[
https://issues.apache.org/jira/browse/HDFS-17590?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18063976#comment-18063976
]
ASF GitHub Bot commented on HDFS-17590:
---------------------------------------
balodesecurity opened a new pull request, #8328:
URL: https://github.com/apache/hadoop/pull/8328
## Summary
`DFSStripedInputStream.createBlockReader()` initialises `dnInfo` to a
sentinel `DNAddrPair(null, null, null, null)` before entering the retry loop.
If `refreshLocatedBlock()` throws an `IOException` (e.g. the block's start
offset is out of range after a file truncation or stale cache) _before_
`getBestNodeDNAddrPair()` is called, the catch block tries:
```java
addToLocalDeadNodes(dnInfo.info); // dnInfo.info is still null
```
`addToLocalDeadNodes` calls `deadNodes.put(null, null)`, and
`ConcurrentHashMap` does not permit null keys, so a `NullPointerException` is
thrown — masking the original `IOException`.
**Fix:** add a null guard at the top of
`DFSInputStream.addToLocalDeadNodes()`:
```java
protected void addToLocalDeadNodes(DatanodeInfo dnInfo) {
if (dnInfo == null) {
return;
}
...
}
```
This is the safest fix location because it protects all callers, not just
the one in `createBlockReader`.
## Test plan
- [x] New test
`TestDFSStripedInputStream#testAddNullToLocalDeadNodesIsIgnored` creates a
striped file, opens a `DFSStripedInputStream`, calls
`addToLocalDeadNodes(null)`, and asserts: no exception thrown and dead-nodes
map remains empty.
- [x] Test passes with MiniDFSCluster (EC RS-6-3 policy).
> `NullPointerException` triggered in `createBlockReader` during retry iteration
> ------------------------------------------------------------------------------
>
> Key: HDFS-17590
> URL: https://issues.apache.org/jira/browse/HDFS-17590
> Project: Hadoop HDFS
> Issue Type: Bug
> Affects Versions: 3.4.0
> Reporter: Elmer J Fudd
> Priority: Critical
>
> While reading blocks of data using `DFSInputStream` in `createBlockReader`, a
> `IOException` originating from `getBlockAt()` that triggers a retry iteration
> results in a `NullPointerException` when passing `dnInfo` to
> `addToLocalDeadNodes` in the catch block.
> This is the relevant callstack portion from our logs (from 3.4.0, but we
> still experience this with versions going into 3.4.1 as recently as late
> June):
> {noformat}
> ...
> java.lang.NullPointerException
> at
> java.base/java.util.concurrent.ConcurrentHashMap.putVal(ConcurrentHashMap.java:1011)
>
> at java.base/
> java.util.concurrent.ConcurrentHashMap.put(ConcurrentHashMap.java:1006)
> at
> org.apache.hadoop.hdfs.DFSInputStream.addToLocalDeadNodes(DFSInputStream.java:184)
>
> at
> org.apache.hadoop.hdfs.DFSStripedInputStream.createBlockReader(DFSStripedInputStream.java:279)
>
> at org.apache.hadoop.hdfs.StripeReader.readChunk(StripeReader.java:304)
> at org.apache.hadoop.hdfs.StripeReader.readStripe(StripeReader.java:335)
> at
> org.apache.hadoop.hdfs.DFSStripedInputStream.fetchBlockByteRange(DFSStripedInputStream.java:504)
>
> at org.apache.hadoop.hdfs.DFSInputStream.pread(DFSInputStream.java:1472)
> at org.apache.hadoop.hdfs.DFSInputStream.read(DFSInputStream.java:1436)
> at org.apache.hadoop.fs.FSInputStream.readFully(FSInputStream.java:124)
> at
> org.apache.hadoop.fs.FSDataInputStream.readFully(FSDataInputStream.java:119)
> ...{noformat}
> What we observe is that `getBlockAt()` thorws an `IOException`
> [here|https://github.com/apache/hadoop/blob/trunk/hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSInputStream.java#L478]
> {code:java}
> //check offset
> if (offset < 0 || offset >= getFileLength()) {
> throw new IOException("offset < 0 || offset >= getFileLength(), offset="
> + offset
> + ", locatedBlocks=" + locatedBlocks);
> }
> {code}
> This is eventually caught in `createBlockReader`. The catch block attempts to
> handle the error and, as part of the error handling, invokes the
> `addToLocalDeadNodes` method. The `dnInfo` object passed to this method is
> actually `NULL` as it wasn't fully allocated
> ([here)|https://github.com/apache/hadoop/blob/4525c7e35ea22d7a6350b8af10eb8d2ff68376e7/hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSStripedInputStream.java#L247],
> which results in a `NullPointerException`.
> To sum up, this is the failure path according to the logs:
> # `IOException` is thrown in `getBlockAt`
> ([code|https://github.com/apache/hadoop/blob/trunk/hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSInputStream.java#L479])
> # The exception propagates to `getBlockGroupAt`
> ([code|https://github.com/apache/hadoop/blob/4525c7e35ea22d7a6350b8af10eb8d2ff68376e7/hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSStripedInputStream.java#L476])
> # It further propagates to `refreshLocatedBlock`
> ([code)|https://github.com/apache/hadoop/blob/4525c7e35ea22d7a6350b8af10eb8d2ff68376e7/hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSStripedInputStream.java#L459]
> # `IOException` caught in `createBlockReader`
> ([code|https://github.com/apache/hadoop/blob/4525c7e35ea22d7a6350b8af10eb8d2ff68376e7/hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSStripedInputStream.java#L247])
> # Error handling in the catch block of `createBlockReader` invokes
> `addToLocalDeadNodes`
> ([code|https://github.com/apache/hadoop/blob/4525c7e35ea22d7a6350b8af10eb8d2ff68376e7/hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSStripedInputStream.java#L281])
> # Execution throws `NullPointerException` since `dnInfo` is NULL
> A simple fix as a `NULL` check before calling `addToLocalDeadNodes`, and
> similarly adjusting the log messages in the `catch` block where `dnInfo` is
> dereferenced, should solve the issue.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
