hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jean-Adrien <a...@jeanjean.ch>
Subject Re: Regionserver fails to serve region
Date Tue, 21 Oct 2008 16:50:48 GMT

I made more tests.

Regarding to HADOOP-3831, it is possible to disable the channel timeout
using the property


I tried this, but I couldn't launch hbase anymore:
During the startup phase a lot of accesses are made on the mapfile of the
-ROOT- region, and it is like the sockets are not closed: The regionserver
responsible to serve the -ROOT- region suddenly fails to get the concerned
mapfile blocks because of:

2008-10-21 14:21:09,212 ERROR org.apache.hadoop.dfs.DataNode:
storageID=DS-316339081-, infoPort=50075,
ipcPort=50020):DataXceiver: java.io.IOException: xceiverCount 257 exceeds
the limit of concurrent xcievers 256
        at java.lang.Thread.run(Thread.java:619)

Which correspond to the limit introduced in HADOOP-3633.

Maybe the "Premeture" error caused by the channel timeout comes from a
socket that is not closed by HBase regionserver ?

It is possible to configure this xcievers limit (see HADOOP-3859) but the
name of the parameter is some kind of secret. Anyway, depends the cause of
the high number of concurrent access, maybe it is useless to increase this
parameter. Then I returned to my previous configuration, removing the
dfs.datanode.socket.write.timeout=0 property.

I noticed there was a lot of stuff done at startup time of HBase. I was
thinking that it was only regionservers that open regions, but it seems that
longer the cluster was running more are the files to be processed. Is that
correct ?

Have a nice day.

-- J.-A.

View this message in context: http://www.nabble.com/Regionserver-fails-to-serve-region-tp20028553p20094637.html
Sent from the HBase User mailing list archive at Nabble.com.

View raw message