hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sebastian Bauer <ad...@ugame.net.pl>
Subject Re: hprof from RS which encountered OOME
Date Tue, 17 Aug 2010 11:38:40 GMT
 Hi, with new trunk i get OOME when inserting even 100 rows in
BatchMutations(thrift) to empty table. One row is about less than 256B long
hprof and logs are there: http://e-back.pl/hbase/ . I  know java a
little, but hprof and jhat are black magic for me :)


W dniu 12.08.2010 20:33, Ted Yu pisze:
> There were 20 handlers running.
> I have seen the fix for HBASE-900
> On Thu, Aug 12, 2010 at 11:23 AM, Stack <stack@duboce.net> wrote:
>> Handler is only carrying what the client gave it.  It holds it just
>> long enough to pass it into the server (or out).  So, can you see what
>> the data is?  Is it a big cell?  How many handlers do you have
>> running?  If 100 handlers and each is carrying a 50MB cell then thats
>> 5G.  Is that whats going on?  There was also an old issue where rpc
>> would hold the ghost of the largest byte array it had ever carried but
>> that should be fixed in your  hbase Ted.
>> St.Ack
>> On Thu, Aug 12, 2010 at 11:09 AM, Ted Yu <yuzhihong@gmail.com> wrote:
>>> From the snapshot, we can see that HBaseServer$Handler's had the highest
>>> retained size.
>>> Can we make HBaseServer$Handler aware of the memory pressure of the
>>> RegionServer so that OOME can be avoided ?
>>> Cheers
>>> On Wed, Aug 11, 2010 at 1:19 PM, Ted Yu <tyu@carrieriq.com> wrote:
>>>> See if this is more helpful:

View raw message