Hi guys,
I have a small setup with 3 macs running os x snow leopard and the
native java 1.6.0.something.something ;-)
I am doing a lot of inserts, without any problems and I can read the
data again.
But
At some point, around 1G Commitlog size, cassandra starts a rise in
ram usage, unt
Forgot to say - It will eat up 1.13GB of RAM and 1.32GB VRAM, and
still throw and OutOfMemory, though 700MB left...
On 07/10/2009, at 19.38, Dan Larsen wrote:
Hi guys,
I have a small setup with 3 macs running os x snow leopard and the
native java 1.6.0.something.something ;-)
I am doing a
What version are you running? Take 0.4 from trunk.
Default config has
128
Did you change it allow your commit log chunks be 1Gb?
On Wed, Oct 7, 2009 at 1:38 PM, Dan Larsen wrote:
> Hi guys,
>
> I have a small setup with 3 macs running os x snow leopard and the native
> java 1.6.0.something.som
O.k... I experimented a bit: changed "-Xmx1G \" to "-Xmx2G \" in bin/
cassandra.in.sh ...
Is it really nescessary to have as much RAM as you have data?!?!?
Best regards
Dan
On 07/10/2009, at 19.50, Dan Larsen wrote:
Forgot to say - It will eat up 1.13GB of RAM and 1.32GB VRAM, and
still thro
Sorry about that... running 0.4 taken from trunk.
I didn't change CommitLogRotationThresholdInMB - it's still 128
Thanks!
On 07/10/2009, at 20.12, Igor Katkov wrote:
What version are you running? Take 0.4 from trunk.
Default config has
128
Did you change it allow your commit log chunks be 1G
I had similar OutOfMemory issues, see "commit logs are not deleted" thread,
but it was fixed/commited to trunk.
What I don't get is how your commit log segment grows beyond threshold.
>Is it really nescessary to have as much RAM as you have data?!?!?
No, I was able to insert 50Gb worth of data wi
I don't know why... But for some reason, the commitlogs grew bigger
than 128MB - so I suppose a manual compact was needed...?!?!
Anyways... I restarted with the 2GB value in cassandra.in.sh,
everything got compacted, the commitlog rotated - and I could stop,
change the settings back and start
I will check that thread out! Thanks :-)
Maybe I should try to refresh the code...
I am starting to think, it might be the "native" java of snow leopard,
that is causing some of my problems...
This is what has kept me from coding Java for more than a decade...
Java is never just Java...
I kn
One tip: don't use ConcurrencyLevel.ZERO when bulk inserting. It's
easy to OOM yourself that way.
-Jonathan
Thanks for the tip!
I'm using QUORUM - I hate losing data :-P
/Dan
On 07/10/2009, at 22.56, Jonathan Ellis wrote:
One tip: don't use ConcurrencyLevel.ZERO when bulk inserting. It's
easy to OOM yourself that way.
-Jonathan
10 matches
Mail list logo
