各种原因导致.rollback、.clean、.clean.requested和.clean.inflight文件大小为0，也就是空文件，而在archive和clean时无法处理空文件，就报错上面的异常。
前言
记录写Hudi时的一个异常的解决方法，其实这个异常从去年就发现并找到解决方法了，而且已经提交到社区merge了，PR：[HUDI-2675] Fix the exception 'Not an Avro data file' when archive and clean。之所以现在又要总结这个异常的处理方法是因为:我们生产环境用的 Hudi0.9.0版本，且没有升级，因为升级Hudi版本可能会有不兼容的问题，需要测试，比较费时，所以目前还没有升级版本，而这个PR合入的版本为0.11.0，所以本文主要总结在0.9.0版本如何解决这个问题，当然也适用于0.11.0版本之前的其他有同样问题的版本。
异常信息
archive和clean时都会有这个异常，主要异常信息：
Caused by: java.io.IOException: Not an Avro data file
异常产生原因
各种原因导致.rollback、.clean、.clean.requested和.clean.inflight文件大小为0，也就是空文件，而在archive和clean时无法处理空文件，就报错上面的异常。有一个已知原因，就是HDFS配了满了之后会产生空文件，更多的是PMC也不清楚的未知原因，上面的PR中有体现。
解决方案
这是在不升级Hudi版本的前提下，如果可以升级Hudi版本，直接升级到Hudi最新版即可。
解决方案1
当发生该异常时，由运维人员删除对应的空文件即可，当然这适用于表不多且异常偶发的情况，具体命令放在最后。但是当表比较多时，运维人员处理起来比较麻烦，这就需要第二种解决方案了。
解决方案2
基于Hudi0.9.0源码将文章开头提到的PR合进去，然后install本地仓库或者公司自己的内部仓库中，然后Maven pom依赖中引用自己的仓库地址就可以了。基于0.9.0的代码我已经提交，有需要的可以自行下载，其他版本就需要大家自己合了。
- gitee: https://gitee.com/dongkelun/hudi/tree/0.9.0-fixNotAvro/
- github: https://github.com/dongkelun/hudi/tree/0.9.0-fixNotAvro
Hudi maven install命令:
mvn clean install -DskipTest
验证
直接本地运行测试用例中的testArchiveCompletedRollbackAndClean和testCleanEmptyInstants，这俩测试用例通过了应该就没有问题
方案1具体处理方法
不管是什么原因导致的异常，不管用何种方式，只要确保找到正确的对应的大小为0的空文件删掉即可，一定不要删错
异常信息1
ERROR [Timer-Driven Process Thread-4] o.a.hudi.table.HoodieTimelineArchiveLog Failed to archive commits, .commit file: 20220726050533.rollback
java.io.IOException: Not an Avro data file
at org.apache.avro.file.DataFileReader.openReader(DataFileReader.java:50)
at org.apache.hudi.common.table.timeline.TimelineMetadataUtils.deserializeAvroMetadata(TimelineMetadataUtils.java:178)
at org.apache.hudi.client.utils.MetadataConversionUtils.createMetaWrapper(MetadataConversionUtils.java:103)
at org.apache.hudi.table.HoodieTimelineArchiveLog.convertToAvroRecord(HoodieTimelineArchiveLog.java:341)
at org.apache.hudi.table.HoodieTimelineArchiveLog.archive(HoodieTimelineArchiveLog.java:305)
at org.apache.hudi.table.HoodieTimelineArchiveLog.archiveIfRequired(HoodieTimelineArchiveLog.java:128)
at org.apache.hudi.client.AbstractHoodieWriteClient.postCommit(AbstractHoodieWriteClient.java:439)
at org.apache.hudi.client.HoodieJavaWriteClient.postWrite(HoodieJavaWriteClient.java:187)
at org.apache.hudi.client.HoodieJavaWriteClient.insert(HoodieJavaWriteClient.java:129)
at org.apache.nifi.processors.javaHudi.JavaHudi.write(JavaHudi.java:523)
at org.apache.nifi.processors.javaHudi.JavaHudi.onTrigger(JavaHudi.java:404)
at org.apache.nifi.processor.AbstractProcessor.onTrigger(AbstractProcessor.java:27)
at org.apache.nifi.controller.StandardProcessorNode.onTrigger(StandardProcessorNode.java:1167)
at org.apache.nifi.controller.tasks.ConnectableTask.invoke(ConnectableTask.java:208)
at org.apache.nifi.controller.scheduling.TimerDrivenSchedulingAgent$1.run(TimerDrivenSchedulingAgent.java:117)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
原因
.rollback文件大小为0
解决方法
在表元数据路径下查看异常信息里的文件，确认是否大小为0
hadoop fs -ls hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie/20220726050533.rollback
0 2022-07-26 07:05 hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie/20220726050533.rollback
确认为0后，删掉该文件即可
hadoop fs -rm -r hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie/20220726050533.rollback
注意不要删错，也可将该文件改名避免删错再启动组件验证是否正常，如果还有异常，排查其他.rollback大小为0文件，一起删掉
最好不要用grep的方式删除，避免误删，只有配额不足导致的文件特别多的情况下才建议使用
查找所有符合条件的文件（一般只有一符合条件的文件，目前发现只有配额不足的情况才会有多个）
hadoop fs -ls -R hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie | grep .rollback | grep -v .rollback.inflight | awk '{ if ($5 == 0) print $8 }'删除所有符合条件的文件
hadoop fs -ls -R hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie | grep .rollback | grep -v .rollback.inflight | awk '{ if ($5 == 0) print $8 }' | xargs hadoop fs -rm异常信息2：
ERROR [Timer-Driven Process Thread-4] o.a.hudi.table.HoodieTimelineArchiveLog Failed to archive commits, .commit file: 20220726050533.rollback
java.io.IOException: Not an Avro data file
at org.apache.avro.file.DataFileReader.openReader(DataFileReader.java:50)
at org.apache.hudi.common.table.timeline.TimelineMetadataUtils.deserializeAvroMetadata(TimelineMetadataUtils.java:178)
at org.apache.hudi.client.utils.MetadataConversionUtils.createMetaWrapper(MetadataConversionUtils.java:103)
at org.apache.hudi.table.HoodieTimelineArchiveLog.convertToAvroRecord(HoodieTimelineArchiveLog.java:341)
at org.apache.hudi.table.HoodieTimelineArchiveLog.archive(HoodieTimelineArchiveLog.java:305)
at org.apache.hudi.table.HoodieTimelineArchiveLog.archiveIfRequired(HoodieTimelineArchiveLog.java:128)
at org.apache.hudi.client.AbstractHoodieWriteClient.postCommit(AbstractHoodieWriteClient.java:439)
at org.apache.hudi.client.HoodieJavaWriteClient.postWrite(HoodieJavaWriteClient.java:187)
at org.apache.hudi.client.HoodieJavaWriteClient.insert(HoodieJavaWriteClient.java:129)
at org.apache.nifi.processors.javaHudi.JavaHudi.write(JavaHudi.java:523)
at org.apache.nifi.processors.javaHudi.JavaHudi.onTrigger(JavaHudi.java:404)
at org.apache.nifi.processor.AbstractProcessor.onTrigger(AbstractProcessor.java:27)
at org.apache.nifi.controller.StandardProcessorNode.onTrigger(StandardProcessorNode.java:1167)
at org.apache.nifi.controller.tasks.ConnectableTask.invoke(ConnectableTask.java:208)
at org.apache.nifi.controller.scheduling.TimerDrivenSchedulingAgent$1.run(TimerDrivenSchedulingAgent.java:117)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
异常原因
.clean文件大为0
解决方法
找到对应表元数据路径下.clean文件大小为0 的文件并删除，目前遇到的情况只有一个文件且是最新的.clean文件最好不要用grep的方式删除，避免误删
hadoop fs -ls hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie/20220726050533.rollback
0 2022-07-26 07:05 hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie/20220726050533.rollback
异常信息3：
o.a.h.t.a.clean.BaseCleanActionExecutor Failed to perform previous clean operation, instant: [==>20211011143809__clean__REQUESTED]
org.apache.hudi.exception.HoodieIOException: Not an Avro data file
at org.apache.hudi.table.action.clean.BaseCleanActionExecutor.runPendingClean(BaseCleanActionExecutor.java:87)
at org.apache.hudi.table.action.clean.BaseCleanActionExecutor.lambda$execute$0(BaseCleanActionExecutor.java:137)
at java.util.ArrayList.forEach(ArrayList.java:1257)
at org.apache.hudi.table.action.clean.BaseCleanActionExecutor.execute(BaseCleanActionExecutor.java:134)
at org.apache.hudi.table.HoodieJavaCopyOnWriteTable.clean(HoodieJavaCopyOnWriteTable.java:188)
at org.apache.hudi.client.AbstractHoodieWriteClient.clean(AbstractHoodieWriteClient.java:660)
at org.apache.hudi.client.AbstractHoodieWriteClient.clean(AbstractHoodieWriteClient.java:641)
at org.apache.hudi.client.AbstractHoodieWriteClient.clean(AbstractHoodieWriteClient.java:672)
at org.apache.hudi.client.AbstractHoodieWriteClient.autoCleanOnCommit(AbstractHoodieWriteClient.java:505)
at org.apache.hudi.client.AbstractHoodieWriteClient.postCommit(AbstractHoodieWriteClient.java:440)
at org.apache.hudi.client.HoodieJavaWriteClient.postWrite(HoodieJavaWriteClient.java:187)
at org.apache.hudi.client.HoodieJavaWriteClient.insert(HoodieJavaWriteClient.java:129)
at org.apache.nifi.processors.javaHudi.JavaHudi.write(JavaHudi.java:401)
at org.apache.nifi.processors.javaHudi.JavaHudi.onTrigger(JavaHudi.java:305)
at org.apache.nifi.processor.AbstractProcessor.onTrigger(AbstractProcessor.java:27)
at org.apache.nifi.controller.StandardProcessorNode.onTrigger(StandardProcessorNode.java:1166)
at org.apache.nifi.controller.tasks.ConnectableTask.invoke(ConnectableTask.java:208)
at org.apache.nifi.controller.scheduling.TimerDrivenSchedulingAgent$1.run(TimerDrivenSchedulingAgent.java:117)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
Caused by: java.io.IOException: Not an Avro data file
at org.apache.avro.file.DataFileReader.openReader(DataFileReader.java:50)
at org.apache.hudi.common.table.timeline.TimelineMetadataUtils.deserializeAvroMetadata(TimelineMetadataUtils.java:178)
at org.apache.hudi.common.util.CleanerUtils.getCleanerPlan(CleanerUtils.java:106)
at org.apache.hudi.table.action.clean.BaseCleanActionExecutor.runPendingClean(BaseCleanActionExecutor.java:84)
... 24 common frames omitted
异常原因
. clean.requested 或者 . clean.inflight
解决方法
删除对应的大小为0的文件，文件名异常信息里已经有了
最好不要用grep的方式删除，避免误删，只有配额不足导致的文件特别多的情况下才建议使用
hadoop fs -ls -R hdfs://cluster1/apps/hive/tenant/zxqzk_smzt_mztgx/sam_exp/.hoodie | grep .clean.requested | awk '{ if ($5 == 0) print $8 }' | xargs hadoop fs -rm
