-Re: Failed recovery after master/tservers IP chages
John Vines 2012-09-23, 17:04
No Jim, it's a thrift transport exception and the dfsclient doesn't use
thrift. Dfs is fairly well designed to avoid any sort of host identity
needed for persistence.
Sent from my phone, so pardon thetypos and brevity.
On Sep 23, 2012 1:00 PM, "Jim Klucar" <[EMAIL PROTECTED]> wrote:
> Hadoop has some weird DNS/Reverse DNS lookup requirements. My guess
> would be that Hadoop is bonking.
> Sent from my iPhone
> On Sep 23, 2012, at 11:07 AM, Kristopher Kane <[EMAIL PROTECTED]>
> > I left some parts out:
> > This is 1.4 and the GC process for fille collection has been running
> > since the cluster turned on. So, does that mean things are being held
> > up in the WA logs?
> > -KRis
> > On Sun, Sep 23, 2012 at 10:50 AM, Kristopher Kane <[EMAIL PROTECTED]>
> >> All,
> >> I was doing some shuffling around at home and changed IPs on my master
> >> and all tservers. I thought this would be OK as I had configured
> >> everything via hostnames but I've got some log entries that say
> >> otherwise:
> >> Unable to recover
> >> org.apache.thrift.transport.TTransportException:
> >> java.net.ConnectException: Connection timed out)
> >> java.io.IOException:
> >> java.net.ConnectException: Connection timed out
> >> This is the master reporting on two tservers of the four I have. Also
> >> to note: I did have an unclean shutdown prior to the IP changes and
> >> the monitor shows no tablets loaded for any table with the recovery
> >> directory in HDFS empty.
> >> I don't need the data and I can always init, but I was curious about
> >> fixing this to learn more about the system.
> >> Where is a good place to start?
> >> Thanks,
> >> -Kris Kane