2011/3/8 Peter Schuller <peter.schuller@infidyne.com>
>                 $client->batch_mutate($mutations,
> cassandra_ConsistencyLevel::QUORUM);

Btw, what are the mutations? Are you doing something like inserting
both very small values and very large ones?

I have big xml file (5 GB) (mysql dump in xml format) and read data from it with SAX xml parser, all records on that file looks like this:

        <row>
                <field name="uid">5</field>
                <field name="aid">3619780:1</field>
                <field name="cleanness">0</field>
                <field name="counter">7</field>
                <field name="gcount">0</field>
                <field name="lastchange">1291053619</field>
                <field name="disaster">0</field>
                <field name="tdisaster">0</field>
        </row>


mutations in that case is 10 similar records (follow fragment of code, describes situation )

    $l_supercolumn = new cassandra_SuperColumn(array("name" => $l_row["aid"], "columns" => $l_columns));
    $l_c_or_sc = new cassandra_ColumnOrSuperColumn(array("super_column" => $l_supercolumn));
    $l_mutation = new cassandra_Mutation(array("column_or_supercolumn" => $l_c_or_sc));

    if(array_key_exists($l_key, $mutations))
    {
        array_push($mutations[$l_key]['aquarium_friend'], $l_mutation);
    }
    else
    {
        $mutations[$l_key] = array('aquarium_friend' => array($l_mutation));
    };

    if(!($l_i % 10))
    {
        make_mutation($client, $mutations, $g_loger, $g_rloger);
        $mutations = array();

        if(!($l_i % 1000))
        {
            $g_loger->info(sprintf("inserted: %s", $l_i));
        };
    };



That's why I asked about the frequency. If you're doing a long-term
stress test and seeing a 30 second pause once per week, that's a lot
more likely to be "normal" for your workload than if you're seeing it
happen once ever three minutes. The issue is that if you want to fix
your problem, one must first figure out what the problem *is*. Based
on past mailing list traffic, it seems most people's problems that are
seemingly "due to GC" end up being because of a too high live set size
or the CMS phase triggering too late. These are fixable issues if are
running into them.


In may case this happen from time to time. For example insert all 5GB xml took about 30-40 minutes, and nodes frozen about 5-10 time on that period (avg time of frozen 15 secs)

 
If all you have is a single column family with a 64 mb flush threshold
and doing a bunch of insertions, and have a heap size of 5 (was it?)
gig, you should not be having these issues. But stating that helps no
one, which is why I'm asking for more information. (Widely
extrapolating and suggesting that all Cassandra nodes will always
freeze for 30 seconds every now and then is also helping no one, other
than not being true.)

 At initial state HEAP was 6GB. When i increase HEAP size to 7GB nodes frozen only one time, but took much greater time (40 secs)