cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Amir <amir7...@yahoo.com>
Subject Re: Problem with Deletes
Date Sun, 20 Jun 2010 10:21:47 GMT
Benjamin Black <b <at> b3k.us> writes:

> 
> http://wiki.apache.org/cassandra/DistributedDeletes
> 
> On Thu, Jun 17, 2010 at 9:10 AM, Amir <amir74tt <at> yahoo.com> wrote:
> > Hi All,
> >
> > I'm running a benchmark on Cassandra while using a benchmark client which 
I've
> > written myself.
> >
> > I'm running the following scenario:
> > One Cassandra node on the same machine as the client.
> > The client writes a new key every 1 second and deletes it after 10 seconds, 
so
> > at any given time there should be only 10 keys.
> > The key value size is 2K.
> >
> > When I ran this scenario I looked at the data folder and saw that initially
> > cassandra created 4 files (SSTables) of ~130K and then compacted them to 20K
> > (which is exactly what I expected 10 keys * 2K = 20K).
> > And afterwards there where another 3 130K files which were compacted 
together
> > with the first 20K file to a new 20K file, and so on...
> >
> > This scenario ran exactly as I excepted.
> >
> > Now I ran the same scenario but this time with key value size=2M.
> >
> > When I ran this scenario initially cassandra created 4 files (SSTables) of 
~64M
> > and then compacted them to 20M (which is exactly what I expected 10 keys * 
2M =
> > 20M). But after creating another 3 64M files - the problem started!
> > It didn't compact them with the first 20M - instead it created another 64M 
file
> > and compacted all 4 of them to 260M file (!), and after creating another 4 
64M
> > files it compacted them to another 260M file and so on...
> >
> > It looks to me like in this scenario for some reason the compaction doesn't
> > make any deletes. I don't have any idea why 
> >
> > Additional info which I should mention:
> > In the storage-conf.xml the following are not default:
> > GCGraceSeconds = 0
> > MemtableFlushAfterMinutes = 1
> > <ColumnFamily Name="Standard2"
> >                    CompareWith="UTF8Type"
> >                    KeysCached="99%"/>
> >
> >
> > Thanks a lot for your help,
> >
> > Amir
> >
> >
> >
> >


Thanks, but I didn't find an answer to my problem in this article.

BTW, the article says "And of course if you are only running a single Cassandra 
node, you can reduce GCGraceSecondszero to zero, and tombstones will be GC'd at 
the first major compaction".
This is exactly my configuration (1 node and GCGraceSecondszero=0) - But still 
the compaction doesn't make the SSTables smaller. 

They just get bigger and bigger until the disk runs out of space





Mime
View raw message