My mapreduce.map.java.opts is 1024MB Thanks, Manu
On Thu, Oct 24, 2013 at 3:11 PM, Tsuyoshi OZAWA <ozawa.tsuyo...@gmail.com>wrote: > Hi, > > How about checking the value of mapreduce.map.java.opts? Are your JVMs > launched with assumed heap memory? > > On Thu, Oct 24, 2013 at 11:31 AM, Manu Zhang <owenzhang1...@gmail.com> > wrote: > > Just confirmed the problem still existed even the "mapred-site.xml"s on > all > > nodes have the same configuration (mapreduce.map.memory.mb = 2560). > > > > Any more thoughts ? > > > > Thanks, > > Manu > > > > > > On Thu, Oct 24, 2013 at 8:59 AM, Manu Zhang <owenzhang1...@gmail.com> > wrote: > >> > >> Thanks Ravi. > >> > >> I do have mapred-site.xml under /etc/hadoop/conf/ on those nodes but it > >> sounds weird to me should they read configuration from those > mapred-site.xml > >> since it's the client who applies for the resource. I have another > >> mapred-site.xml in the directory where I run my job. I suppose my job > should > >> read conf from that mapred-site.xml. Please correct me if I am mistaken. > >> > >> Also, not always the same nodes. The number of failures is random, too. > >> > >> Anyway, I will have my settings in all the nodes' mapred-site.xml and > see > >> if the problem goes away. > >> > >> Manu > >> > >> > >> On Thu, Oct 24, 2013 at 1:40 AM, Ravi Prakash <ravi...@ymail.com> > wrote: > >>> > >>> Manu! > >>> > >>> This should not be the case. All tasks should have the configuration > >>> values you specified propagated to them. Are you sure your setup is > correct? > >>> Are they always the same nodes which run with 1024Mb? Perhaps you have > >>> mapred-site.xml on those nodes? > >>> > >>> HTH > >>> Ravi > >>> > >>> > >>> On Tuesday, October 22, 2013 9:09 PM, Manu Zhang > >>> <owenzhang1...@gmail.com> wrote: > >>> Hi, > >>> > >>> I've been running Terasort on Hadoop-2.0.4. > >>> > >>> Every time there is s a small number of Map failures (like 4 or 5) > >>> because of container's running beyond virtual memory limit. > >>> > >>> I've set mapreduce.map.memory.mb to a safe value (like 2560MB) so most > >>> TaskAttempt goes fine while the values of those failed maps are the > default > >>> 1024MB. > >>> > >>> My question is thus, why a small number of container's memory values > are > >>> set to default rather than that of user-configured ? > >>> > >>> Any thoughts ? > >>> > >>> Thanks, > >>> Manu Zhang > >>> > >>> > >>> > >> > > > > > > -- > - Tsuyoshi >