org.apache.hadoop.hbase.master.snapshot

Class SnapshotFileCache

java.lang.Object

org.apache.hadoop.hbase.master.snapshot.SnapshotFileCache

All Implemented Interfaces:

Stoppable

@InterfaceAudience.Private
 @InterfaceStability.Evolving
public class SnapshotFileCache
extends Object
implements Stoppable

Intelligently keep track of all the files for all the snapshots.

A cache of files is kept to avoid querying the FileSystem frequently. If there is a cache miss the directory modification time is used to ensure that we don't rescan directories that we already have in cache. We only check the modification times of the snapshot directories (/hbase/.snapshot/[snapshot_name]) to determine if the files need to be loaded into the cache.

New snapshots will be added to the cache and deleted snapshots will be removed when we refresh the cache. If the files underneath a snapshot directory are changed, but not the snapshot itself, we will ignore updates to that snapshot's files.

This is sufficient because each snapshot has its own directory and is added via an atomic rename once, when the snapshot is created. We don't need to worry about the data in the snapshot being run.

Further, the cache is periodically refreshed ensure that files in snapshots that were deleted are also removed from the cache.

A SnapshotFileCache.SnapshotFileInspector must be passed when creating this to allow extraction of files under /hbase/.snapshot/[snapshot name] directory, for each snapshot. This allows you to only cache files under, for instance, all the logs in the .logs directory or all the files under all the regions.

this also considers all running snapshots (those under /hbase/.snapshot/.tmp) as valid snapshots and will attempt to cache files from those snapshots as well.

Queries about a given file are thread-safe with respect to multiple queries and cache refreshes.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	SnapshotFileCache.RefreshCacheTask Simple helper task that just periodically attempts to refresh the cache
private static class	SnapshotFileCache.SnapshotDirectoryInfo Information about a snapshot directory
(package private) static interface	SnapshotFileCache.SnapshotFileInspector

Field Summary

Fields

Modifier and Type	Field and Description
private Set<String>	cache
private SnapshotFileCache.SnapshotFileInspector	fileInspector
private org.apache.hadoop.fs.FileSystem	fs
private static org.slf4j.Logger	LOG
private Timer	refreshTimer
private org.apache.hadoop.fs.Path	snapshotDir
private Map<String,SnapshotFileCache.SnapshotDirectoryInfo>	snapshots This is a helper map of information about the snapshot directories so we don't need to rescan them if they haven't changed since the last time we looked.
private boolean	stop
private org.apache.hadoop.fs.FileSystem	workingFs
private org.apache.hadoop.fs.Path	workingSnapshotDir

Constructor Summary

Constructors

Constructor and Description
SnapshotFileCache(org.apache.hadoop.conf.Configuration conf, long cacheRefreshPeriod, long cacheRefreshDelay, String refreshThreadName, SnapshotFileCache.SnapshotFileInspector inspectSnapshotFiles) Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the filesystem.
SnapshotFileCache(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path rootDir, org.apache.hadoop.fs.FileSystem workingFs, org.apache.hadoop.fs.Path workingDir, long cacheRefreshPeriod, long cacheRefreshDelay, String refreshThreadName, SnapshotFileCache.SnapshotFileInspector inspectSnapshotFiles) Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the filesystem

Method Summary

Modifier and Type	Method and Description
(package private) List<String>	getSnapshotsInProgress()
Iterable<org.apache.hadoop.fs.FileStatus>	getUnreferencedFiles(Iterable<org.apache.hadoop.fs.FileStatus> files, SnapshotManager snapshotManager) Check to see if any of the passed file names is contained in any of the snapshots.
boolean	isStopped()
private void	refreshCache()
void	stop(String why) Stop this service.
void	triggerCacheRefreshForTesting() Trigger a cache refresh, even if its before the next cache refresh.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

LOG

private static final org.slf4j.Logger LOG

stop

private volatile boolean stop

fs

private final org.apache.hadoop.fs.FileSystem fs

workingFs

private final org.apache.hadoop.fs.FileSystem workingFs

fileInspector

private final SnapshotFileCache.SnapshotFileInspector fileInspector

snapshotDir

private final org.apache.hadoop.fs.Path snapshotDir

workingSnapshotDir

private final org.apache.hadoop.fs.Path workingSnapshotDir

cache

private final Set<String> cache

snapshots

private final Map<String,SnapshotFileCache.SnapshotDirectoryInfo> snapshots

This is a helper map of information about the snapshot directories so we don't need to rescan them if they haven't changed since the last time we looked.

refreshTimer

private final Timer refreshTimer

Constructor Detail

SnapshotFileCache

public SnapshotFileCache(org.apache.hadoop.conf.Configuration conf,
                         long cacheRefreshPeriod,
                         long cacheRefreshDelay,
                         String refreshThreadName,
                         SnapshotFileCache.SnapshotFileInspector inspectSnapshotFiles)
                  throws IOException

Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the filesystem.

Immediately loads the file cache.

Parameters:

conf - to extract the configured FileSystem where the snapshots are stored and hbase root directory

cacheRefreshPeriod - frequency (ms) with which the cache should be refreshed

cacheRefreshDelay - amount of time to wait for the cache to be refreshed

refreshThreadName - name of the cache refresh thread

inspectSnapshotFiles - Filter to apply to each snapshot to extract the files.

Throws:

IOException - if the FileSystem or root directory cannot be loaded

SnapshotFileCache

public SnapshotFileCache(org.apache.hadoop.fs.FileSystem fs,
                         org.apache.hadoop.fs.Path rootDir,
                         org.apache.hadoop.fs.FileSystem workingFs,
                         org.apache.hadoop.fs.Path workingDir,
                         long cacheRefreshPeriod,
                         long cacheRefreshDelay,
                         String refreshThreadName,
                         SnapshotFileCache.SnapshotFileInspector inspectSnapshotFiles)

Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the filesystem

Parameters:

fs - FileSystem where the snapshots are stored

rootDir - hbase root directory

workingFs - FileSystem where ongoing snapshot mainifest files are stored

workingDir - Location to store ongoing snapshot manifest files

cacheRefreshPeriod - period (ms) with which the cache should be refreshed

cacheRefreshDelay - amount of time to wait for the cache to be refreshed

refreshThreadName - name of the cache refresh thread

inspectSnapshotFiles - Filter to apply to each snapshot to extract the files.

Method Detail

triggerCacheRefreshForTesting

public void triggerCacheRefreshForTesting()

Trigger a cache refresh, even if its before the next cache refresh. Does not affect pending cache refreshes.

Blocks until the cache is refreshed.

Exposed for TESTING.

getUnreferencedFiles

public Iterable<org.apache.hadoop.fs.FileStatus> getUnreferencedFiles(Iterable<org.apache.hadoop.fs.FileStatus> files,
                                                                      SnapshotManager snapshotManager)
                                                               throws IOException

Check to see if any of the passed file names is contained in any of the snapshots. First checks an in-memory cache of the files to keep. If its not in the cache, then the cache is refreshed and the cache checked again for that file. This ensures that we never return files that exist.

Note this may lead to periodic false positives for the file being referenced. Periodically, the cache is refreshed even if there are no requests to ensure that the false negatives get removed eventually. For instance, suppose you have a file in the snapshot and it gets loaded into the cache. Then at some point later that snapshot is deleted. If the cache has not been refreshed at that point, cache will still think the file system contains that file and return true, even if it is no longer present (false positive). However, if the file never was on the filesystem, we will never find it and always return false.

Parameters:

files - file to check, NOTE: Relies that files are loaded from hdfs before method is called (NOT LAZY)

Returns:

unReferencedFiles the collection of files that do not have snapshot references

Throws:

IOException - if there is an unexpected error reaching the filesystem.

refreshCache

private void refreshCache()
                   throws IOException

Throws:

IOException

getSnapshotsInProgress

List<String> getSnapshotsInProgress()
                             throws IOException

Throws:

IOException

stop

public void stop(String why)

Description copied from interface: Stoppable

Stop this service. Implementers should favor logging errors over throwing RuntimeExceptions.

Specified by:

stop in interface Stoppable

Parameters:

why - Why we're stopping.

isStopped

public boolean isStopped()

Specified by:

isStopped in interface Stoppable

Returns:

True if Stoppable.stop(String) has been closed.