Apache
/
hbase
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

HBASE-954 Don't reassign root region until ProcessServerShutdown has split the former region server's log 

git-svn-id: https://svn.apache.org/repos/asf/hadoop/hbase/trunk@707710 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Jim Kellerman 2008-10-24 18:42:56 +00:00
parent 9b91e0bfe4
commit ae1653c001
6 changed files with 52 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CHANGES.txt
View File

@ -41,6 +41,8 @@ Release 0.19.0 - Unreleased
HBASE-950 HTable.commit no longer works with existing RowLocks though it's still in API
HBASE-728 Support for HLog appends
HBASE-952 Deadlock in HRegion.batchUpdate
HBASE-954 Don't reassign root region until ProcessServerShutdown has split
the former region server's log
IMPROVEMENTS
HBASE-901 Add a limit to key length, check key and value length on client side

9
src/java/org/apache/hadoop/hbase/master/ProcessRegionOpen.java
View File

@ -96,13 +96,8 @@ class ProcessRegionOpen extends ProcessRegionStatusChange {
regionInfo.getRegionName(), regionInfo.getStartKey());
if (!master.regionManager.isInitialMetaScanComplete()) {
// Put it on the queue to be scanned for the first time.
try {
LOG.debug("Adding " + m.toString() + " to regions to scan");
master.regionManager.addMetaRegionToScan(m);
} catch (InterruptedException e) {
throw new RuntimeException(
"Putting into metaRegionsToScan was interrupted.", e);
}
LOG.debug("Adding " + m.toString() + " to regions to scan");
master.regionManager.addMetaRegionToScan(m);
} else {
// Add it to the online meta regions
LOG.debug("Adding to onlineMetaRegions: " + m.toString());

15
src/java/org/apache/hadoop/hbase/master/ProcessServerShutdown.java
View File

@ -20,7 +20,6 @@
package org.apache.hadoop.hbase.master;
import java.io.IOException;
import java.io.UnsupportedEncodingException; //TODO: remove
import java.util.ArrayList;
import java.util.HashSet;
import java.util.List;
@ -45,11 +44,13 @@ import org.apache.hadoop.hbase.io.RowResult;
* serving, and the regions need to get reassigned.
*/
class ProcessServerShutdown extends RegionServerOperation {
private HServerAddress deadServer;
private String deadServerName;
private final HServerAddress deadServer;
private final String deadServerName;
private final boolean rootRegionServer;
private Path oldLogDir;
private boolean logSplit;
private boolean rootRescanned;
private class ToDoEntry {
boolean regionOffline;
@ -66,11 +67,14 @@ class ProcessServerShutdown extends RegionServerOperation {
/**
* @param master
* @param serverInfo
* @param rootRegionServer
*/
public ProcessServerShutdown(HMaster master, HServerInfo serverInfo) {
public ProcessServerShutdown(HMaster master, HServerInfo serverInfo,
boolean rootRegionServer) {
super(master);
this.deadServer = serverInfo.getServerAddress();
this.deadServerName = this.deadServer.toString();
this.rootRegionServer = rootRegionServer;
this.logSplit = false;
this.rootRescanned = false;
StringBuilder dirName = new StringBuilder("log_");
@ -253,6 +257,9 @@ class ProcessServerShutdown extends RegionServerOperation {
}
if (!rootAvailable()) {
// Get root region assigned now that log has been split
master.regionManager.reassignRootRegion();
// Return true so that worker does not put this request back on the
// toDoQueue.
// rootAvailable() has already put it on the delayedToDoQueue

23
src/java/org/apache/hadoop/hbase/master/RegionManager.java
View File

@ -148,7 +148,7 @@ class RegionManager implements HConstants {
// Scans the meta table
metaScannerThread = new MetaScanner(master, this);
unassignRootRegion();
reassignRootRegion();
}
void start() {
@ -158,15 +158,11 @@ class RegionManager implements HConstants {
"RegionManager.metaScanner");
}
/*
* Unassign the root region.
* This method would be used in case where root region server had died
* without reporting in. Currently, we just flounder and never recover. We
* could 'notice' dead region server in root scanner -- if we failed access
* multiple times -- but reassigning root is catastrophic.
*
*/
void unassignRootRegion() {
void unsetRootRegion() {
rootRegionLocation.set(null);
}
void reassignRootRegion() {
rootRegionLocation.set(null);
if (!master.shutdownRequested) {
unassignedRegions.put(HRegionInfo.ROOT_REGIONINFO, ZERO_L);
@ -818,9 +814,8 @@ class RegionManager implements HConstants {
/**
* Add a meta region to the scan queue
* @param m MetaRegion that needs to get scanned
* @throws InterruptedException
*/
public void addMetaRegionToScan(MetaRegion m) throws InterruptedException {
public void addMetaRegionToScan(MetaRegion m) {
metaScannerThread.addMetaRegionToScan(m);
}
@ -937,6 +932,9 @@ class RegionManager implements HConstants {
/**
* @param regionName
* @param info
* @param server
* @param op
*/
public void startAction(byte[] regionName, HRegionInfo info,
HServerAddress server, int op) {
@ -956,6 +954,7 @@ class RegionManager implements HConstants {
/**
* @param regionName
* @param op
*/
public void endAction(byte[] regionName, int op) {
switch (op) {

34
src/java/org/apache/hadoop/hbase/master/ServerManager.java
View File

@ -125,10 +125,17 @@ class ServerManager implements HConstants {
// The startup message was from a known server with the same name.
// Timeout the old one right away.
HServerAddress root = master.getRootRegionLocation();
boolean rootServer = false;
if (root != null && root.equals(storedInfo.getServerAddress())) {
master.regionManager.unassignRootRegion();
master.regionManager.setRootRegionLocation(null);
rootServer = true;
}
try {
master.toDoQueue.put(
new ProcessServerShutdown(master, storedInfo, rootServer));
} catch (InterruptedException e) {
LOG.error("Insertion into toDoQueue was interrupted", e);
}
master.delayedToDoQueue.put(new ProcessServerShutdown(master, storedInfo));
}
// record new server
@ -254,7 +261,7 @@ class ServerManager implements HConstants {
LOG.info("Processing " + msgs[i] + " from " + serverName);
HRegionInfo info = msgs[i].getRegionInfo();
if (info.isRootRegion()) {
master.regionManager.unassignRootRegion();
master.regionManager.reassignRootRegion();
} else if (info.isMetaTable()) {
master.regionManager.offlineMetaRegion(info.getStartKey());
}
@ -484,7 +491,7 @@ class ServerManager implements HConstants {
master.shutdown();
}
master.connection.setRootRegionLocation(null);
master.regionManager.unassignRootRegion();
master.regionManager.reassignRootRegion();
} else {
boolean reassignRegion = !region.isOffline();
@ -537,7 +544,7 @@ class ServerManager implements HConstants {
// This method can be called a couple of times during shutdown.
if (master.getRootRegionLocation() != null &&
info.getServerAddress().equals(master.getRootRegionLocation())) {
master.regionManager.unassignRootRegion();
master.regionManager.reassignRootRegion();
}
LOG.info("Cancelling lease for " + serverName);
try {
@ -682,10 +689,15 @@ class ServerManager implements HConstants {
LOG.info(server + " lease expired");
// Remove the server from the known servers list and update load info
HServerInfo info = serversToServerInfo.remove(server);
boolean rootServer = false;
if (info != null) {
HServerAddress root = master.getRootRegionLocation();
if (root != null && root.equals(info.getServerAddress())) {
master.regionManager.unassignRootRegion();
// NOTE: If the server was serving the root region, we cannot reassign
// it here because the new server will start serving the root region
// before ProcessServerShutdown has a chance to split the log file.
master.regionManager.unsetRootRegion();
rootServer = true;
}
String serverName = info.getServerAddress().toString();
HServerLoad load = serversToLoad.remove(serverName);
@ -704,11 +716,13 @@ class ServerManager implements HConstants {
serversToServerInfo.notifyAll();
}
// NOTE: If the server was serving the root region, we cannot reassign it
// here because the new server will start serving the root region before
// the ProcessServerShutdown operation has a chance to split the log file.
if (info != null) {
master.delayedToDoQueue.put(new ProcessServerShutdown(master, info));
try {
master.toDoQueue.put(
new ProcessServerShutdown(master, info, rootServer));
} catch (InterruptedException e) {
LOG.error("Insertion into toDoQueue was interrupted", e);
}
}
}
}

4
src/java/org/apache/hadoop/hbase/regionserver/HLog.java
View File

@ -662,7 +662,7 @@ public class HLog extends Thread implements HConstants, Syncable {
try {
for (int i = 0; i < logfiles.length; i++) {
if (LOG.isDebugEnabled()) {
LOG.debug("Splitting " + i + " of " + logfiles.length + ": " +
LOG.debug("Splitting " + (i + 1) + " of " + logfiles.length + ": " +
logfiles[i].getPath());
}
// Check for empty file.
@ -703,7 +703,7 @@ public class HLog extends Thread implements HConstants, Syncable {
logWriters.put(regionName, w);
if (LOG.isDebugEnabled()) {
LOG.debug("Creating new log file writer for path " + logfile +
" and region " + regionName);
" and region " + Bytes.toString(regionName));
}
if (old != null) {