-Re: java.io.IOEcxeption key k1 followed by a smaller key k2
lars hofhansl 2012-09-17, 04:06
It would be good track this down. Any way you can share the tools you use to load the data?
Is it easy for you to reproduce this problem?
It's possible (but not likely) that there is a bug in Hadoop 1.0. You should use 1.0.3.
From: Mohamed Ibrahim <[EMAIL PROTECTED]>
To: [EMAIL PROTECTED]; lars hofhansl <[EMAIL PROTECTED]>; Stack <[EMAIL PROTECTED]>
Sent: Sunday, September 16, 2012 7:13 PM
Subject: Re: java.io.IOEcxeption key k1 followed by a smaller key k2
Hello Lars / Stack,
Thank you for responding.
The date on the files is March 9th 2012. It's been up since then, I restarted hbase and hadoop once. I only have a single node that I'm running my tests on. I'm currently running 0.92.1 on hadoop 1.0 . I hope I'm using the correct mix.
I'm not using any external tools other than hbase java api, and inspecting the data using the shell. I ran my program, the following day I found the stack dump on the console. I checked by scanning the table that had the exception for the smaller key k1, with limit 2 from the shell and no exceptions were thrown. I can also see that the following key is larger than the first one, so nothing is wrong.
I faced the same exception before and it happened more frequently when I did a lot of Puts and Deletes using Htable.batch . Essentially I was updating several inverted indexes each in its own table on the data of rows when they get inserted, and instead of doing single Deletes & Puts on the indexes I used batch. batch improved the performance but threw this exception more. I stopped using batch and now doing single Puts & Deletes. The exception rarely gets thrown, but still sporadically gets thrown.
I will read about the hfile tool, thanks for the pointers. I will also try figure out a set of steps that would repeat the exception so it is more helpful. I will also try 0.94.1 with batch and see if it will happen again, and will let you know and will file a bug if I can repeat it consistently.
On Sun, Sep 16, 2012 at 7:33 PM, lars hofhansl <[EMAIL PROTECTED]> wrote:
Hmm... HBASE-6579 gets rid of that check, because we thought it no longer necessary.
>No do you remember what you did leading up to this?
>Did you write these HFiles with some other tool? Done some bulk import, etc?
> From: Mohamed Ibrahim <[EMAIL PROTECTED]>
>To: [EMAIL PROTECTED]
>Sent: Sunday, September 16, 2012 5:59 AM
>Subject: java.io.IOEcxeption key k1 followed by a smaller key k2
>I am using hbase 0.92.1 on hadoop 1 . I am getting those exceptions, and it
>seems to me that it means that the hbase file is not sorted in order. So
>when the scanner goes through it, it finds a smaller key after its current.
>Is that related to https://issues.apache.org/jira/browse/HBASE-6579 ??
>It looks like upgrading to 0.94.1 (current stable) won't fix the issue. Any
>Here is the stack dump:
> at sun.reflect.GeneratedMethodAccessor25.invoke(Unknown Source)
> at java.lang.reflect.Method.invoke(Method.java:616)