[jira] [Commented] (CASSANDRA-8016) Auth tables should use higher consistency level
[ https://issues.apache.org/jira/browse/CASSANDRA-8016?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14170277#comment-14170277 ] Vishy Kasar commented on CASSANDRA-8016: We need to fallback to CL.LOCAL_QUORUM also in case of CL.LOCAL_ONE failure (as opposed to miss) > Auth tables should use higher consistency level > --- > > Key: CASSANDRA-8016 > URL: https://issues.apache.org/jira/browse/CASSANDRA-8016 > Project: Cassandra > Issue Type: Bug >Reporter: T Jake Luciani >Assignee: Aleksey Yeschenko > Fix For: 2.0.11 > > > The Auth code in Cassandra uses CL.ONE or CL.LOCAL_ONE except in the case of > the superuser. > Since the Auth keyspace is created with RF=1 the default experience is fine. > However if you change to RF > 1 suddenly the select statements are open to > misses. > We should change reads/writes in Auth, PasswordAuthenticator, > CassandraAuthorizer to always use LOCAL_QUORUM/QUORUM. > For reads we could optimize the code to start with CL.ONE and on a miss > increase to CL.QUORUM -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (CASSANDRA-8016) Auth tables should use higher consistency level
[ https://issues.apache.org/jira/browse/CASSANDRA-8016?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14153808#comment-14153808 ] Aleksey Yeschenko commented on CASSANDRA-8016: -- We can bump it to CL.LOCAL_QUORUM in case of a miss w/ CL.LOCAL_ONE though - for making it easier to increase the RF on a live cluster, so +1 to that. > Auth tables should use higher consistency level > --- > > Key: CASSANDRA-8016 > URL: https://issues.apache.org/jira/browse/CASSANDRA-8016 > Project: Cassandra > Issue Type: Bug >Reporter: T Jake Luciani > Fix For: 2.0.11 > > > The Auth code in Cassandra uses CL.ONE or CL.LOCAL_ONE except in the case of > the superuser. > Since the Auth keyspace is created with RF=1 the default experience is fine. > However if you change to RF > 1 suddenly the select statements are open to > misses. > We should change reads/writes in Auth, PasswordAuthenticator, > CassandraAuthorizer to always use LOCAL_QUORUM/QUORUM. > For reads we could optimize the code to start with CL.ONE and on a miss > increase to CL.QUORUM -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (CASSANDRA-8016) Auth tables should use higher consistency level
[ https://issues.apache.org/jira/browse/CASSANDRA-8016?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14152179#comment-14152179 ] Jonathan Ellis commented on CASSANDRA-8016: --- I think the reasoning was to maximize availability. Better to run with out of date permissions, than not run at all. > Auth tables should use higher consistency level > --- > > Key: CASSANDRA-8016 > URL: https://issues.apache.org/jira/browse/CASSANDRA-8016 > Project: Cassandra > Issue Type: Bug >Reporter: T Jake Luciani > Fix For: 2.0.11 > > > The Auth code in Cassandra uses CL.ONE or CL.LOCAL_ONE except in the case of > the superuser. > Since the Auth keyspace is created with RF=1 the default experience is fine. > However if you change to RF > 1 suddenly the select statements are open to > misses. > We should change reads/writes in Auth, PasswordAuthenticator, > CassandraAuthorizer to always use LOCAL_QUORUM/QUORUM. > For reads we could optimize the code to start with CL.ONE and on a miss > increase to CL.QUORUM -- This message was sent by Atlassian JIRA (v6.3.4#6332)