Errata

max_val should be initialized to the smallest negative integer possible instead of 0.

You could have the unlikely case that a year consists entirely of negative temperatures. If this happened, the code as currently written would then erroneously return 0 as the maximum temperature instead of the largest negative temperature from the actual data set.

This is also true for the Python code in example 2-10 on page 36.

"In this case, the keys are the weather station identifiers" should read "In this case, the keys are the years."

The key-value pair output from the map on page 33 is the year and temperature, not the weather station identifier and temperature.

The discussion in that and the previous page is about the PositionedReadable interface. Here's the confusing snippet:

All of these methods preserve the current offset in the file and are thread-safe.. In fact, they are just implemented using the
Seekable interface using the following pattern:

long oldPos = getPos();
try {
seek(position);
// read data
} finally {
seek(oldPos);
}

**

Now, clearly, that is _not_ a thread-safe pattern. On checking the actual source code, I found that the pattern is in fact wrapped in a synchronized block (as it should be).

So the implementation is thread safe, but not exactly concurrent. And since I/O is many orders of magnitude slower than most any other operation, the window in which this mutex is held is indeed quite long.

It would be better to clear this all up, by either (i) omitting the implementation details altogether, (ii) explicitly wrapping the code snippet in a synchronized block, and/or (iii) noting that while the operation is thread safe, it's not designed for concurrent access.

(By contrast, if this were implement via an nio FileChannel then it would be both thread safe _and_ concurrent--something many a reader knows.)

Note from the Author or Editor:
Thanks for your analysis - I agree entirely. For the third edition I've opted to omit the implementation details (i) and mention that a single instance of FSDataInputStream is not designed for concurrent access (iii), and it's better to create multiple instances.

"if set in the tasktracker?s mapred-site.html file" should be "if set in the tasktracker?s mapred-site.xml file"

Note from the Author or Editor:
This error is on page 136 in the print version second edition.

"Combined with SequenceFile.Reader's appendRaw() method" should be "nextRaw()".

Note from the Author or Editor:
The replacement should be: "Combined with a process that creates sequence files with SequenceFile.Writer?s appendRaw() method or SequenceFileAsBinaryOutputFormat"

In section "Audit Logging", there are two appearances of "org.apache.hadoop.fs.FSNamesystem.audit". They should be "org.apache.hadoop.hdfs.server.namenode.FSNamesystem.audit". http://wiki.apache.org/hadoop/HowToConfigure has the same mistake. I have updated it.

Note from the Author or Editor:
Change "org.apache.hadoop.fs.FSNamesystem.audit" to "org.apache.hadoop.hdfs.server.namenode.FSNamesystem.audit" (I could only see one occurrence on the page.)

The third sentence of the second paragraph says, "B's second field is the third field of A ($1) with one added to it." The "$1" should be changed to a "$2".

> % ls /user/hive/warehouse/record/

should be

> % ls /user/hive/warehouse/records/

miss "s"