Package org.apache.flink.connector.file.sink

Class FileSink<IN>

java.lang.Object
org.apache.flink.connector.file.sink.FileSink<IN>
Type Parameters:
IN - Type of the elements in the input of the sink that are also the elements to be written to its output
All Implemented Interfaces:
Serializable, org.apache.flink.api.common.SupportsConcurrentExecutionAttempts, org.apache.flink.api.connector.sink2.Sink<IN>, org.apache.flink.api.connector.sink2.SupportsCommitter<FileSinkCommittable>, org.apache.flink.api.connector.sink2.SupportsWriterState<IN,FileWriterBucketState>, org.apache.flink.api.connector.sink2.SupportsWriterState.WithCompatibleState, org.apache.flink.streaming.api.connector.sink2.SupportsPreCommitTopology<FileSinkCommittable,FileSinkCommittable>
@Experimental public class FileSink<IN> extends Object implements org.apache.flink.api.connector.sink2.Sink<IN>, org.apache.flink.api.connector.sink2.SupportsWriterState<IN,FileWriterBucketState>, org.apache.flink.api.connector.sink2.SupportsCommitter<FileSinkCommittable>, org.apache.flink.api.connector.sink2.SupportsWriterState.WithCompatibleState, org.apache.flink.streaming.api.connector.sink2.SupportsPreCommitTopology<FileSinkCommittable,FileSinkCommittable>, org.apache.flink.api.common.SupportsConcurrentExecutionAttempts
A unified sink that emits its input elements to FileSystem files within buckets. This sink achieves exactly-once semantics for both BATCH and STREAMING.

When creating the sink a basePath must be specified. The base directory contains one directory for every bucket. The bucket directories themselves contain several part files, with at least one for each parallel subtask of the sink which is writing data to that bucket. These part files contain the actual output data.

The sink uses a BucketAssigner to determine in which bucket directory each element should be written to inside the base directory. The BucketAssigner can, for example, roll on every checkpoint or use time or a property of the element to determine the bucket directory. The default BucketAssigner is a DateTimeBucketAssigner which will create one new bucket every hour. You can specify a custom BucketAssigner using the setBucketAssigner(bucketAssigner) method, after calling forRowFormat(Path, Encoder) or forBulkFormat(Path, BulkWriter.Factory).

The names of the part files could be defined using OutputFileConfig. This configuration contains a part prefix and a part suffix that will be used with a random uid assigned to each subtask of the sink and a rolling counter to determine the file names. For example with a prefix "prefix" and a suffix ".ext", a file named "prefix-81fc4980-a6af-41c8-9937-9939408a734b-17.ext" contains the data from subtask with uid 81fc4980-a6af-41c8-9937-9939408a734b of the sink and is the 17th part-file created by that subtask.

Part files roll based on the user-specified RollingPolicy. By default, a DefaultRollingPolicy is used for row-encoded sink output; a OnCheckpointRollingPolicy is used for bulk-encoded sink output.

In some scenarios, the open buckets are required to change based on time. In these cases, the user can specify a bucketCheckInterval (by default 1m) and the sink will check periodically and roll the part file if the specified rolling policy says so.

Part files can be in one of three states: in-progress, pending or finished. The reason for this is how the sink works to provide exactly-once semantics and fault-tolerance. The part file that is currently being written to is in-progress. Once a part file is closed for writing it becomes pending. When a checkpoint is successful (for STREAMING) or at the end of the job (for BATCH) the currently pending files will be moved to finished.

For STREAMING in order to guarantee exactly-once semantics in case of a failure, the sink should roll back to the state it had when that last successful checkpoint occurred. To this end, when restoring, the restored files in pending state are transferred into the finished state while any in-progress files are rolled back, so that they do not contain data that arrived after the checkpoint from which we restore.

FileSink also support compacting small files to accelerate the access speed of the resulted files. Compaction could be enabled via enableCompact. Once enabled, the compaction could only be disabled via calling disableCompact explicitly, otherwise there might be data loss.

See Also:

  • Method Details

    • createWriter

      public FileWriter<IN> createWriter(org.apache.flink.api.connector.sink2.WriterInitContext context) throws IOException
      Specified by:
      createWriter in interface org.apache.flink.api.connector.sink2.Sink<IN>
      Throws:
      IOException

    • restoreWriter

      public FileWriter<IN> restoreWriter(org.apache.flink.api.connector.sink2.WriterInitContext context, Collection<FileWriterBucketState> recoveredState) throws IOException
      Specified by:
      restoreWriter in interface org.apache.flink.api.connector.sink2.SupportsWriterState<IN,FileWriterBucketState>
      Throws:
      IOException

    • getWriterStateSerializer

      public org.apache.flink.core.io.SimpleVersionedSerializer<FileWriterBucketState> getWriterStateSerializer()
      Specified by:
      getWriterStateSerializer in interface org.apache.flink.api.connector.sink2.SupportsWriterState<IN,FileWriterBucketState>

    • createCommitter

      public org.apache.flink.api.connector.sink2.Committer<FileSinkCommittable> createCommitter(org.apache.flink.api.connector.sink2.CommitterInitContext context) throws IOException
      Specified by:
      createCommitter in interface org.apache.flink.api.connector.sink2.SupportsCommitter<IN>
      Throws:
      IOException

    • getCommittableSerializer

      public org.apache.flink.core.io.SimpleVersionedSerializer<FileSinkCommittable> getCommittableSerializer()
      Specified by:
      getCommittableSerializer in interface org.apache.flink.api.connector.sink2.SupportsCommitter<IN>

    • getWriteResultSerializer

      public org.apache.flink.core.io.SimpleVersionedSerializer<FileSinkCommittable> getWriteResultSerializer()
      Specified by:
      getWriteResultSerializer in interface org.apache.flink.streaming.api.connector.sink2.SupportsPreCommitTopology<FileSinkCommittable,FileSinkCommittable>

    • getCompatibleWriterStateNames

      public Collection<String> getCompatibleWriterStateNames()
      Specified by:
      getCompatibleWriterStateNames in interface org.apache.flink.api.connector.sink2.SupportsWriterState.WithCompatibleState

    • forRowFormat

      public static <IN> FileSink.DefaultRowFormatBuilder<IN> forRowFormat(org.apache.flink.core.fs.Path basePath, org.apache.flink.api.common.serialization.Encoder<IN> encoder)

    • forBulkFormat

      public static <IN> FileSink.DefaultBulkFormatBuilder<IN> forBulkFormat(org.apache.flink.core.fs.Path basePath, org.apache.flink.api.common.serialization.BulkWriter.Factory<IN> bulkWriterFactory)

    • addPreCommitTopology

      public org.apache.flink.streaming.api.datastream.DataStream<org.apache.flink.streaming.api.connector.sink2.CommittableMessage<FileSinkCommittable>> addPreCommitTopology(org.apache.flink.streaming.api.datastream.DataStream<org.apache.flink.streaming.api.connector.sink2.CommittableMessage<FileSinkCommittable>> committableStream)
      Specified by:
      addPreCommitTopology in interface org.apache.flink.streaming.api.connector.sink2.SupportsPreCommitTopology<FileSinkCommittable,FileSinkCommittable>