一. 问题描述

搭建了一个Hadoop的demo环境，用于一些功能测试，使用了一段时间之后发现flink任务提交不到hadoop上了。查看资源也都充足，查看hdfs后发现文件出现丢失和损坏的情况。此文章用于解决hdfs文件的问题。

二. 问题分析与解决

1. HDFS 块损坏

1.1. 问题表述

执行命令：

hdfs fsck /

发现文件存在丢失和损坏的情况

.....
/dodb/datalake/jars/110/e24d18b0014183c95f56a26724353c15.jar:  Under replicated BP-1704786246-10.101.1.140-1663251681207:blk_1073742003_1179. Target Replicas is 3 but found 1 live replica(s), 0 decommissioned replica(s), 0 decommissioning replica(s).
.................................................................................
/flink-savepoint/34c63dd8507daaa028860d984baa6597/chk-142/_metadata: CORRUPT blockpool BP-1704786246-10.101.1.140-1663251681207 block blk_1073742316

/flink-savepoint/34c63dd8507daaa028860d984baa6597/chk-142/_metadata: MISSING 1 blocks of total size 17009 B................
..................
/flink-savepoint/ced56a39eca402a8050c99a5187b995a/chk-151/_metadata: CORRUPT blockpool BP-1704786246-10.101.1.140-1663251681207 block blk_1073742317

/flink-savepoint/ced56a39eca402a8050c99a5187b995a/chk-151/_metadata: MISSING 1 blocks of total size 2439 B.............
/user/commonuser/.flink/application_1663251697076_0001/dataflow-flink.jar:  Under replicated BP-1704786246-10.101.1.140-1663251681207:blk_1073741830_1006. Target Replicas is 3 but found 1 live replica(s), 0 decommissioned replica(s), 0 decommissioning replica(s).
.
.....

1.2. 问题解决

看到是执行flink任务时做的savepoint存在丢失的情况，一般savepoint用于集群迁移等情况。

直接删除文件的情况

根据上述命令找到有问题的块，查看无用之后，执行删除，然后再看hdfs的健康状态。

hdfs dfs -rm  -f -R  /flink-savepoint/34c63dd8507daaa028860d984baa6597

需要文件恢复的情况

参考：
How to fix corrupt HDFS FIles

2. 副本同步问题

2.1. 问题表述

修复完损坏的文件之后，再次查看hadoop的健康状态

hdfs fsck /

Status: HEALTHY
 Total size:    1039836161 B
 Total dirs:    179
 Total files:   188
 Total symlinks:                0
 Total blocks (validated):      188 (avg. block size 5531043 B)
 Minimally replicated blocks:   188 (100.0 %)
 Over-replicated blocks:        0 (0.0 %)
 Under-replicated blocks:       69 (36.70213 %)
 Mis-replicated blocks:         0 (0.0 %)
 Default replication factor:    1
 Average block replication:     1.0
 Corrupt blocks:                0
 Missing replicas:              138 (42.331287 %)
 Number of data-nodes:          1
 Number of racks:               1
FSCK ended at Thu Sep 29 13:53:02 CST 2022 in 7 milliseconds


The filesystem under path '/' is HEALTHY

看到大量的块存在副本问题

 Missing replicas:              138 (42.331287 %)

demo环境是一个单节点，只有一个副本，但是日志却提示有三个副本

/user/commonuser/.flink/application_1664426842010_0004/log4j.properties:  Under replicated BP-1704786246-10.101.1.140-1663251681207:blk_1073752973_12149. 
Target Replicas is 3 but found 1 live replica(s), 0 decommissioned replica(s), 0 decommissioning replica(s).

2.2. 问题解决

首先查看副本数的设置：dfs.replication=1，没有问题，这会保证新建的文件副本数=1。

然后对已存在的文件修改副本数为1

hadoop dfs -setrep -w 1 -R /

再查看HDFS的健康状态，看到HDFS恢复正常

hdfs fsck /


Status: HEALTHY
。。。。
 Corrupt blocks:                0
 Missing replicas:              0
。。。。
FSCK ended at Fri Sep 30 10:49:05 CST 2022 in 5 milliseconds


The filesystem under path '/' is HEALTHY