There's everything you need there I think, although the lines you pasted are about catalog tables so you might not relate to them as if you saw your own data instead.
So when you look at the first line, you see which table is being edited, the region's encoded name (gives you the location in HDFS), then you have the edit itself which in this case is a Put with the row key .META.,,1, then you can see the family, qualifier, timestamp, and finally the value is there but the tool you used doesn't print it out (makes the output sane). What else would be missing? J-D On Tue, Jun 14, 2011 at 3:12 AM, James Ram <[email protected]> wrote: > Hi, > > I would like to dispel some of my doubts about WriteAheadLogs (WAL). I > believe they are stored under the .logs directory in hdfs. > But it seems that the information contained in the WAL is insufficient for > recovery after a regionserver goes down. > I have given the sample contents of a WAL of a particular regionserver in my > cluster: > > -ROOT-/70236052/1 [#edits: 1 = > <.META.,,1/info:regioninfo/1306420231906/Put/vlen=272; >] > -ROOT-/70236052/2 [#edits: 1 = > <METAROW/METAFAMILY:/1306420232051/Put/vlen=17; >] > > This doesn't seem to give any info on the data in the regionserver. Is it > meant only for the master to use this or can we write our own code to handle > failures? > > > -- > With Regards, > Jr. >
