HBASE-12987 Pare repeated hbck output and increase verbosity in long-running tasks.

This commit is contained in:
Josh Elser 2015-04-08 16:22:22 -04:00 committed by Nick Dimiduk
parent 65bf10a47f
commit ffd7bbfd6b
2 changed files with 85 additions and 8 deletions

View File

@ -67,6 +67,7 @@ import org.apache.hadoop.hbase.fs.HFileSystem;
import org.apache.hadoop.hbase.master.HMaster;
import org.apache.hadoop.hbase.regionserver.StoreFileInfo;
import org.apache.hadoop.hbase.security.AccessDeniedException;
import org.apache.hadoop.hbase.util.HBaseFsck.ErrorReporter;
import org.apache.hadoop.hbase.protobuf.ProtobufUtil;
import org.apache.hadoop.hbase.protobuf.generated.FSProtos;
import org.apache.hadoop.hbase.regionserver.HRegion;
@ -1542,6 +1543,28 @@ public abstract class FSUtils {
*/
public static Map<String, Path> getTableStoreFilePathMap(Map<String, Path> map,
final FileSystem fs, final Path hbaseRootDir, TableName tableName)
throws IOException {
return getTableStoreFilePathMap(map, fs, hbaseRootDir, tableName, null);
}
/**
* Runs through the HBase rootdir/tablename and creates a reverse lookup map for
* table StoreFile names to the full Path.
* <br>
* Example...<br>
* Key = 3944417774205889744 <br>
* Value = hdfs://localhost:51169/user/userid/-ROOT-/70236052/info/3944417774205889744
*
* @param map map to add values. If null, this method will create and populate one to return
* @param fs The file system to use.
* @param hbaseRootDir The root directory to scan.
* @param tableName name of the table to scan.
* @param errors ErrorReporter instance or null
* @return Map keyed by StoreFile name with a value of the full Path.
* @throws IOException When scanning the directory fails.
*/
public static Map<String, Path> getTableStoreFilePathMap(Map<String, Path> map,
final FileSystem fs, final Path hbaseRootDir, TableName tableName, ErrorReporter errors)
throws IOException {
if (map == null) {
map = new HashMap<String, Path>();
@ -1554,10 +1577,16 @@ public abstract class FSUtils {
PathFilter familyFilter = new FamilyDirFilter(fs);
FileStatus[] regionDirs = fs.listStatus(tableDir, new RegionDirFilter(fs));
for (FileStatus regionDir : regionDirs) {
if (null != errors) {
errors.progress();
}
Path dd = regionDir.getPath();
// else its a region name, now look in region for families
FileStatus[] familyDirs = fs.listStatus(dd, familyFilter);
for (FileStatus familyDir : familyDirs) {
if (null != errors) {
errors.progress();
}
Path family = familyDir.getPath();
if (family.getName().equals(HConstants.RECOVERED_EDITS_DIR)) {
continue;
@ -1566,6 +1595,9 @@ public abstract class FSUtils {
// put in map
FileStatus[] familyStatus = fs.listStatus(family);
for (FileStatus sfStatus : familyStatus) {
if (null != errors) {
errors.progress();
}
Path sf = sfStatus.getPath();
map.put( sf.getName(), sf);
}
@ -1586,7 +1618,6 @@ public abstract class FSUtils {
return result;
}
/**
* Runs through the HBase rootdir and creates a reverse lookup map for
* table StoreFile names to the full Path.
@ -1602,6 +1633,26 @@ public abstract class FSUtils {
*/
public static Map<String, Path> getTableStoreFilePathMap(
final FileSystem fs, final Path hbaseRootDir)
throws IOException {
return getTableStoreFilePathMap(fs, hbaseRootDir, null);
}
/**
* Runs through the HBase rootdir and creates a reverse lookup map for
* table StoreFile names to the full Path.
* <br>
* Example...<br>
* Key = 3944417774205889744 <br>
* Value = hdfs://localhost:51169/user/userid/-ROOT-/70236052/info/3944417774205889744
*
* @param fs The file system to use.
* @param hbaseRootDir The root directory to scan.
* @param errors ErrorReporter instance or null
* @return Map keyed by StoreFile name with a value of the full Path.
* @throws IOException When scanning the directory fails.
*/
public static Map<String, Path> getTableStoreFilePathMap(
final FileSystem fs, final Path hbaseRootDir, ErrorReporter errors)
throws IOException {
Map<String, Path> map = new HashMap<String, Path>();
@ -1611,7 +1662,7 @@ public abstract class FSUtils {
// only include the directory paths to tables
for (Path tableDir : FSUtils.getTableDirs(fs, hbaseRootDir)) {
getTableStoreFilePathMap(map, fs, hbaseRootDir,
FSUtils.getTableName(tableDir));
FSUtils.getTableName(tableDir), errors);
}
return map;
}

View File

@ -648,7 +648,9 @@ public class HBaseFsck extends Configured implements Closeable {
// load regiondirs and regioninfos from HDFS
if (shouldCheckHdfs()) {
LOG.info("Loading region directories from HDFS");
loadHdfsRegionDirs();
LOG.info("Loading region information from HDFS");
loadHdfsRegionInfos();
}
@ -658,6 +660,8 @@ public class HBaseFsck extends Configured implements Closeable {
// fix the orphan tables
fixOrphanTables();
LOG.info("Checking and fixing region consistency");
// Check and fix consistency
checkAndFixConsistency();
@ -978,7 +982,10 @@ public class HBaseFsck extends Configured implements Closeable {
Configuration conf = getConf();
Path hbaseRoot = FSUtils.getRootDir(conf);
FileSystem fs = hbaseRoot.getFileSystem(conf);
Map<String, Path> allFiles = FSUtils.getTableStoreFilePathMap(fs, hbaseRoot);
LOG.info("Computing mapping of all store files");
Map<String, Path> allFiles = FSUtils.getTableStoreFilePathMap(fs, hbaseRoot, errors);
errors.print("");
LOG.info("Validating mapping using HDFS state");
for (Path path: allFiles.values()) {
boolean isReference = false;
try {
@ -1176,6 +1183,7 @@ public class HBaseFsck extends Configured implements Closeable {
}
loadTableInfosForTablesWithNoRegion();
errors.print("");
return tablesInfo;
}
@ -1355,6 +1363,7 @@ public class HBaseFsck extends Configured implements Closeable {
*/
private void suggestFixes(
SortedMap<TableName, TableInfo> tablesInfo) throws IOException {
logParallelMerge();
for (TableInfo tInfo : tablesInfo.values()) {
TableIntegrityErrorHandler handler = tInfo.new IntegrityFixSuggester(tInfo, errors);
tInfo.checkRegionChain(handler);
@ -1425,9 +1434,23 @@ public class HBaseFsck extends Configured implements Closeable {
return true;
}
/**
* Log an appropriate message about whether or not overlapping merges are computed in parallel.
*/
private void logParallelMerge() {
if (getConf().getBoolean("hbasefsck.overlap.merge.parallel", true)) {
LOG.info("Handling overlap merges in parallel. set hbasefsck.overlap.merge.parallel to" +
" false to run serially.");
} else {
LOG.info("Handling overlap merges serially. set hbasefsck.overlap.merge.parallel to" +
" true to run in parallel.");
}
}
private SortedMap<TableName, TableInfo> checkHdfsIntegrity(boolean fixHoles,
boolean fixOverlaps) throws IOException {
LOG.info("Checking HBase region split map from HDFS data...");
logParallelMerge();
for (TableInfo tInfo : tablesInfo.values()) {
TableIntegrityErrorHandler handler;
if (fixHoles || fixOverlaps) {
@ -1671,6 +1694,7 @@ public class HBaseFsck extends Configured implements Closeable {
LOG.warn("Could not load region dir " , e.getCause());
}
}
errors.print("");
}
/**
@ -2346,6 +2370,7 @@ public class HBaseFsck extends Configured implements Closeable {
loadTableInfosForTablesWithNoRegion();
logParallelMerge();
for (TableInfo tInfo : tablesInfo.values()) {
TableIntegrityErrorHandler handler = tInfo.new IntegrityFixSuggester(tInfo, errors);
if (!tInfo.checkRegionChain(handler)) {
@ -2963,15 +2988,11 @@ public class HBaseFsck extends Configured implements Closeable {
// TODO fold this into the TableIntegrityHandler
if (getConf().getBoolean("hbasefsck.overlap.merge.parallel", true)) {
LOG.info("Handling overlap merges in parallel. set hbasefsck.overlap.merge.parallel to" +
" false to run serially.");
boolean ok = handleOverlapsParallel(handler, prevKey);
if (!ok) {
return false;
}
} else {
LOG.info("Handling overlap merges serially. set hbasefsck.overlap.merge.parallel to" +
" true to run in parallel.");
for (Collection<HbckInfo> overlap : overlapGroups.asMap().values()) {
handler.handleOverlapGroup(overlap);
}
@ -3753,6 +3774,8 @@ public class HBaseFsck extends Configured implements Closeable {
static class PrintingErrorReporter implements ErrorReporter {
public int errorCount = 0;
private int showProgress;
// How frequently calls to progress() will create output
private static final int progressThreshold = 100;
Set<TableInfo> errorTables = new HashSet<TableInfo>();
@ -3867,7 +3890,7 @@ public class HBaseFsck extends Configured implements Closeable {
@Override
public synchronized void progress() {
if (showProgress++ == 10) {
if (showProgress++ == progressThreshold) {
if (!summary) {
System.out.print(".");
}
@ -3964,6 +3987,7 @@ public class HBaseFsck extends Configured implements Closeable {
// level 2: <HBASE_DIR>/<table>/*
FileStatus[] regionDirs = fs.listStatus(tableDir.getPath());
for (FileStatus regionDir : regionDirs) {
errors.progress();
String encodedName = regionDir.getPath().getName();
// ignore directories that aren't hexadecimal
if (!encodedName.toLowerCase().matches("[0-9a-f]+")) {
@ -3991,6 +4015,7 @@ public class HBaseFsck extends Configured implements Closeable {
FileStatus[] subDirs = fs.listStatus(regionDir.getPath());
Path ePath = WALSplitter.getRegionDirRecoveredEditsDir(regionDir.getPath());
for (FileStatus subDir : subDirs) {
errors.progress();
String sdName = subDir.getPath().getName();
if (!sdName.startsWith(".") && !sdName.equals(ePath.getName())) {
he.hdfsOnlyEdits = false;
@ -4031,6 +4056,7 @@ public class HBaseFsck extends Configured implements Closeable {
// only load entries that haven't been loaded yet.
if (hbi.getHdfsHRI() == null) {
try {
errors.progress();
hbck.loadHdfsRegioninfo(hbi);
} catch (IOException ioe) {
String msg = "Orphan region in HDFS: Unable to load .regioninfo from table "