You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "stack (JIRA)" <ji...@apache.org> on 2018/01/11 22:05:00 UTC
[jira] [Commented] (HBASE-19769) IllegalAccessError on
package-private Hadoop metrics2 classes in MapReduce jobs
[ https://issues.apache.org/jira/browse/HBASE-19769?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16323051#comment-16323051 ]
stack commented on HBASE-19769:
-------------------------------
We don't want to use hadoops metrics long-term if thats of any use to. Its only been a pain. Modules for hadoop1 vs hadoop2. We have hacks in place to kill it and restart it because no other way of onlining new metrics -- e.g. region added/removed. We spend most of our CPU counting so we should probably own the metrics system rather than defer to another.
#1 sounds easy-to-do? Nice to have but an aspect of general client metrics and user is not completely lost in that they can look at zk ensemble metrics. #1 has never shipped in any hbase?
> IllegalAccessError on package-private Hadoop metrics2 classes in MapReduce jobs
> -------------------------------------------------------------------------------
>
> Key: HBASE-19769
> URL: https://issues.apache.org/jira/browse/HBASE-19769
> Project: HBase
> Issue Type: Bug
> Components: mapreduce, metrics
> Affects Versions: 2.0.0-beta-1
> Reporter: Josh Elser
> Assignee: Josh Elser
> Priority: Critical
> Fix For: 2.0.0-beta-2
>
>
> issues for context: HBASE-17170, HBASE-17448, TEZ-3299, HADOOP-10893
> Since Hadoop 2.6.0, the {{yarn jar}} entry point to submit a YARN job has been using a custom classloader to separate Hadoop dependencies from the user's JAR being run. A separate classloader is created for the user-provided jar, and then this classloader is set as the contextClassLoader before the Tool is executed by Hadoop's RunJar class. This has been (mostly?) fine for us to date because we don't try to access any Hadoop internal classes client-side.
> However, with the ZK metrics, clients are pushing ZK metrics to metrics2. The problem is that Hadoop metrics2 implementations which we reference from the same package are loaded by a different classloader than our HBase code is loaded from. This makes the expected package-private access of these Metrics2 classes (e.g. MetricsInfoImpl) fail with an IllegalAccessError.
> {noformat}
> java.lang.RuntimeException: Could not create interface org.apache.hadoop.hbase.zookeeper.MetricsZooKeeperSource Is the hadoop compatibility jar on the classpath?
> at org.apache.hadoop.hbase.CompatibilitySingletonFactory.getInstance(CompatibilitySingletonFactory.java:75)
> at org.apache.hadoop.hbase.zookeeper.ZKMetrics.<init>(ZKMetrics.java:36)
> at org.apache.hadoop.hbase.zookeeper.RecoverableZooKeeper.<init>(RecoverableZooKeeper.java:115)
> at org.apache.hadoop.hbase.zookeeper.ZKUtil.connect(ZKUtil.java:139)
> at org.apache.hadoop.hbase.zookeeper.ZKWatcher.<init>(ZKWatcher.java:128)
> at org.apache.hadoop.hbase.zookeeper.ZKWatcher.<init>(ZKWatcher.java:102)
> at org.apache.hadoop.hbase.security.token.TokenUtil.getAuthToken(TokenUtil.java:293)
> at org.apache.hadoop.hbase.security.token.TokenUtil.addTokenForJob(TokenUtil.java:259)
> at org.apache.hadoop.hbase.mapreduce.TableMapReduceUtil.initCredentials(TableMapReduceUtil.java:535)
> at org.apache.phoenix.mapreduce.MultiHfileOutputFormat.configureIncrementalLoad(MultiHfileOutputFormat.java:712)
> at org.apache.phoenix.mapreduce.AbstractBulkLoadTool.submitJob(AbstractBulkLoadTool.java:300)
> at org.apache.phoenix.mapreduce.AbstractBulkLoadTool.loadData(AbstractBulkLoadTool.java:267)
> at org.apache.phoenix.mapreduce.AbstractBulkLoadTool.run(AbstractBulkLoadTool.java:180)
> at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:76)
> at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:90)
> at org.apache.phoenix.mapreduce.CsvBulkLoadTool.main(CsvBulkLoadTool.java:109)
> at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
> at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
> at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
> at java.lang.reflect.Method.invoke(Method.java:498)
> at org.apache.hadoop.util.RunJar.run(RunJar.java:239)
> at org.apache.hadoop.util.RunJar.main(RunJar.java:153)
> Caused by: java.util.ServiceConfigurationError: org.apache.hadoop.hbase.zookeeper.MetricsZooKeeperSource: Provider org.apache.hadoop.hbase.zookeeper.MetricsZooKeeperSourceImpl could not be instantiated
> at java.util.ServiceLoader.fail(ServiceLoader.java:232)
> at java.util.ServiceLoader.access$100(ServiceLoader.java:185)
> at java.util.ServiceLoader$LazyIterator.nextService(ServiceLoader.java:384)
> at java.util.ServiceLoader$LazyIterator.next(ServiceLoader.java:404)
> at java.util.ServiceLoader$1.next(ServiceLoader.java:480)
> at org.apache.hadoop.hbase.CompatibilitySingletonFactory.getInstance(CompatibilitySingletonFactory.java:59)
> ... 21 more
> Caused by: java.lang.IllegalAccessError: tried to access class org.apache.hadoop.metrics2.lib.MetricsInfoImpl from class org.apache.hadoop.metrics2.lib.DynamicMetricsRegistry
> at org.apache.hadoop.metrics2.lib.DynamicMetricsRegistry.newGauge(DynamicMetricsRegistry.java:139)
> at org.apache.hadoop.hbase.zookeeper.MetricsZooKeeperSourceImpl.<init>(MetricsZooKeeperSourceImpl.java:59)
> at org.apache.hadoop.hbase.zookeeper.MetricsZooKeeperSourceImpl.<init>(MetricsZooKeeperSourceImpl.java:51)
> at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
> at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62)
> at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)
> at java.lang.reflect.Constructor.newInstance(Constructor.java:423)
> at java.lang.Class.newInstance(Class.java:442)
> at java.util.ServiceLoader$LazyIterator.nextService(ServiceLoader.java:380)
> ... 24 more
> {noformat}
> There are two ways to fix this problem that I see now:
> # Remove client-side ZK metrics and disallow any more client-side metrics that directly use metrics2 API
> # Use public API for metrics2 only in HBase, or copy implementations from Hadoop to avoid the need to access the package-private classes from this goofy state.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)