-Re: Scan exception when running MR
J Mohamed Zahoor 2012-10-23, 06:51
It was a stray hbase jar that caused the problem.
On 22-Oct-2012, at 10:43 PM, Bryan Beaudreault <[EMAIL PROTECTED]> wrote:
> Oh, sorry.
> This sounds like a version mismatch. Do you have the same version
> installed on your servers and being pulled in your hadoop job?
> On Mon, Oct 22, 2012 at 12:32 PM, J Mohamed Zahoor <[EMAIL PROTECTED]> wrote:
>> Cool… But my map reduce doesn't even start…
>> It fails while creating a record reader...
>> The record reader fails in
>> and throws a
>> java.io.IOException: version not supported
>> at org.apache.hadoop.hbase.client.Scan.readFields(Scan.java:558)
>> at org.apache.hadoop.mapred.MapTask.runNewMapper(MapTask.java:723)
>> at org.apache.hadoop.mapred.MapTask.run(MapTask.java:370)
>> at org.apache.hadoop.mapred.Child$4.run(Child.java:255)
>> at java.security.AccessController.doPrivileged(Native Method)
>> at javax.security.auth.Subject.doAs(Subject.java:415)
>> at org.apache.hadoop.mapred.Child.main(Child.java:249)
>> Basically the deserializing of Scan object from the conf is failing for
>> some reason.
>> On 22-Oct-2012, at 9:52 PM, Bryan Beaudreault <[EMAIL PROTECTED]>
>>> I'm not on 0.94.1, but I've found a lot of situations that can cause
>>> scanner timeouts and other scanner exceptions from M/R. The primary ones
>>> probably still apply in later versions:
>>> - Caching or batching set too high. If caching is set to, e.g. 1000,
>>> and hbase.rpc.timeout is set to 30 seconds, it means you need to be
>> able to
>>> be able to process 1000 records in your mapper in under 30 seconds
>>> the overhead of actually returning that many records). Otherwise the
>>> mapper's next call to next() will throw a timeout.
>>> - Similar to the above, this can happen if just the logic in your
>>> is too heavy and taking too long. Just keep in mind that the
>>> hbase.rpc.timeout can be triggered by the time between calls to next()
>>> - hbase.rpc.timeout > hbase.regionserver.lease.period. If this is the
>>> case, the RS will timeout first and kill the scan. Then your mapper
>>> call next() and since the scan no longer exists will throw a scan/lease
>>> - The filters on the mapper scan are causing too many rows to be
>>> skipped, such that not enough rows can be collected to return within
>>> Hope this helps and/or is still accurate for your version.
>>> On Mon, Oct 22, 2012 at 11:51 AM, J Mohamed Zahoor <[EMAIL PROTECTED]>
>>>> I am using 0.94.1
>>>> On 22-Oct-2012, at 9:17 PM, J Mohamed Zahoor <[EMAIL PROTECTED]> wrote:
>>>>> I am facing a scanner exception like this when i run a mr job.
>>>>> Both the input and output are hbase tables (different tables)…
>>>>> This comes sporadically on some mapper and all other mappers runs
>>>>> Even the failed mapper gets passed in the next attempt.
>>>>> Any clue on what might be wrong?
>>>>> at org.apache.hadoop.hbase.client.Scan.<init>(Scan.java:147)