Hrm, looking at the code, it appears that the !METADATA table has some inconsistencies. A (at least?) one row lacks of the "prev-row" column. Something like:

some_row ~tab:~pr

Sadly, I'm not positive about the cause given the information. Can you scan all of the !METADATA table without error?

That said, I don't think that the above would prevent you from being able to change the iterators on a table because that's stored purely in ZooKeeper. Maybe it tries to read some data from !METADATA before doing the ZooKeeper update?

On 2/26/14, 11:49 PM, Russ Weeks wrote:
Hi, Josh,

Thanks for responding so quickly... sorry, I should've mentioned that
I'm running Accumulo 1.5.0.

Nothing really helpful that I can see in the monitor or debug monitor
logs. It just interleaves the 2 errors above from the
state.MetaDataTableScanner and util.TabletIterator loggers.

-Russ


On Wed, Feb 26, 2014 at 8:37 PM, Josh Elser <[email protected]
<mailto:[email protected]>> wrote:

    Hi Russ,

    What version of Accumulo are you running?

    Any other exceptions in the Monitor?

    - Josh


    On 2/26/14, 11:10 PM, Russ Weeks wrote:

        Hi, All,

        I can't delete any tables through the accumulo shell, I keep
        getting the
        message: "Internal error processing waitForTableOperation".

        Recently my accumulo cluster has started showing this error
        repeatedly
        in the master log:
        [state.MetaDataTableScanner] WARN : No prev-row for key extent:
        {z< : []
        9223372036854775807
        false=  loc243f8e8ad6f5875>. 10.128.1.17:9997
        <http://10.128.1.17:9997> ~tab ~pr-�  �}

        And this error in the monitor log:
        [util.TabletIterator] WARN : gc:agile01 Metadata problem :
        tablet !0;!0<
        has no prev end row

        And this in the GC log:
        [util.TabletIterator] WARN : Metadata problem : tablet !0;!0< has no
        prev end row
        [util.TabletIterator] INFO : Resetting !METADATA scanner to [!0 : []
        9223372036854775807 false,~ : [] 9223372036854775807 false)

        Maybe coincidence, but it seems like things started getting
        weird when I
        added a WholeRowIterator through the accumulo shell (which is
        nonsense,
        I know). The iterators on one of my tables (36M entries) look
        like this:

        -    Iterator WholeRowIterator, scan scope options:
        -        iteratorPriority = 50
        -        iteratorClassName =
        org.apache.accumulo.core.__iterators.user.__WholeRowIterator
        -    Iterator rowdel, majc scope options:
        -        iteratorPriority = 30
        -        iteratorClassName =
        org.apache.accumulo.core.__iterators.user.__RowDeletingIterator
        -    Iterator transformer, majc scope options:
        -        iteratorPriority = 15
        -        iteratorClassName =
        com.phemi.agile.datapolicy.__DataColumnTransformer
        -        maxBufferSize = 256M
        -    Iterator vers, majc scope options:
        -        iteratorPriority = 20
        -        iteratorClassName =
        org.apache.accumulo.core.__iterators.user.__VersioningIterator
        -        maxVersions = 1
        -    Iterator vers, minc scope options:
        -        iteratorPriority = 20
        -        iteratorClassName =
        org.apache.accumulo.core.__iterators.user.__VersioningIterator
        -        maxVersions = 1
        -    Iterator vers, scan scope options:
        -        iteratorPriority = 20
        -        iteratorClassName =
        org.apache.accumulo.core.__iterators.user.__VersioningIterator
        -        maxVersions = 1

        I tried deleting DataColumnTransformer, RowDeletingIterator and
        WholeRowIterator. No errors were returned from the deleteiter
        command in
        the shell but after running listiter, the WholeRowIterator and
        RowDeletingIterator are still there.

        I *am* still able to create tables, but tables that I create
        can't then
        be deleted.

        Any advice appreciated,
        -Russ


Reply via email to