenario but requires additional worker capacity to support. In general, it is a good idea to have taskDuration be large enough that the previous set of tasks finishes publishing before the current set begins.

Shards and segment handoff

Each Kinesis indexing task writes the events it consumes from Kinesis shards into a single segment for the segment granularity interval until it reaches one of the following limits: maxRowsPerSegment, maxTotalRows, or intermediateHandoffPeriod. At this point, the task creates a new shard for this segment granularity to contain subsequent events.

The Kinesis indexing task also performs incremental hand-offs so that the segments created by the task are not held up until the task duration is over. When the task reaches one of the maxRowsPerSegment, maxTotalRows, or intermediateHandoffPeriod limits, it hands off all the segments and creates a new set of segments for further events. This allows the task to run for longer durations without accumulating old segments locally on Middle Manager processes.

The Kinesis indexing service may still produce some small segments. For example, consider the following scenario:

• Task duration is 4 hours
• Segment granularity is set to an HOUR
• The supervisor was started at 9:10

After 4 hours at 13:10, Druid starts a new set of tasks. The events for the interval 13:00 - 14:00 may be split across existing tasks and the new set of tasks which could result in small segments. To merge them together into new segments of an ideal size (in the range of ~500-700 MB per segment), you can schedule re-indexing tasks, optionally with a different segment granularity.

For more detail, see Segment size optimization.

Determine fetch settings

Kinesis indexing tasks fetch records using fetchThreads threads. If fetchThreads is higher than the number of Kinesis shards, the excess threads are unused. Each fetch thread fetches up to recordsPerFetch records at once from a Kinesis shard, with a delay between fetches of fetchDelayMillis. The records fetched by each thread are pushed into a shared queue of size recordBufferSize. The main runner thread for each task polls up to maxRecordsPerPoll records from the queue at once.

When using Kinesis Producer Library’s aggregation feature, that is when deaggregate is set, each of these parameters refers to aggregated records rather than individual records.

The default values for these parameters are:

• fetchThreads: Twice the number of processors available to the task. The number of processors available to the task is the total number of processors on the server, divided by druid.worker.capacity (the number of task slots on that particular server).
• fetchDelayMillis: 0 (no delay between fetches).
• recordsPerFetch: 100 MB or an estimated 5% of available heap, whichever is smaller, divided by fetchThreads. For estimation purposes, Druid uses a figure of 10 KB for regular records and 1 MB for aggregated records.
• recordBufferSize: 100 MB or an estimated 10% of available heap, whichever is smaller. For estimation purposes, Druid uses a figure of 10 KB for regular records and 1 MB for aggregated records.
• maxRecordsPerPoll: 100 for regular records, 1 for aggregated records.

Kinesis places the following restrictions on calls to fetch records:

• Each data record can be up to 1 MB in size.
• Each shard can support up to 5 transactions per second for reads.
• Each shard can read up to 2 MB per second.
• The maximum size of data that GetRecords can return is 10 MB.

If the above limits are exceeded, Kinesis throws ProvisionedThroughputExceededException errors. If this happens, Druid Kinesis tasks pause by fetchDelayMillis or 3 seconds, whichever is larger, and then attempt the call again.

In most cases, the default settings for fetch parameters are sufficient to achieve good performance without excessive memory usage. However, in some cases, you may need to adjust these parameters to control fetch rate and memory usage more finely. Optimal values depend on the average size of a record and the number of consumers you have reading from a given shard, which will be replicas unless you have other consumers also reading from this Kinesis stream.

Deaggregation

The Kinesis indexing service supports de-aggregation of multiple rows packed into a single record by the Kinesis Producer Library’s aggregate method for more efficient data transfer.

To enable this feature, set deaggregate to true in your ioConfig when submitting a supervisor spec.

Resharding

When changing the shard count for a Kinesis stream, there is a window of time around the resharding operation with early shutdown of Kinesis ingestion tasks and possible task failures.

The early shutdowns and task failures are expected. They occur because the supervisor updates the shard to task group mappings as shards are closed and fully read. This ensures that tasks are not running with an assignment of closed shards that have been fully read and balances distribution of active shards across tasks.

This window with early task shutdowns and possible task failures concludes when:

• All closed shards have been fully read and the Kinesis ingestion tasks have published the data from those shards, committing the “closed” state to metadata storage.
• Any remaining tasks that had inactive shards in the assignment have been shut down. These tasks would have been created before the closed shards were completely drained.

Kinesis known issues

Before you deploy the Kinesis extension to production, consider the following known issues:

• Avoid implementing more than one Kinesis supervisor that reads from the same Kinesis stream for ingestion. Kinesis has a per-shard read throughput limit and having multiple supervisors on the same stream can reduce available read throughput for an individual supervisor’s tasks. Multiple supervisors ingesting to the same Druid datasource can also cause increased contention for locks on the datasource.
• The only way to change the stream reset policy is to submit a new ingestion spec and set up a new supervisor.
• If ingestion tasks get stuck, the supervisor does not automatically recover. You should monitor ingestion tasks and investigate if your ingestion falls behind.
• A Kinesis supervisor can sometimes compare the checkpoint offset to retention window of the stream to see if it has fallen behind. These checks fetch the earliest sequence number for Kinesis which can result in IteratorAgeMilliseconds becoming very high in AWS CloudWatch.