Yan Cui created CASSANDRA-11459: ----------------------------------- Summary: cassandra performance problem when streaming large data Key: CASSANDRA-11459 URL: https://issues.apache.org/jira/browse/CASSANDRA-11459 Project: Cassandra Issue Type: Bug Components: Core Environment: ubuntu 14.04, 3 nodes in each datacenter, 1g networking, each node has 128G ram, 3*300G SSD in RAID5, dual E5-2620v3 processors Reporter: Yan Cui
We found the problem on Cassandra 2.0.15, and have not tested on other versions. there is one core table, and the schema is [user_id int, device_token text, deleted bool, device_info map<text, text>, human_code text] user_id and device token is the primary key, and user_id is the partition key, we have the statement that caused latency spike (3500ms to 4000 ms). select * from table where user_id = <hotuserid>. the hotuserid has roughly 80000 rows. On average, there is 200 bytes for each row. We feel this should be slow because of more results out there, but it is not expected to be that slow. -- This message was sent by Atlassian JIRA (v6.3.4#6332)