[ https://issues.apache.org/jira/browse/CASSANDRA-9683?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14631408#comment-14631408 ]
Loic Lambiel commented on CASSANDRA-9683: ----------------------------------------- That stats comes from Opscenter. We're using durable_writes = false in the blobstore Keyspace, were most data are written. This may explain the low write latency. I'm going to try to reproduce this on a new single node setup as I don't want to kill this cluster. I'll do it in the coming days, as soon as I have the pipe. We're using Cassandra as a backend for our object storage service based on our pithos (http://pithos.io) API frontend. Data can then be uploaded using any S3 compatible tools like s3cmd. I don't know if you want to go into such setup. (we could help or do it remotely if needed). I don't know either how to reproduce the data pattern and usage without it. > Get mucher higher load and latencies after upgrading from 2.1.6 to cassandra > 2.1.7 > ---------------------------------------------------------------------------------- > > Key: CASSANDRA-9683 > URL: https://issues.apache.org/jira/browse/CASSANDRA-9683 > Project: Cassandra > Issue Type: Bug > Environment: Ubuntu 12.04 (3.13 Kernel) * 3 > JDK: Oracle JDK 7 > RAM: 32GB > Cores 4 (+4 HT) > Reporter: Loic Lambiel > Assignee: Ariel Weisberg > Fix For: 2.1.x > > Attachments: cassandra-env.sh, cassandra.yaml, cfstats.txt, > os_load.png, pending_compactions.png, read_latency.png, schema.txt, > system.log, write_latency.png > > > After upgrading our cassandra staging cluster version from 2.1.6 to 2.1.7, > the average load grows from 0.1-0.3 to 1.8. > Latencies did increase as well. > We see an increase of pending compactions, probably due to CASSANDRA-9592. > This cluster has almost no workload (staging environment) -- This message was sent by Atlassian JIRA (v6.3.4#6332)