Drop AWS SDK v1: replace KCL-based streaming with direct v2 poller

Remove the spark-kinesis-dynamodb module (5 files, ~800 lines) and its
dynamodb-streams-kinesis-adapter dependency, which was the sole reason
for depending on AWS SDK v1.

Replace with DynamoStreamPoller (~130 lines) that uses the v2
DynamoDbStreamsClient directly to poll shards. This eliminates KCL's
unnecessary checkpointing, lease management, and CloudWatch metrics.

- Delete spark-kinesis-dynamodb/ module entirely
- Delete AttributeValueUtils.scala (v1→v2 conversion no longer needed)
- Rewrite DynamoStreamReplication to use v2 AttributeValue natively
- Update DynamoStreamReplicationIntegrationTest to use v2 types
- Remove unused import in DynamoDBS3Export.scala
- Clean up build.sbt (remove module, adapter dep, version val)

The only remaining v1 dependency is hadoop-aws (transitive, no code imports).

Author: dkropachev

.github/workflows/build.yml

@@ -21,7 +21,6 @@ jobs:
             target/
             migrator/target/scala-2.13/classes/
             tests/target/scala-2.13/classes/
-            spark-kinesis-dynamodb/target/scala-2.13/classes/
             project/target/
           key: sbt-target-${{ runner.os }}-${{ hashFiles('**/*.scala', '**/*.sbt', 'project/build.properties') }}
           restore-keys: |

.github/workflows/tests.yml

@@ -59,7 +59,6 @@ jobs:
             target/
             migrator/target/scala-2.13/classes/
             tests/target/scala-2.13/classes/
-            spark-kinesis-dynamodb/target/scala-2.13/classes/
             project/target/
           key: sbt-target-${{ runner.os }}-${{ hashFiles('**/*.scala', '**/*.sbt', 'project/build.properties') }}
           restore-keys: |

.gitignore

@@ -178,4 +178,12 @@ AGENTS.md
 claude.log
 claude_history.json
 claude_config.json
-CLAUDE.md
+CLAUDE.md
+
+# Docker test runtime data (written by Scylla containers)
+tests/docker/scylla/*
+!tests/docker/scylla/.gitkeep
+tests/docker/scylla-source/*
+!tests/docker/scylla-source/.gitkeep
+tests/docker/parquet/*
+!tests/docker/parquet/.gitkeep

ansible/files/config.dynamodb.yml

@@ -4,7 +4,7 @@ source:
   # if using DynamoDB local - specify endpoint with http:// prefix
   # Otherwise, if using aws configure, leave commented out and it will use that
   #endpoint:
-  #  host: http://dynamo-local-db-source
+  #  host: dynamo-local-db-source
   #  port: 8000
   #credentials:
   #  accessKey: empty
@@ -36,7 +36,7 @@ target:
   # The easiest and least performant option is just point to a single nodes IP.
   # YOUR-SCYLLA-NODE0-IP scylla
   endpoint:
-    host: http://YOUR_ALTERNATOR_INSTANCE
+    host: YOUR_ALTERNATOR_INSTANCE
     port: 8000
   credentials:
     accessKey: empty

build.sbt

@@ -4,8 +4,6 @@ import sbt.librarymanagement.InclExclRule
 val awsSdkVersion = "2.23.19"
 val sparkVersion = "3.5.8"
 val hadoopVersion = "3.3.4"
-val dynamodbStreamsKinesisAdapterVersion =
-  "1.5.4" // Note This version still depends on AWS SDK 1.x, but there is no more recent version that supports AWS SDK v2.
 
 inThisBuild(
   List(
@@ -15,20 +13,6 @@ inThisBuild(
   )
 )
 
-// Augmentation of spark-streaming-kinesis-asl to also work with DynamoDB Streams
-lazy val `spark-kinesis-dynamodb` = project
-  .in(file("spark-kinesis-dynamodb"))
-  .settings(
-    libraryDependencies ++= Seq(
-      ("org.apache.spark" %% "spark-streaming-kinesis-asl" % sparkVersion)
-        .excludeAll(
-          InclExclRule("org.apache.spark", s"spark-streaming_${scalaBinaryVersion.value}")
-        ), // For some reason, the Spark dependency is not marked as provided in spark-streaming-kinesis-asl. We exclude it and then add it as provided.
-      "org.apache.spark" %% "spark-streaming" % sparkVersion % Provided,
-      "com.amazonaws" % "dynamodb-streams-kinesis-adapter" % dynamodbStreamsKinesisAdapterVersion
-    )
-  )
-
 lazy val migrator = (project in file("migrator"))
   .enablePlugins(BuildInfoPlugin)
   .settings(
@@ -43,8 +27,7 @@ lazy val migrator = (project in file("migrator"))
     Test / fork     := false,
     scalafmtOnCompile := true,
     libraryDependencies ++= Seq(
-      "org.apache.spark" %% "spark-streaming" % sparkVersion % "provided",
-      "org.apache.spark" %% "spark-sql"       % sparkVersion % "provided",
+      "org.apache.spark" %% "spark-sql" % sparkVersion % "provided",
       ("org.apache.hadoop" % "hadoop-aws" % hadoopVersion) // Note: this package still depends on the AWS SDK v1
         // Exclude the AWS bundle because it creates many conflicts when generating the assembly
         .excludeAll(
@@ -54,6 +37,7 @@ lazy val migrator = (project in file("migrator"))
       "software.amazon.awssdk" % "dynamodb"                 % awsSdkVersion,
       "software.amazon.awssdk" % "s3"                       % awsSdkVersion,
       "software.amazon.awssdk" % "sts"                      % awsSdkVersion,
+      "software.amazon.awssdk" % "cloudwatch"               % awsSdkVersion,
       "com.scylladb"          %% "spark-scylladb-connector" % "4.0.0",
       "com.github.jnr" % "jnr-posix" % "3.1.19", // Needed by the Spark ScyllaDB connector
       "com.scylladb.alternator" % "emr-dynamodb-hadoop"  % "5.8.0",
@@ -71,9 +55,9 @@ lazy val migrator = (project in file("migrator"))
       case "mime.types"                                         => MergeStrategy.first
       case PathList("META-INF", "io.netty.versions.properties") => MergeStrategy.concat
       case PathList("META-INF", "versions", _, "module-info.class") =>
-        MergeStrategy.discard // OK as long as we don’t rely on Java 9+ features such as SPI
+        MergeStrategy.discard // OK as long as we don't rely on Java 9+ features such as SPI
       case "module-info.class" =>
-        MergeStrategy.discard // OK as long as we don’t rely on Java 9+ features such as SPI
+        MergeStrategy.discard // OK as long as we don't rely on Java 9+ features such as SPI
       case x =>
         val oldStrategy = (assembly / assemblyMergeStrategy).value
         oldStrategy(x)
@@ -86,15 +70,13 @@ lazy val migrator = (project in file("migrator"))
       .runTask(Compile / fullClasspath, Compile / run / mainClass, Compile / run / runner)
       .evaluated
   )
-  .dependsOn(`spark-kinesis-dynamodb`)
 
 lazy val tests = project
   .in(file("tests"))
   .settings(
     libraryDependencies ++= Seq(
       "software.amazon.awssdk"    % "dynamodb"                  % awsSdkVersion,
       "org.apache.spark"         %% "spark-sql"                 % sparkVersion,
-      "org.apache.spark"         %% "spark-streaming"           % sparkVersion,
       "org.apache.cassandra"      % "java-driver-query-builder" % "4.18.0",
       "com.github.mjakubowski84" %% "parquet4s-core"            % "1.9.4",
       "org.apache.hadoop"         % "hadoop-client"             % hadoopVersion,
@@ -118,4 +100,4 @@ lazy val tests = project
 
 lazy val root = project
   .in(file("."))
-  .aggregate(migrator, `spark-kinesis-dynamodb`, tests)
+  .aggregate(migrator, tests)

config.yaml.example

@@ -104,6 +104,41 @@ source:
 #
 #   # how many tasks per executor?
 #   maxMapTasks: 1
+#
+#   # --- Stream replication tuning (only relevant when target.streamChanges is true) ---
+#
+#   # How often to poll DynamoDB Streams for new records (seconds). Default: 5.
+#   # streamingPollIntervalSeconds: 5
+#
+#   # Maximum consecutive poll failures before stopping stream replication. Default: 50.
+#   # streamingMaxConsecutiveErrors: 50
+#
+#   # Thread pool size for polling shards in parallel. Default: max(4, available CPUs).
+#   # streamingPollingPoolSize: 4
+#
+#   # Lease duration in milliseconds. If a worker doesn't renew within this window,
+#   # other workers can claim the shard. Default: 60000 (60 seconds).
+#   # streamingLeaseDurationMs: 60000
+#
+#   # Maximum records to fetch per GetRecords call. Default: DynamoDB default (1000).
+#   # streamingMaxRecordsPerPoll: 1000
+#
+#   # Maximum records processed per second across all shards. Default: unlimited.
+#   # streamingMaxRecordsPerSecond: 5000
+#
+#   # Publish stream replication metrics to CloudWatch. Default: false.
+#   # Requires valid AWS credentials in the source region.
+#   # streamingEnableCloudWatchMetrics: false
+#
+#   # Overall timeout for DynamoDB Streams API calls (seconds). Default: 30.
+#   # streamApiCallTimeoutSeconds: 30
+#
+#   # Per-attempt timeout for DynamoDB Streams API calls (seconds). Default: 10.
+#   # streamApiCallAttemptTimeoutSeconds: 10
+#
+#   # Timeout for awaiting parallel shard poll results (seconds). Default: 60.
+#   # Increase if the polling pool is saturated and polls take longer than expected.
+#   # streamingPollFutureTimeoutSeconds: 60
 
 # Example for loading from a DynamoDB S3 export (see https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/S3DataExport.Output.html)
 # source:

docs/source/migrate-from-dynamodb.rst

@@ -38,7 +38,7 @@ To read from the Alternator, you need to provide an ``endpoint`` instead of a ``
     type: dynamodb
     table: <table>
     endpoint:
-      host: http://<host>
+      host: <host>
       port: <port>
 
 Where ``<host>`` and ``<port>`` should be replaced with the host name and TCP port of your Alternator instance.

docs/source/stream-changes.rst

@@ -20,7 +20,7 @@ Enable this feature by setting the property ``streamChanges`` to ``true`` in the
 
 In this mode, the migrator has to be interrupted manually with ``Control`` + ``C`` (or by sending a ``SIGINT`` signal to the ``spark-submit`` process). Currently, the created stream is not deleted when the migrator is stopped. You have to delete it manually (e.g. via the AWS Console).
 
-Note that for the migration to be performed without loosing writes, the initial snapshot transfer must complete within 24 hours. Otherwise, some captured changes may be lost due to the retention period of the table’s stream.
+Note that for the migration to be performed without losing writes, the initial snapshot transfer must complete within 24 hours. Otherwise, some captured changes may be lost due to the retention period of the table’s stream.
 
 Optionally, you can skip the initial snapshot transfer and only replicate the changed items by setting the property ``skipInitialSnapshotTransfer`` to ``true``:
 
@@ -31,3 +31,137 @@ Optionally, you can skip the initial snapshot transfer and only replicate the ch
     # ...
     streamChanges: true
     skipInitialSnapshotTransfer: true
+
+Tuning Stream Replication
+-------------------------
+
+The following optional properties can be set in the **source** configuration to tune stream replication behavior:
+
+.. list-table::
+   :header-rows: 1
+   :widths: 35 15 50
+
+   * - Property
+     - Default
+     - Description
+   * - ``streamingPollIntervalSeconds``
+     - ``5``
+     - How often (in seconds) to poll DynamoDB Streams for new records.
+   * - ``streamingMaxConsecutiveErrors``
+     - ``50``
+     - Maximum consecutive poll failures before stopping stream replication.
+   * - ``streamingPollingPoolSize``
+     - ``max(4, CPUs)``
+     - Thread pool size for polling shards in parallel.
+   * - ``streamingLeaseDurationMs``
+     - ``60000``
+     - Lease duration in milliseconds. If a worker doesn't renew within this window, other workers can claim the shard.
+   * - ``streamingMaxRecordsPerPoll``
+     - unset (DynamoDB service default: ``1000``)
+     - Maximum records to fetch per ``GetRecords`` call. When not set, the DynamoDB Streams service default of 1000 is used.
+   * - ``streamingMaxRecordsPerSecond``
+     - unlimited
+     - Maximum records processed per second across all shards. Use this to avoid overwhelming the target.
+   * - ``streamingEnableCloudWatchMetrics``
+     - ``false``
+     - Publish stream replication metrics (records processed, active shards, iterator age) to CloudWatch.
+   * - ``streamApiCallTimeoutSeconds``
+     - ``30``
+     - Overall timeout for DynamoDB Streams API calls (seconds).
+   * - ``streamApiCallAttemptTimeoutSeconds``
+     - ``10``
+     - Per-attempt timeout for DynamoDB Streams API calls (seconds).
+   * - ``streamingPollFutureTimeoutSeconds``
+     - ``60``
+     - Timeout for awaiting parallel shard poll results. Increase if the polling pool is saturated and polls take longer than expected.
+
+Example:
+
+.. code-block:: yaml
+
+  source:
+    type: dynamodb
+    table: my-table
+    # ...
+    streamingPollIntervalSeconds: 2
+    streamingMaxRecordsPerSecond: 5000
+    streamingPollingPoolSize: 8
+
+Checkpoint Table
+----------------
+
+The migrator creates a DynamoDB table named ``migrator_<tableName>_<hash>`` in the source account to track replication progress. This table stores the last processed sequence number for each shard and is used for lease coordination when multiple workers are running.
+
+The checkpoint table has a single hash key (``leaseKey``) and the following columns:
+
+.. list-table::
+   :header-rows: 1
+   :widths: 25 10 65
+
+   * - Column
+     - Type
+     - Description
+   * - ``leaseKey``
+     - S
+     - Hash key. The DynamoDB Streams shard ID.
+   * - ``checkpoint``
+     - S
+     - Last successfully processed sequence number, or ``SHARD_END`` when the shard is fully consumed.
+   * - ``leaseOwner``
+     - S
+     - Worker ID that currently owns this shard's lease.
+   * - ``leaseExpiryEpochMs``
+     - N
+     - Epoch milliseconds when the lease expires. Set to ``0`` when released.
+   * - ``parentShardId``
+     - S
+     - Parent shard ID (if any), used to ensure child shards are processed after their parent is drained.
+   * - ``leaseTransferTo``
+     - S
+     - Worker ID requesting a graceful lease transfer. Removed when the lease is released.
+   * - ``leaseCounter``
+     - N
+     - Monotonically increasing counter incremented on each lease renewal.
+
+The checkpoint table persists across restarts. If the migrator is restarted, it automatically resumes from the last checkpointed position rather than re-processing the entire stream from the beginning. To force re-processing from ``TRIM_HORIZON``, manually delete the checkpoint table before restarting the migrator.
+
+CloudWatch Metrics
+------------------
+
+When ``streamingEnableCloudWatchMetrics`` is enabled, the migrator publishes the following metrics to the ``ScyllaMigrator/StreamReplication`` namespace. All metrics include a ``TableName`` dimension set to the source table name.
+
+.. list-table::
+   :header-rows: 1
+   :widths: 30 15 55
+
+   * - Metric Name
+     - Unit
+     - Description
+   * - ``RecordsProcessed``
+     - Count
+     - Number of records processed since the last publish interval (delta).
+   * - ``ActiveShards``
+     - Count
+     - Number of shards currently owned by this worker.
+   * - ``MaxIteratorAgeMs``
+     - Milliseconds
+     - Age of the oldest record seen in the most recent poll cycle, indicating how far behind the stream consumer is.
+   * - ``PollDurationMs``
+     - Milliseconds
+     - Wall-clock time of the most recent poll cycle.
+   * - ``WriteFailures``
+     - Count
+     - Number of write failures since the last publish interval (delta).
+   * - ``CheckpointFailures``
+     - Count
+     - Number of checkpoint failures since the last publish interval (delta).
+   * - ``DeadLetterItems``
+     - Count
+     - Number of items that exhausted write retries since the last publish interval (delta).
+
+Metrics are published every 60 poll cycles (e.g., every 5 minutes with the default 5-second poll interval).
+
+Reserved Attribute Names
+------------------------
+
+The stream replication process uses an internal attribute named ``_dynamo_op_type`` to distinguish between put and delete operations. If a source table has an attribute with this exact name, it will be silently overwritten during stream replication. Avoid using ``_dynamo_op_type`` as an attribute name in tables being migrated.

docs/source/tutorials/dynamodb-to-scylladb-alternator/spark-data/config.yaml

@@ -1,7 +1,7 @@
 source:
   type: dynamodb
   endpoint:
-    host: http://dynamodb
+    host: dynamodb
     port: 8000
   region: us-west-1
   credentials:
@@ -12,7 +12,7 @@ source:
 target:
   type: dynamodb
   endpoint:
-    host: http://scylla
+    host: scylla
     port: 8001
   region: us-west-1
   credentials: