Ting Dai created MAPREDUCE-6990:
-----------------------------------
Summary: FileInputStream.skip function can return 0 when the file
is corrupted, causing an infinite loop
Key: MAPREDUCE-6990
URL: https://issues.apache.org/jira/browse/MAPREDUCE-6990
Project: Hadoop Map/Reduce
Issue Type: Bug
Affects Versions: 0.23.0
Reporter: Ting Dai
When file is corrupted, for example, bad encoding,
[Yarn-2724](https://issues.apache.org/jira/browse/YARN-2724), the
FileInputStream can return 0, causing the while loop in TaskLog$Reader become
infinite.
{code:java}
// Some comments here
public Reader(TaskAttemptID taskid, LogName kind,
long start, long end, boolean isCleanup) throws IOException {
// find the right log file
LogFileDetail fileDetail = getLogFileDetail(taskid, kind, isCleanup);
// calculate the start and stop
long size = fileDetail.length;
if (start < 0) {
start += size + 1;
}
if (end < 0) {
end += size + 1;
}
start = Math.max(0, Math.min(start, size));
end = Math.max(0, Math.min(end, size));
start += fileDetail.start;
end += fileDetail.start;
bytesRemaining = end - start;
String owner = obtainLogDirOwner(taskid);
file = SecureIOUtils.openForRead(new File(fileDetail.location,
kind.toString()),
owner, null);
// skip upto start
long pos = 0;
while (pos < start) {
long result = file.skip(start - pos);
if (result < 0) {
bytesRemaining = 0;
break;
}
pos += result;
}
}
{code}
Similar bugs are
[Hadoop-8614](https://issues.apache.org/jira/browse/HADOOP-8614),
[Yarn-2905](https://issues.apache.org/jira/browse/YARN-2905)
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: mapreduce-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: mapreduce-dev-h...@hadoop.apache.org
