Hadoopメーリングリスト

ここ１週間ほど見れなかったんだけど，760スレッド程度(gmail的な数え方)になってしまった．orz...きついぜ．

そういえば，primaryサーバが落ちたなぁ．落ちるとそれで終了名ところがつらい．とおもったら，

This patch implements a separate process called the Watcher. It currently monitors only the Namenode. If the NameNode dies, then the watcher restarts it. A testcase to test this code is being written.

One question is whether this watcher process is helpful to real deployments. In can so happen, that a cluster administrator might actually want to watch a variety of services, Hadoop being only one of them. In that case, the sys admin probably would like to integrate the hadoop servers into that framework. Most OS's have software that can watch over a set of services.

ということで，ウォッチャープロセスによって落ちたら自動的に再起動するらしいです．便利そうだ．