> It started with Cassandra 1.1.10 and on Friday we've migrated to 1.2.6 and > now we have the issue that flushes are no longer executed automatically - we > have to do them every hour by hand (nodetool flush) in order to keep the > nodes on an acceptable utilizatzion level. Without manual flush the CPU goes > mad after a couple of hours on each instance. It may be that because 1.2 manages it's memory better it has to flush to disk less frequently in your use case. We flush to disk based on memory usage and log file growth.
Can you explain what happens when the CPU goes mad ? > Is there a way to find out which tasks are pending? Is there a way to cancel > them? This looks like an error in the way the count is being handled, if there are pending tasks they would be running. There was a recent issue like this that was fixed in 1.2.5 https://issues.apache.org/jira/browse/CASSANDRA-5554?attachmentOrder=asc If this is happening reliability can you create a new ticket and include the version and the compaction strategy. Cheers ----------------- Aaron Morton Freelance Cassandra Consultant New Zealand @aaronmorton http://www.thelastpickle.com On 8/07/2013, at 9:21 PM, Steffen Rusitschka <[email protected]> wrote: > The heap size is 3GB and the GC has to run only every 30 minutes... I'm > pretty sure increasing heap won't help. > > Am 08.07.2013 11:15, schrieb Radim Kolar: >> >>> Without manual flush the CPU goes mad after a couple of hours on each >>> instance. >> increase heap size > > > -- > Steffen Rusitschka > CTO > MegaZebra GmbH > Steinsdorfstraße 2 > 81538 München > Phone +49 89 80929577 > > > [email protected] > Challenge me at www.megazebra.com > > MegaZebra GmbH > Geschäftsführer: Henning Kosmack, Christian Meister, Steffen Rusitschka > Sitz der Gesellschaft: München, HRB 177947
