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

HBASE-24295 [Chaos Monkey] abstract logging through the class hierarchy 

Adds `protected abstract Logger getLogger()` to `Action` so that
implementation's names are logged when actions are performed.

Signed-off-by: stack <stack@apache.org>
Signed-off-by: Jan Hentschel <jan.hentschel@ultratendency.com>

foo
This commit is contained in:
Nick Dimiduk 2020-04-30 15:00:57 -07:00 committed by Nick Dimiduk
parent 9e975d1b98
commit 8d1228ece7
37 changed files with 337 additions and 193 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

79
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/Action.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -39,12 +39,11 @@ import org.apache.hadoop.hbase.chaos.monkies.PolicyBasedChaosMonkey;
import org.apache.hadoop.hbase.client.Admin;
import org.apache.hadoop.hbase.util.Bytes;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
/**
* A (possibly mischievous) action that the ChaosMonkey can perform.
*/
public class Action {
public abstract class Action {
public static final String KILL_MASTER_TIMEOUT_KEY =
"hbase.chaosmonkey.action.killmastertimeout";
@ -65,8 +64,6 @@ public class Action {
public static final String START_NAMENODE_TIMEOUT_KEY =
"hbase.chaosmonkey.action.startnamenodetimeout";
private static final Logger LOG = LoggerFactory.getLogger(Action.class);
protected static final long KILL_MASTER_TIMEOUT_DEFAULT = PolicyBasedChaosMonkey.TIMEOUT;
protected static final long START_MASTER_TIMEOUT_DEFAULT = PolicyBasedChaosMonkey.TIMEOUT;
protected static final long KILL_RS_TIMEOUT_DEFAULT = PolicyBasedChaosMonkey.TIMEOUT;
@ -121,6 +118,11 @@ public class Action {
cluster.getConf().getLong(START_NAMENODE_TIMEOUT_KEY, START_NAMENODE_TIMEOUT_DEFAULT);
}
/**
* Retrieve the instance's {@link Logger}, for use throughout the class hierarchy.
*/
protected abstract Logger getLogger();
public void perform() throws Exception { }
/** Returns current region servers - active master */
@ -138,110 +140,110 @@ public class Action {
ArrayList<ServerName> tmp = new ArrayList<>(count);
tmp.addAll(regionServers);
tmp.removeAll(masters);
return tmp.toArray(new ServerName[tmp.size()]);
return tmp.toArray(new ServerName[0]);
}
protected void killMaster(ServerName server) throws IOException {
LOG.info("Killing master:" + server);
getLogger().info("Killing master:" + server);
cluster.killMaster(server);
cluster.waitForMasterToStop(server, killMasterTimeout);
LOG.info("Killed master server:" + server);
getLogger().info("Killed master server:" + server);
}
protected void startMaster(ServerName server) throws IOException {
LOG.info("Starting master:" + server.getHostname());
getLogger().info("Starting master:" + server.getHostname());
cluster.startMaster(server.getHostname(), server.getPort());
cluster.waitForActiveAndReadyMaster(startMasterTimeout);
LOG.info("Started master: " + server);
getLogger().info("Started master: " + server);
}
protected void stopRs(ServerName server) throws IOException {
LOG.info("Stopping regionserver " + server);
getLogger().info("Stopping regionserver " + server);
cluster.stopRegionServer(server);
cluster.waitForRegionServerToStop(server, killRsTimeout);
LOG.info(String.format("Stopping regionserver %s. Reported num of rs: %s", server,
getLogger().info(String.format("Stopping regionserver %s. Reported num of rs: %s", server,
cluster.getClusterStatus().getLiveServersLoad().size()));
}
protected void suspendRs(ServerName server) throws IOException {
LOG.info("Suspending regionserver %s" + server);
getLogger().info("Suspending regionserver %s" + server);
cluster.suspendRegionServer(server);
if(!(cluster instanceof MiniHBaseCluster)){
cluster.waitForRegionServerToStop(server, killRsTimeout);
}
LOG.info(String.format("Suspending regionserver %s. Reported num of rs: %s", server,
getLogger().info(String.format("Suspending regionserver %s. Reported num of rs: %s", server,
cluster.getClusterStatus().getLiveServersLoad().size()));
}
protected void resumeRs(ServerName server) throws IOException {
LOG.info("Resuming regionserver " + server);
getLogger().info("Resuming regionserver " + server);
cluster.resumeRegionServer(server);
if(!(cluster instanceof MiniHBaseCluster)){
cluster.waitForRegionServerToStart(server.getHostname(), server.getPort(), startRsTimeout);
}
LOG.info(String.format("Resuming regionserver %s. Reported num of rs: %s", server,
getLogger().info(String.format("Resuming regionserver %s. Reported num of rs: %s", server,
cluster.getClusterStatus().getLiveServersLoad().size()));
}
protected void killRs(ServerName server) throws IOException {
LOG.info("Killing regionserver " + server);
getLogger().info("Killing regionserver " + server);
cluster.killRegionServer(server);
cluster.waitForRegionServerToStop(server, killRsTimeout);
LOG.info(String.format("Killed regionserver %s. Reported num of rs: %s", server,
getLogger().info(String.format("Killed regionserver %s. Reported num of rs: %s", server,
cluster.getClusterStatus().getLiveServersLoad().size()));
}
protected void startRs(ServerName server) throws IOException {
LOG.info("Starting regionserver " + server.getAddress());
getLogger().info("Starting regionserver " + server.getAddress());
cluster.startRegionServer(server.getHostname(), server.getPort());
cluster.waitForRegionServerToStart(server.getHostname(), server.getPort(), startRsTimeout);
LOG.info(String.format("Started regionserver %s. Reported num of rs: %s", server.getAddress(),
cluster.getClusterStatus().getLiveServersLoad().size()));
getLogger().info(String.format("Started regionserver %s. Reported num of rs: %s",
server.getAddress(), cluster.getClusterStatus().getLiveServersLoad().size()));
}
protected void killZKNode(ServerName server) throws IOException {
LOG.info("Killing zookeeper node " + server);
getLogger().info("Killing zookeeper node " + server);
cluster.killZkNode(server);
cluster.waitForZkNodeToStop(server, killZkNodeTimeout);
LOG.info(String.format("Killed zookeeper node %s. Reported num of rs: %s", server,
getLogger().info(String.format("Killed zookeeper node %s. Reported num of rs: %s", server,
cluster.getClusterStatus().getLiveServersLoad().size()));
}
protected void startZKNode(ServerName server) throws IOException {
LOG.info("Starting zookeeper node " + server.getHostname());
getLogger().info("Starting zookeeper node " + server.getHostname());
cluster.startZkNode(server.getHostname(), server.getPort());
cluster.waitForZkNodeToStart(server, startZkNodeTimeout);
LOG.info("Started zookeeper node " + server);
getLogger().info("Started zookeeper node " + server);
}
protected void killDataNode(ServerName server) throws IOException {
LOG.info("Killing datanode " + server);
getLogger().info("Killing datanode " + server);
cluster.killDataNode(server);
cluster.waitForDataNodeToStop(server, killDataNodeTimeout);
LOG.info(String.format("Killed datanode %s. Reported num of rs: %s", server,
getLogger().info(String.format("Killed datanode %s. Reported num of rs: %s", server,
cluster.getClusterStatus().getLiveServersLoad().size()));
}
protected void startDataNode(ServerName server) throws IOException {
LOG.info("Starting datanode " + server.getHostname());
getLogger().info("Starting datanode " + server.getHostname());
cluster.startDataNode(server);
cluster.waitForDataNodeToStart(server, startDataNodeTimeout);
LOG.info("Started datanode " + server);
getLogger().info("Started datanode " + server);
}
protected void killNameNode(ServerName server) throws IOException {
LOG.info("Killing namenode : " + server.getHostname());
getLogger().info("Killing namenode : " + server.getHostname());
cluster.killNameNode(server);
cluster.waitForNameNodeToStop(server, killNameNodeTimeout);
LOG.info("Killed namenode: " + server + ". Reported num of rs:"
getLogger().info("Killed namenode: " + server + ". Reported num of rs:"
+ cluster.getClusterStatus().getServersSize());
}
protected void startNameNode(ServerName server) throws IOException {
LOG.info("Starting Namenode : " + server.getHostname());
getLogger().info("Starting Namenode : " + server.getHostname());
cluster.startNameNode(server);
cluster.waitForNameNodeToStart(server, startNameNodeTimeout);
LOG.info("Started namenode: " + server);
getLogger().info("Started namenode: " + server);
}
protected void unbalanceRegions(ClusterStatus clusterStatus,
@ -253,7 +255,8 @@ public class Action {
// Ugh.
List<byte[]> regions = new LinkedList<byte[]>(serverLoad.getRegionsLoad().keySet());
int victimRegionCount = (int)Math.ceil(fractionOfRegions * regions.size());
LOG.debug("Removing " + victimRegionCount + " regions from " + server.getServerName());
getLogger().debug("Removing " + victimRegionCount + " regions from "
+ server.getServerName());
for (int i = 0; i < victimRegionCount; ++i) {
int victimIx = RandomUtils.nextInt(regions.size());
String regionId = HRegionInfo.encodeRegionName(regions.remove(victimIx));
@ -261,7 +264,7 @@ public class Action {
}
}
LOG.info("Moving " + victimRegions.size() + " regions from " + fromServers.size()
getLogger().info("Moving " + victimRegions.size() + " regions from " + fromServers.size()
+ " servers to " + toServers.size() + " different servers");
Admin admin = this.context.getHBaseIntegrationTestingUtility().getHBaseAdmin();
for (byte[] victimRegion : victimRegions) {
@ -281,10 +284,10 @@ public class Action {
try {
result = admin.balancer();
} catch (Exception e) {
LOG.warn("Got exception while doing balance ", e);
getLogger().warn("Got exception while doing balance ", e);
}
if (!result) {
LOG.error("Balancer didn't succeed");
getLogger().error("Balancer didn't succeed");
}
}
@ -293,7 +296,7 @@ public class Action {
try {
admin.setBalancerRunning(onOrOff, synchronous);
} catch (Exception e) {
LOG.warn("Got exception while switching balance ", e);
getLogger().warn("Got exception while switching balance ", e);
}
}

8
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/AddColumnAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -40,6 +40,10 @@ public class AddColumnAction extends Action {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void init(ActionContext context) throws IOException {
super.init(context);
@ -61,7 +65,7 @@ public class AddColumnAction extends Action {
return;
}
LOG.debug("Performing action: Adding " + columnDescriptor + " to " + tableName);
getLogger().debug("Performing action: Adding " + columnDescriptor + " to " + tableName);
tableDescriptor.addFamily(columnDescriptor);
admin.modifyTable(tableName, tableDescriptor);

20
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/BatchRestartRsAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -21,7 +21,6 @@ package org.apache.hadoop.hbase.chaos.actions;
import java.util.HashSet;
import java.util.List;
import java.util.Set;
import org.apache.hadoop.hbase.ServerName;
import org.apache.hadoop.hbase.chaos.monkies.PolicyBasedChaosMonkey;
import org.slf4j.Logger;
@ -32,17 +31,20 @@ import org.slf4j.LoggerFactory;
*/
public class BatchRestartRsAction extends RestartActionBaseAction {
float ratio; //ratio of regionservers to restart
private static final Logger LOG =
LoggerFactory.getLogger(BatchRestartRsAction.class);
private static final Logger LOG = LoggerFactory.getLogger(BatchRestartRsAction.class);
public BatchRestartRsAction(long sleepTime, float ratio) {
super(sleepTime);
this.ratio = ratio;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info(String.format("Performing action: Batch restarting %d%% of region servers",
getLogger().info(String.format("Performing action: Batch restarting %d%% of region servers",
(int)(ratio * 100)));
List<ServerName> selectedServers = PolicyBasedChaosMonkey.selectRandomItems(getCurrentServers(),
ratio);
@ -55,7 +57,7 @@ public class BatchRestartRsAction extends RestartActionBaseAction {
if (context.isStopping()) {
break;
}
LOG.info("Killing region server:" + server);
getLogger().info("Killing region server:" + server);
cluster.killRegionServer(server);
killedServers.add(server);
}
@ -64,13 +66,13 @@ public class BatchRestartRsAction extends RestartActionBaseAction {
cluster.waitForRegionServerToStop(server, PolicyBasedChaosMonkey.TIMEOUT);
}
LOG.info("Killed " + killedServers.size() + " region servers. Reported num of rs:"
getLogger().info("Killed " + killedServers.size() + " region servers. Reported num of rs:"
+ cluster.getClusterStatus().getServersSize());
sleep(sleepTime);
for (ServerName server : killedServers) {
LOG.info("Starting region server:" + server.getHostname());
getLogger().info("Starting region server:" + server.getHostname());
cluster.startRegionServer(server.getHostname(), server.getPort());
}
@ -79,7 +81,7 @@ public class BatchRestartRsAction extends RestartActionBaseAction {
server.getPort(),
PolicyBasedChaosMonkey.TIMEOUT);
}
LOG.info("Started " + killedServers.size() +" region servers. Reported num of rs:"
getLogger().info("Started " + killedServers.size() +" region servers. Reported num of rs:"
+ cluster.getClusterStatus().getServersSize());
}
}

12
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/ChangeBloomFilterAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -47,13 +47,17 @@ public class ChangeBloomFilterAction extends Action {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
Random random = new Random();
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
LOG.info("Performing action: Change bloom filter on all columns of table "
getLogger().info("Performing action: Change bloom filter on all columns of table "
+ tableName);
HTableDescriptor tableDescriptor = admin.getTableDescriptor(tableName);
HColumnDescriptor[] columnDescriptors = tableDescriptor.getColumnFamilies();
@ -67,11 +71,11 @@ public class ChangeBloomFilterAction extends Action {
for (HColumnDescriptor descriptor : columnDescriptors) {
int bloomFilterIndex = random.nextInt(bloomArraySize);
LOG.debug("Performing action: About to set bloom filter type to "
getLogger().debug("Performing action: About to set bloom filter type to "
+ bloomArray[bloomFilterIndex] + " on column "
+ descriptor.getNameAsString() + " of table " + tableName);
descriptor.setBloomFilterType(bloomArray[bloomFilterIndex]);
LOG.debug("Performing action: Just set bloom filter type to "
getLogger().debug("Performing action: Just set bloom filter type to "
+ bloomArray[bloomFilterIndex] + " on column "
+ descriptor.getNameAsString() + " of table " + tableName);
}

10
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/ChangeCompressionAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -46,6 +46,10 @@ public class ChangeCompressionAction extends Action {
this.random = new Random();
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void init(ActionContext context) throws IOException {
super.init(context);
@ -82,12 +86,12 @@ public class ChangeCompressionAction extends Action {
algo.returnCompressor(c);
break;
} catch (Throwable t) {
LOG.info("Performing action: Changing compression algorithms to " + algo +
getLogger().info("Performing action: Changing compression algorithms to " + algo +
" is not supported, pick another one");
}
} while (true);
LOG.debug("Performing action: Changing compression algorithms on "
getLogger().debug("Performing action: Changing compression algorithms on "
+ tableName.getNameAsString() + " to " + algo);
for (HColumnDescriptor descriptor : columnDescriptors) {
if (random.nextBoolean()) {

10
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/ChangeEncodingAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -44,6 +44,10 @@ public class ChangeEncodingAction extends Action {
this.random = new Random();
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void init(ActionContext context) throws IOException {
super.init(context);
@ -59,13 +63,13 @@ public class ChangeEncodingAction extends Action {
return;
}
LOG.debug("Performing action: Changing encodings on " + tableName);
getLogger().debug("Performing action: Changing encodings on " + tableName);
// possible DataBlockEncoding id's
int[] possibleIds = {0, 2, 3, 4, 6};
for (HColumnDescriptor descriptor : columnDescriptors) {
short id = (short) possibleIds[random.nextInt(possibleIds.length)];
descriptor.setDataBlockEncoding(DataBlockEncoding.getEncodingById(id));
LOG.debug("Set encoding of column family " + descriptor.getNameAsString()
getLogger().debug("Set encoding of column family " + descriptor.getNameAsString()
+ " to: " + descriptor.getDataBlockEncoding());
}

9
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/ChangeSplitPolicyAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -45,17 +45,20 @@ public class ChangeSplitPolicyAction extends Action {
this.random = new Random();
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
LOG.info("Performing action: Change split policy of table " + tableName);
getLogger().info("Performing action: Change split policy of table " + tableName);
HTableDescriptor tableDescriptor = admin.getTableDescriptor(tableName);
String chosenPolicy = possiblePolicies[random.nextInt(possiblePolicies.length)];
tableDescriptor.setRegionSplitPolicyClassName(chosenPolicy);
LOG.info("Changing " + tableName + " split policy to " + chosenPolicy);
getLogger().info("Changing " + tableName + " split policy to " + chosenPolicy);
admin.modifyTable(tableName, tableDescriptor);
}
}

12
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/ChangeVersionsAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -34,17 +34,21 @@ import org.slf4j.LoggerFactory;
* Always keeps at least 1 as the number of versions.
*/
public class ChangeVersionsAction extends Action {
private final TableName tableName;
private static final Logger LOG = LoggerFactory.getLogger(ChangeVersionsAction.class);
private final TableName tableName;
private final Random random;
private Admin admin;
private Random random;
public ChangeVersionsAction(TableName tableName) {
this.tableName = tableName;
this.random = new Random();
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void init(ActionContext context) throws IOException {
super.init(context);
@ -68,7 +72,7 @@ public class ChangeVersionsAction extends Action {
if (context.isStopping()) {
return;
}
LOG.debug("Performing action: Changing versions on " + tableName.getNameAsString());
getLogger().debug("Performing action: Changing versions on " + tableName.getNameAsString());
admin.modifyTable(tableName, tableDescriptor);
}
}

22
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/CompactRandomRegionOfTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -33,11 +33,11 @@ import org.slf4j.LoggerFactory;
* Region that queues a compaction of a random region from the table.
*/
public class CompactRandomRegionOfTableAction extends Action {
private static final Logger LOG = LoggerFactory.getLogger(CompactRandomRegionOfTableAction.class);
private final int majorRatio;
private final long sleepTime;
private final TableName tableName;
private static final Logger LOG =
LoggerFactory.getLogger(CompactRandomRegionOfTableAction.class);
public CompactRandomRegionOfTableAction(
TableName tableName, float majorRatio) {
@ -51,33 +51,37 @@ public class CompactRandomRegionOfTableAction extends Action {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
boolean major = RandomUtils.nextInt(100) < majorRatio;
LOG.info("Performing action: Compact random region of table "
getLogger().info("Performing action: Compact random region of table "
+ tableName + ", major=" + major);
List<HRegionInfo> regions = admin.getTableRegions(tableName);
if (regions == null || regions.isEmpty()) {
LOG.info("Table " + tableName + " doesn't have regions to compact");
getLogger().info("Table " + tableName + " doesn't have regions to compact");
return;
}
HRegionInfo region = PolicyBasedChaosMonkey.selectRandomItem(
regions.toArray(new HRegionInfo[regions.size()]));
regions.toArray(new HRegionInfo[0]));
try {
if (major) {
LOG.debug("Major compacting region " + region.getRegionNameAsString());
getLogger().debug("Major compacting region " + region.getRegionNameAsString());
admin.majorCompactRegion(region.getRegionName());
} else {
LOG.debug("Compacting region " + region.getRegionNameAsString());
getLogger().debug("Compacting region " + region.getRegionNameAsString());
admin.compactRegion(region.getRegionName());
}
} catch (Exception ex) {
LOG.warn("Compaction failed, might be caused by other chaos: " + ex.getMessage());
getLogger().warn("Compaction failed, might be caused by other chaos: " + ex.getMessage());
}
if (sleepTime > 0) {
Thread.sleep(sleepTime);

13
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/CompactTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -29,10 +29,11 @@ import org.slf4j.LoggerFactory;
* Action that queues a table compaction.
*/
public class CompactTableAction extends Action {
private static final Logger LOG = LoggerFactory.getLogger(CompactTableAction.class);
private final TableName tableName;
private final int majorRatio;
private final long sleepTime;
private static final Logger LOG = LoggerFactory.getLogger(CompactTableAction.class);
public CompactTableAction(TableName tableName, float majorRatio) {
this(-1, tableName, majorRatio);
@ -45,13 +46,17 @@ public class CompactTableAction extends Action {
this.sleepTime = sleepTime;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
boolean major = RandomUtils.nextInt(100) < majorRatio;
LOG.info("Performing action: Compact table " + tableName + ", major=" + major);
getLogger().info("Performing action: Compact table " + tableName + ", major=" + major);
try {
if (major) {
admin.majorCompact(tableName);
@ -59,7 +64,7 @@ public class CompactTableAction extends Action {
admin.compact(tableName);
}
} catch (Exception ex) {
LOG.warn("Compaction failed, might be caused by other chaos: " + ex.getMessage());
getLogger().warn("Compaction failed, might be caused by other chaos: " + ex.getMessage());
}
if (sleepTime > 0) {
Thread.sleep(sleepTime);

11
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/DecreaseMaxHFileSizeAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -25,10 +25,13 @@ import org.apache.hadoop.hbase.TableName;
import org.apache.hadoop.hbase.client.Admin;
import java.util.Random;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
public class DecreaseMaxHFileSizeAction extends Action {
private static final Logger LOG = LoggerFactory.getLogger(DecreaseMaxHFileSizeAction.class);
private static final long minFileSize = 1 * 1024 * 1024 * 1024L;
private static final long minFileSize = 1024 * 1024 * 1024L;
private final long sleepTime;
private final TableName tableName;
@ -40,6 +43,10 @@ public class DecreaseMaxHFileSizeAction extends Action {
this.random = new Random();
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();

14
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/DumpClusterStatusAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -19,7 +19,6 @@
package org.apache.hadoop.hbase.chaos.actions;
import java.io.IOException;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
@ -27,8 +26,11 @@ import org.slf4j.LoggerFactory;
* Action to dump the cluster status.
*/
public class DumpClusterStatusAction extends Action {
private static final Logger LOG =
LoggerFactory.getLogger(DumpClusterStatusAction.class);
private static final Logger LOG = LoggerFactory.getLogger(DumpClusterStatusAction.class);
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void init(ActionContext context) throws IOException {
@ -37,7 +39,7 @@ public class DumpClusterStatusAction extends Action {
@Override
public void perform() throws Exception {
LOG.debug("Performing action: Dump cluster status");
LOG.info("Cluster status\n" + cluster.getClusterStatus());
getLogger().debug("Performing action: Dump cluster status");
getLogger().info("Cluster status\n" + cluster.getClusterStatus());
}
}

16
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/FlushRandomRegionOfTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -46,25 +46,29 @@ public class FlushRandomRegionOfTableAction extends Action {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
LOG.info("Performing action: Flush random region of table " + tableName);
getLogger().info("Performing action: Flush random region of table " + tableName);
List<HRegionInfo> regions = admin.getTableRegions(tableName);
if (regions == null || regions.isEmpty()) {
LOG.info("Table " + tableName + " doesn't have regions to flush");
getLogger().info("Table " + tableName + " doesn't have regions to flush");
return;
}
HRegionInfo region = PolicyBasedChaosMonkey.selectRandomItem(
regions.toArray(new HRegionInfo[regions.size()]));
LOG.debug("Flushing region " + region.getRegionNameAsString());
regions.toArray(new HRegionInfo[0]));
getLogger().debug("Flushing region " + region.getRegionNameAsString());
try {
admin.flushRegion(region.getRegionName());
} catch (Exception ex) {
LOG.warn("Flush failed, might be caused by other chaos: " + ex.getMessage());
getLogger().warn("Flush failed, might be caused by other chaos: " + ex.getMessage());
}
if (sleepTime > 0) {
Thread.sleep(sleepTime);

13
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/FlushTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -28,8 +28,7 @@ import org.slf4j.LoggerFactory;
* Action that tries to flush a table.
*/
public class FlushTableAction extends Action {
private static final Logger LOG =
LoggerFactory.getLogger(FlushTableAction.class);
private static final Logger LOG = LoggerFactory.getLogger(FlushTableAction.class);
private final long sleepTime;
private final TableName tableName;
@ -42,6 +41,10 @@ public class FlushTableAction extends Action {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
@ -52,11 +55,11 @@ public class FlushTableAction extends Action {
return;
}
LOG.info("Performing action: Flush table " + tableName);
getLogger().info("Performing action: Flush table " + tableName);
try {
admin.flush(tableName);
} catch (Exception ex) {
LOG.warn("Flush failed, might be caused by other chaos: " + ex.getMessage());
getLogger().warn("Flush failed, might be caused by other chaos: " + ex.getMessage());
}
if (sleepTime > 0) {
Thread.sleep(sleepTime);

11
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/ForceBalancerAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -25,8 +25,11 @@ import org.slf4j.LoggerFactory;
* Action that tries to force a balancer run.
*/
public class ForceBalancerAction extends Action {
private static final Logger LOG =
LoggerFactory.getLogger(ForceBalancerAction.class);
private static final Logger LOG = LoggerFactory.getLogger(ForceBalancerAction.class);
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
@ -34,7 +37,7 @@ public class ForceBalancerAction extends Action {
if (context.isStopping()) {
return;
}
LOG.info("Balancing regions");
getLogger().info("Balancing regions");
forceBalancer();
}
}

14
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/MergeRandomAdjacentRegionsOfTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -46,22 +46,26 @@ public class MergeRandomAdjacentRegionsOfTableAction extends Action {
this.sleepTime = sleepTime;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
LOG.info("Performing action: Merge random adjacent regions of table " + tableName);
getLogger().info("Performing action: Merge random adjacent regions of table " + tableName);
List<HRegionInfo> regions = admin.getTableRegions(tableName);
if (regions == null || regions.size() < 2) {
LOG.info("Table " + tableName + " doesn't have enough regions to merge");
getLogger().info("Table " + tableName + " doesn't have enough regions to merge");
return;
}
int i = RandomUtils.nextInt(regions.size() - 1);
HRegionInfo a = regions.get(i++);
HRegionInfo b = regions.get(i);
LOG.debug("Merging " + a.getRegionNameAsString() + " and " + b.getRegionNameAsString());
getLogger().debug("Merging " + a.getRegionNameAsString() + " and " + b.getRegionNameAsString());
// Don't try the merge if we're stopping
if (context.isStopping()) {
@ -71,7 +75,7 @@ public class MergeRandomAdjacentRegionsOfTableAction extends Action {
try {
admin.mergeRegions(a.getEncodedNameAsBytes(), b.getEncodedNameAsBytes(), false);
} catch (Exception ex) {
LOG.warn("Merge failed, might be caused by other chaos: " + ex.getMessage());
getLogger().warn("Merge failed, might be caused by other chaos: " + ex.getMessage());
}
if (sleepTime > 0) {
Thread.sleep(sleepTime);

12
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/MoveRandomRegionOfTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -46,6 +46,10 @@ public class MoveRandomRegionOfTableAction extends Action {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
if (sleepTime > 0) {
@ -55,16 +59,16 @@ public class MoveRandomRegionOfTableAction extends Action {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
LOG.info("Performing action: Move random region of table " + tableName);
getLogger().info("Performing action: Move random region of table " + tableName);
List<HRegionInfo> regions = admin.getTableRegions(tableName);
if (regions == null || regions.isEmpty()) {
LOG.info("Table " + tableName + " doesn't have regions to move");
getLogger().info("Table " + tableName + " doesn't have regions to move");
return;
}
HRegionInfo region = PolicyBasedChaosMonkey.selectRandomItem(
regions.toArray(new HRegionInfo[regions.size()]));
LOG.debug("Unassigning region " + region.getRegionNameAsString());
getLogger().debug("Unassigning region " + region.getRegionNameAsString());
admin.unassign(region.getRegionName(), false);
if (sleepTime > 0) {
Thread.sleep(sleepTime);

14
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/MoveRegionsOfTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -46,6 +46,10 @@ public class MoveRegionsOfTableAction extends Action {
this(-1, MonkeyConstants.DEFAULT_MOVE_REGIONS_MAX_TIME, tableName);
}
@Override protected Logger getLogger() {
return LOG;
}
public MoveRegionsOfTableAction(long sleepTime, long maxSleepTime, TableName tableName) {
this.sleepTime = sleepTime;
this.tableName = tableName;
@ -62,10 +66,10 @@ public class MoveRegionsOfTableAction extends Action {
Collection<ServerName> serversList = admin.getClusterStatus().getServers();
ServerName[] servers = serversList.toArray(new ServerName[serversList.size()]);
LOG.info("Performing action: Move regions of table " + tableName);
getLogger().info("Performing action: Move regions of table " + tableName);
List<HRegionInfo> regions = admin.getTableRegions(tableName);
if (regions == null || regions.isEmpty()) {
LOG.info("Table " + tableName + " doesn't have regions to move");
getLogger().info("Table " + tableName + " doesn't have regions to move");
return;
}
@ -82,10 +86,10 @@ public class MoveRegionsOfTableAction extends Action {
try {
String destServerName =
servers[RandomUtils.nextInt(servers.length)].getServerName();
LOG.debug("Moving " + regionInfo.getRegionNameAsString() + " to " + destServerName);
getLogger().debug("Moving " + regionInfo.getRegionNameAsString() + " to " + destServerName);
admin.move(regionInfo.getEncodedNameAsBytes(), Bytes.toBytes(destServerName));
} catch (Exception ex) {
LOG.warn("Move failed, might be caused by other chaos: " + ex.getMessage());
getLogger().warn("Move failed, might be caused by other chaos: " + ex.getMessage());
}
if (sleepTime > 0) {
Thread.sleep(sleepTime);

8
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RemoveColumnAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -47,6 +47,10 @@ public class RemoveColumnAction extends Action {
random = new Random();
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void init(ActionContext context) throws IOException {
super.init(context);
@ -68,7 +72,7 @@ public class RemoveColumnAction extends Action {
index = random.nextInt(columnDescriptors.length);
}
byte[] colDescName = columnDescriptors[index].getName();
LOG.debug("Performing action: Removing " + Bytes.toString(colDescName)+ " from "
getLogger().debug("Performing action: Removing " + Bytes.toString(colDescName)+ " from "
+ tableName.getNameAsString());
tableDescriptor.removeFamily(colDescName);

32
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartActionBaseAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -37,8 +37,12 @@ public class RestartActionBaseAction extends Action {
this.sleepTime = sleepTime;
}
@Override protected Logger getLogger() {
return LOG;
}
void sleep(long sleepTime) {
LOG.info("Sleeping for:" + sleepTime);
getLogger().info("Sleeping for:" + sleepTime);
Threads.sleep(sleepTime);
}
@ -49,10 +53,10 @@ public class RestartActionBaseAction extends Action {
return;
}
LOG.info("Killing master: " + server);
getLogger().info("Killing master: " + server);
killMaster(server);
sleep(sleepTime);
LOG.info("Starting master: " + server);
getLogger().info("Starting master: " + server);
startMaster(server);
}
@ -68,10 +72,10 @@ public class RestartActionBaseAction extends Action {
if (context.isStopping()) {
return;
}
LOG.info("Stopping region server: " + server);
getLogger().info("Stopping region server: " + server);
stopRs(server);
sleep(sleepTime);
LOG.info("Starting region server: " + server);
getLogger().info("Starting region server: " + server);
startRs(server);
}
@ -81,10 +85,10 @@ public class RestartActionBaseAction extends Action {
if (context.isStopping()) {
return;
}
LOG.info("Killing region server: " + server);
getLogger().info("Killing region server: " + server);
killRs(server);
sleep(sleepTime);
LOG.info("Starting region server: " + server);
getLogger().info("Starting region server: " + server);
startRs(server);
}
@ -94,10 +98,10 @@ public class RestartActionBaseAction extends Action {
if (context.isStopping()) {
return;
}
LOG.info("Killing zookeeper node: " + server);
getLogger().info("Killing zookeeper node: " + server);
killZKNode(server);
sleep(sleepTime);
LOG.info("Starting zookeeper node: " + server);
getLogger().info("Starting zookeeper node: " + server);
startZKNode(server);
}
@ -107,10 +111,10 @@ public class RestartActionBaseAction extends Action {
if (context.isStopping()) {
return;
}
LOG.info("Killing data node: " + server);
getLogger().info("Killing data node: " + server);
killDataNode(server);
sleep(sleepTime);
LOG.info("Starting data node: " + server);
getLogger().info("Starting data node: " + server);
startDataNode(server);
}
@ -120,10 +124,10 @@ public class RestartActionBaseAction extends Action {
if (context.isStopping()) {
return;
}
LOG.info("Killing name node: " + server);
getLogger().info("Killing name node: " + server);
killNameNode(server);
sleep(sleepTime);
LOG.info("Starting name node: " + server);
getLogger().info("Starting name node: " + server);
startNameNode(server);
}
}

9
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartActiveMasterAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -31,9 +31,14 @@ public class RestartActiveMasterAction extends RestartActionBaseAction {
public RestartActiveMasterAction(long sleepTime) {
super(sleepTime);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info("Performing action: Restart active master");
getLogger().info("Performing action: Restart active master");
ServerName master = cluster.getClusterStatus().getMaster();
restartMaster(master, sleepTime);

12
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartActiveNameNodeAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -51,9 +51,13 @@ public class RestartActiveNameNodeAction extends RestartActionBaseAction {
super(sleepTime);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info("Performing action: Restart active namenode");
getLogger().info("Performing action: Restart active namenode");
Configuration conf = FSUtils.getRootDir(getConf()).getFileSystem(getConf()).getConf();
String nameServiceID = DFSUtil.getNamenodeNameServiceId(conf);
if (!HAUtil.isHAEnabled(conf, nameServiceID)) {
@ -85,9 +89,9 @@ public class RestartActiveNameNodeAction extends RestartActionBaseAction {
if (activeNamenode == null) {
throw new Exception("No active Name node found in zookeeper under " + hadoopHAZkNode);
}
LOG.info("Found active namenode host:" + activeNamenode);
getLogger().info("Found active namenode host:" + activeNamenode);
ServerName activeNNHost = ServerName.valueOf(activeNamenode, -1, -1);
LOG.info("Restarting Active NameNode :" + activeNamenode);
getLogger().info("Restarting Active NameNode :" + activeNamenode);
restartNameNode(activeNNHost, sleepTime);
}
}

10
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartRandomDataNodeAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -42,9 +42,13 @@ public class RestartRandomDataNodeAction extends RestartActionBaseAction {
super(sleepTime);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info("Performing action: Restart random data node");
getLogger().info("Performing action: Restart random data node");
ServerName server = PolicyBasedChaosMonkey.selectRandomItem(getDataNodes());
restartDataNode(server, sleepTime);
}
@ -57,6 +61,6 @@ public class RestartRandomDataNodeAction extends RestartActionBaseAction {
for (DatanodeInfo dataNode: dfsClient.datanodeReport(HdfsConstants.DatanodeReportType.LIVE)) {
hosts.add(ServerName.valueOf(dataNode.getHostName(), -1, -1));
}
return hosts.toArray(new ServerName[hosts.size()]);
return hosts.toArray(new ServerName[0]);
}
}

8
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartRandomRsAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -33,9 +33,13 @@ public class RestartRandomRsAction extends RestartActionBaseAction {
super(sleepTime);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info("Performing action: Restart random region server");
getLogger().info("Performing action: Restart random region server");
ServerName server = PolicyBasedChaosMonkey.selectRandomItem(getCurrentServers());
restartRs(server, sleepTime);

10
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartRandomRsExceptMetaAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -20,12 +20,20 @@ package org.apache.hadoop.hbase.chaos.actions;
import org.apache.hadoop.hbase.ServerName;
import org.apache.hadoop.hbase.chaos.monkies.PolicyBasedChaosMonkey;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
public class RestartRandomRsExceptMetaAction extends RestartRandomRsAction {
private static final Logger LOG = LoggerFactory.getLogger(RestartRandomRsExceptMetaAction.class);
public RestartRandomRsExceptMetaAction(long sleepTime) {
super(sleepTime);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
int tries = 10;

8
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartRandomZKNodeAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -34,9 +34,13 @@ public class RestartRandomZKNodeAction extends RestartActionBaseAction {
super(sleepTime);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info("Performing action: Restart random zookeeper node");
getLogger().info("Performing action: Restart random zookeeper node");
ServerName server = PolicyBasedChaosMonkey.selectRandomItem(
ZKServerTool.readZKNodes(getConf()));
restartZKNode(server, sleepTime);

11
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartRsHoldingMetaAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -33,12 +33,17 @@ public class RestartRsHoldingMetaAction extends RestartActionBaseAction {
public RestartRsHoldingMetaAction(long sleepTime) {
super(sleepTime);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info("Performing action: Restart region server holding META");
getLogger().info("Performing action: Restart region server holding META");
ServerName server = cluster.getServerHoldingMeta();
if (server == null) {
LOG.warn("No server is holding hbase:meta right now.");
getLogger().warn("No server is holding hbase:meta right now.");
return;
}
ClusterStatus clusterStatus = cluster.getClusterStatus();

10
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RestartRsHoldingTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -43,15 +43,19 @@ public class RestartRsHoldingTableAction extends RestartActionBaseAction {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HTable table = null;
try {
LOG.info("Performing action: Restart random RS holding table " + this.tableName);
getLogger().info("Performing action: Restart random RS holding table " + this.tableName);
Configuration conf = context.getHBaseIntegrationTestingUtility().getConfiguration();
table = new HTable(conf, TableName.valueOf(tableName));
} catch (IOException e) {
LOG.debug("Error creating HTable used to get list of region locations.", e);
getLogger().debug("Error creating HTable used to get list of region locations.", e);
return;
}

25
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RollingBatchRestartRsAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -27,10 +27,10 @@ import java.util.Objects;
import java.util.Queue;
import org.apache.commons.lang.math.RandomUtils;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.hbase.ServerName;
import org.apache.hadoop.hbase.chaos.monkies.PolicyBasedChaosMonkey;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
/**
* Restarts a ratio of the regionservers in a rolling fashion. At each step, either kills a
@ -39,7 +39,7 @@ import org.apache.hadoop.hbase.chaos.monkies.PolicyBasedChaosMonkey;
* can be down at the same time during rolling restarts.
*/
public class RollingBatchRestartRsAction extends BatchRestartRsAction {
private static final Log LOG = LogFactory.getLog(RollingBatchRestartRsAction.class);
private static final Logger LOG = LoggerFactory.getLogger(RollingBatchRestartRsAction.class);
protected int maxDeadServers; // number of maximum dead servers at any given time. Defaults to 5
public RollingBatchRestartRsAction(long sleepTime, float ratio) {
@ -56,9 +56,14 @@ public class RollingBatchRestartRsAction extends BatchRestartRsAction {
START
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info(String.format("Performing action: Rolling batch restarting %d%% of region servers",
getLogger().info(
String.format("Performing action: Rolling batch restarting %d%% of region servers",
(int)(ratio * 100)));
List<ServerName> selectedServers = selectServers();
@ -91,7 +96,7 @@ public class RollingBatchRestartRsAction extends BatchRestartRsAction {
} catch (org.apache.hadoop.util.Shell.ExitCodeException e) {
// We've seen this in test runs where we timeout but the kill went through. HBASE-9743
// So, add to deadServers even if exception so the start gets called.
LOG.info("Problem killing but presume successful; code=" + e.getExitCode(), e);
getLogger().info("Problem killing but presume successful; code=" + e.getExitCode(), e);
}
deadServers.add(server);
break;
@ -105,7 +110,7 @@ public class RollingBatchRestartRsAction extends BatchRestartRsAction {
// The start may fail but better to just keep going though we may lose server.
// Shuffle the dead list to avoid getting stuck on a single stubborn host.
Collections.shuffle(deadServers);
LOG.info(String.format(
getLogger().info(String.format(
"Problem starting %s, will retry; code=%s", server, e.getExitCode(), e));
}
break;
@ -139,7 +144,7 @@ public class RollingBatchRestartRsAction extends BatchRestartRsAction {
@Override
protected void killRs(ServerName server) throws IOException {
LOG.info("Killed " + server);
getLogger().info("Killed " + server);
if (this.invocations++ % 3 == 0) {
throw new org.apache.hadoop.util.Shell.ExitCodeException(-1, "Failed");
}
@ -147,7 +152,7 @@ public class RollingBatchRestartRsAction extends BatchRestartRsAction {
@Override
protected void startRs(ServerName server) throws IOException {
LOG.info("Started " + server);
getLogger().info("Started " + server);
if (this.invocations++ % 3 == 0) {
throw new org.apache.hadoop.util.Shell.ExitCodeException(-1, "Failed");
}
@ -156,4 +161,4 @@ public class RollingBatchRestartRsAction extends BatchRestartRsAction {
action.perform();
}
}
}

16
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/RollingBatchSuspendResumeRsAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -58,9 +58,14 @@ public class RollingBatchSuspendResumeRsAction extends Action {
SUSPEND, RESUME
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info(String.format("Performing action: Rolling batch restarting %d%% of region servers",
getLogger().info(
String.format("Performing action: Rolling batch restarting %d%% of region servers",
(int) (ratio * 100)));
List<ServerName> selectedServers = selectServers();
@ -91,7 +96,8 @@ public class RollingBatchSuspendResumeRsAction extends Action {
try {
suspendRs(server);
} catch (Shell.ExitCodeException e) {
LOG.warn("Problem suspending but presume successful; code=" + e.getExitCode(), e);
getLogger().warn("Problem suspending but presume successful; code="
+ e.getExitCode(), e);
}
suspendedServers.add(server);
break;
@ -100,7 +106,7 @@ public class RollingBatchSuspendResumeRsAction extends Action {
try {
resumeRs(server);
} catch (Shell.ExitCodeException e) {
LOG.info("Problem resuming, will retry; code= " + e.getExitCode(), e);
getLogger().info("Problem resuming, will retry; code= " + e.getExitCode(), e);
}
break;
default:
@ -108,7 +114,7 @@ public class RollingBatchSuspendResumeRsAction extends Action {
"Encountered unexpected action type: " + action.name());
}
LOG.info("Sleeping for: " + sleepTime);
getLogger().info("Sleeping for: " + sleepTime);
Threads.sleep(sleepTime);
}
}

11
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/SnapshotTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -28,8 +28,7 @@ import org.slf4j.LoggerFactory;
* Action that tries to take a snapshot of a table.
*/
public class SnapshotTableAction extends Action {
private static final Logger LOG =
LoggerFactory.getLogger(SnapshotTableAction.class);
private static final Logger LOG = LoggerFactory.getLogger(SnapshotTableAction.class);
private final TableName tableName;
private final long sleepTime;
@ -42,6 +41,10 @@ public class SnapshotTableAction extends Action {
this.sleepTime = sleepTime;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
@ -53,7 +56,7 @@ public class SnapshotTableAction extends Action {
return;
}
LOG.info("Performing action: Snapshot table " + tableName);
getLogger().info("Performing action: Snapshot table {}", tableName);
admin.snapshot(snapshotName, tableName);
if (sleepTime > 0) {
Thread.sleep(sleepTime);

12
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/SplitAllRegionOfTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -19,14 +19,12 @@ package org.apache.hadoop.hbase.chaos.actions;
import java.io.IOException;
import java.util.concurrent.ThreadLocalRandom;
import org.apache.hadoop.hbase.HBaseTestingUtility;
import org.apache.hadoop.hbase.TableName;
import org.apache.hadoop.hbase.client.Admin;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
public class SplitAllRegionOfTableAction extends Action {
private static final Logger LOG =
LoggerFactory.getLogger(SplitAllRegionOfTableAction.class);
@ -47,6 +45,10 @@ public class SplitAllRegionOfTableAction extends Action {
this.maxFullTableSplits = getConf().getInt(MAX_SPLIT_KEY, DEFAULT_MAX_SPLITS);
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
@ -61,10 +63,10 @@ public class SplitAllRegionOfTableAction extends Action {
if (ThreadLocalRandom.current().nextDouble()
< (((double) splits) / ((double) maxFullTableSplits)) / ((double) 2)) {
splits++;
LOG.info("Performing action: Split all regions of " + tableName);
getLogger().info("Performing action: Split all regions of {}", tableName);
admin.split(tableName);
} else {
LOG.info("Skipping split of all regions.");
getLogger().info("Skipping split of all regions.");
}
}
}

14
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/SplitRandomRegionOfTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -46,15 +46,19 @@ public class SplitRandomRegionOfTableAction extends Action {
this.tableName = tableName;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
Admin admin = util.getHBaseAdmin();
LOG.info("Performing action: Split random region of table " + tableName);
getLogger().info("Performing action: Split random region of table " + tableName);
List<HRegionInfo> regions = admin.getTableRegions(tableName);
if (regions == null || regions.isEmpty()) {
LOG.info("Table " + tableName + " doesn't have regions to split");
getLogger().info("Table " + tableName + " doesn't have regions to split");
return;
}
// Don't try the split if we're stopping
@ -64,11 +68,11 @@ public class SplitRandomRegionOfTableAction extends Action {
HRegionInfo region = PolicyBasedChaosMonkey.selectRandomItem(
regions.toArray(new HRegionInfo[regions.size()]));
LOG.debug("Splitting region " + region.getRegionNameAsString());
getLogger().debug("Splitting region " + region.getRegionNameAsString());
try {
admin.splitRegion(region.getRegionName());
} catch (Exception ex) {
LOG.warn("Split failed, might be caused by other chaos: " + ex.getMessage());
getLogger().warn("Split failed, might be caused by other chaos: " + ex.getMessage());
}
if (sleepTime > 0) {
Thread.sleep(sleepTime);

14
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/TruncateTableAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -19,7 +19,6 @@
package org.apache.hadoop.hbase.chaos.actions;
import java.util.Random;
import org.apache.hadoop.hbase.HBaseTestingUtility;
import org.apache.hadoop.hbase.client.Admin;
import org.apache.hadoop.hbase.client.HBaseAdmin;
@ -31,8 +30,7 @@ import org.slf4j.LoggerFactory;
* Action that tries to truncate of a table.
*/
public class TruncateTableAction extends Action {
private static final Logger LOG =
LoggerFactory.getLogger(TruncateTableAction.class);
private static final Logger LOG = LoggerFactory.getLogger(TruncateTableAction.class);
private final TableName tableName;
private final Random random;
@ -41,6 +39,10 @@ public class TruncateTableAction extends Action {
this.random = new Random();
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
HBaseTestingUtility util = context.getHBaseIntegrationTestingUtility();
@ -52,8 +54,8 @@ public class TruncateTableAction extends Action {
}
boolean preserveSplits = random.nextBoolean();
LOG.info("Performing action: Truncate table " + tableName.getNameAsString() +
"preserve splits " + preserveSplits);
getLogger().info("Performing action: Truncate table {} preserve splits {}",
tableName.getNameAsString(), preserveSplits);
admin.truncateTable(tableName, preserveSplits);
}
}

8
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/UnbalanceKillAndRebalanceAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -56,6 +56,10 @@ public class UnbalanceKillAndRebalanceAction extends Action {
this.killMetaRs = killMetaRs;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
ClusterStatus status = this.cluster.getClusterStatus();
@ -86,7 +90,7 @@ public class UnbalanceKillAndRebalanceAction extends Action {
}
if (!killMetaRs && targetServer.equals(metaServer)) {
LOG.info("Not killing server because it holds hbase:meta.");
getLogger().info("Not killing server because it holds hbase:meta.");
} else {
killRs(targetServer);
killedServers.add(targetServer);

8
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/actions/UnbalanceRegionsAction.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -48,9 +48,13 @@ public class UnbalanceRegionsAction extends Action {
this.fractionOfServers = fractionOfServers;
}
@Override protected Logger getLogger() {
return LOG;
}
@Override
public void perform() throws Exception {
LOG.info("Unbalancing regions");
getLogger().info("Unbalancing regions");
ClusterStatus status = this.cluster.getClusterStatus();
List<ServerName> victimServers = new LinkedList<ServerName>(status.getServers());
int targetServerCount = (int)Math.ceil(fractionOfServers * victimServers.size());

4
hbase-it/src/test/java/org/apache/hadoop/hbase/chaos/policies/CompositeSequentialPolicy.java
View File

@ -1,4 +1,4 @@
/**
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
@ -23,7 +23,7 @@ import java.util.List;
/** A policy that runs multiple other policies one after the other */
public class CompositeSequentialPolicy extends Policy {
private List<Policy> policies;
private final List<Policy> policies;
public CompositeSequentialPolicy(Policy... policies) {
this.policies = Arrays.asList(policies);
}