Package org.apache.druid.segment.loading

Interface DataSegmentPusher

All Known Implementing Classes:

NoopDataSegmentPusher

public interface DataSegmentPusher

Field Summary

Fields

Modifier and Type	Field	Description
static final com.google.common.base.Joiner	JOINER

Method Summary

Modifier and Type	Method	Description
static String	generateUniquePath()
static String	getDefaultStorageDir(DataSegment segment, boolean useUniquePath)
static String	getDefaultStorageDirWithExistingUniquePath(DataSegment segment, String uniquePath)
default String	getStorageDir(DataSegment dataSegment)	Deprecated. backward-compatibiliy shim that should be removed on next major release; use getStorageDir(DataSegment, boolean) instead.
default String	getStorageDir(DataSegment dataSegment, boolean useUniquePath)
Map<String,Object>	makeLoadSpec(URI finalIndexZipFilePath)
DataSegment	push(File file, DataSegment segment, boolean useUniquePath)	Pushes index files and segment descriptor to deep storage.
default DataSegment	pushToPath(File indexFilesDir, DataSegment segment, String storageDirSuffix)

Field Details

JOINER

static final com.google.common.base.Joiner JOINER

Method Details

push

DataSegment push(File file,
 DataSegment segment,
 boolean useUniquePath)
          throws IOException

Pushes index files and segment descriptor to deep storage. Expected to perform its own retries, if appropriate.

Parameters:

file - directory containing index files

segment - segment descriptor

useUniquePath - if true, pushes to a unique file path. This prevents situations where task failures or replica tasks can either overwrite or fail to overwrite existing segments leading to the possibility of different versions of the same segment ID containing different data. As an example, a Kafka indexing task starting at offset A and ending at offset B may push a segment to deep storage and then fail before writing the loadSpec to the metadata table, resulting in a replacement task being spawned. This replacement will also start at offset A but will read to offset C and will then push a segment to deep storage and write the loadSpec metadata. Without unique file paths, this can only work correctly if new segments overwrite existing segments. Suppose that at this point the task then fails so that the supervisor retries again from offset A. This 3rd attempt will overwrite the segments in deep storage before failing to write the loadSpec metadata, resulting in inconsistencies in the segment data now in deep storage and copies of the segment already loaded by historicals. If unique paths are used, caller is responsible for cleaning up segments that were pushed but were not written to the metadata table (for example when using replica tasks).

Returns:

segment descriptor

Throws:

IOException

pushToPath

default DataSegment pushToPath(File indexFilesDir,
 DataSegment segment,
 String storageDirSuffix)
                        throws IOException

Throws:

IOException

makeLoadSpec

Map<String,Object> makeLoadSpec(URI finalIndexZipFilePath)

getStorageDir

@Deprecated
default String getStorageDir(DataSegment dataSegment)

Deprecated.

backward-compatibiliy shim that should be removed on next major release; use getStorageDir(DataSegment, boolean) instead.

getStorageDir

default String getStorageDir(DataSegment dataSegment,
 boolean useUniquePath)

getDefaultStorageDir

static String getDefaultStorageDir(DataSegment segment,
 boolean useUniquePath)

getDefaultStorageDirWithExistingUniquePath

static String getDefaultStorageDirWithExistingUniquePath(DataSegment segment,
 String uniquePath)

generateUniquePath

static String generateUniquePath()