错误1：Too many fetch-failures
Reduce task 启动后第一个阶段是 shuffle ，即向 map 端 fetch 数据。每次 fetch 都可能因为 connect 超时， read 超时， checksum 错误等原因而失败。 Reduce task 为每个 map 设置了一个计数器，用以记录 fetch 该 map 输出时失败的次数。当失败次数达到一定阈值时，会通知 JobTracker fetch 该 map 输出操作失败次数太多了，并打印如下 log ：
Failed to fetch map-output from attempt_201105261254_102769_m_001802_0 even after MAX_FETCH_RETRIES_PER_MAP retries... reporting to the JobTracker
其中阈值计算方式为：
max ( MIN_FETCH_RETRIES_PER_MAP ,
getClosestPowerOf2 (( this . maxBackoff * 1000 / BACKOFF_INIT ) + 1));
默认情况下 MIN_FETCH_RETRIES_PER_MAP=2 maxBackoff=300 BACKOFF_INIT=4000 ， 因此默认阈值为 6 ，可通过修改 mapred.reduce.copy.backoff 参数来调整。
当达到阈值后， Reduce task 通过 umbilical 协议告诉 TaskTracker ， TaskTracker 在下一次 heartbeat 时，通知 JobTracker 。当 JobTracker 发现超过 50% 的 Reduce 汇报 fetch 某个 map 的输出多次失败后， JobTracker 会 failed 掉该 map 并重新调度，打印如下 log ：
"Too many fetch-failures for output of task: attempt_201105261254_102769_m_001802_0 ... killing it"
错误2：Task attempt failed to report status for 622 seconds. Killing
The description for mapred.task.timeout which defaults to 600s says "The number of milliseconds before a task will be terminated if it neither reads an input, writes an output, nor updates its status string. "
Increasing the value of mapred.task.timeout might solve the problem, but you need to figure out if more than 600s is actually required for the map task to complete processing the input data or if there is a bug in the code which needs to be debugged.
According to the Hadoop best practices, on average a map task should take a minute or so to process an InputSplit.
错误3：Hadoop: Blacklisted tasktracker
Put following config in conf/hdfs-site.xml:
<property>
<name>dfs.hosts</name>
<value>/full/path/to/whitelisted/node/file</value>
</property>
Use following command to ask Hadoop to refresh node status to based on configuration.
./bin/hadoop dfsadmin -refreshNodeshttp://serverfault.com/questions/288440/hadoop-blacklisted-tasktracker