[ https://issues.apache.org/jira/browse/CASSANDRA-10215?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14723651#comment-14723651 ]
T Jake Luciani commented on CASSANDRA-10215: -------------------------------------------- Overall looks good. - Why use a UUID vs just passing the name? Since the UUID is basically the MD5 of the name? is it to have a fixed length of key-space and index name? For debugging purposes it would be clearer to have the name send in case you have a schema disagreement and you can see what index it's looking for. The exception log in deserializeIndexMetadata would ideally have the name. I think you should make that message clearer as to what just happened. - Why do you have IndexMetadata in the constructor of the read command when it's always passing Optional.empty() in the code? > Reduce redundant secondary index selection lookups > -------------------------------------------------- > > Key: CASSANDRA-10215 > URL: https://issues.apache.org/jira/browse/CASSANDRA-10215 > Project: Cassandra > Issue Type: Improvement > Reporter: Sam Tunnicliffe > Assignee: Sam Tunnicliffe > Fix For: 3.0.0 rc1 > > > Once an index has been identified as being the one that should be used for a > given query, it should be included as part of the {{ReadCommand}} so that it > can be retrieved when needed during execution. Currently we don't do this, > meaning that the index selection process implemented in > {{SIM::getBestIndexFor}} is run multiple times on both the coordinator and > each replica when performing a range command. > This is also a pre-requisite of CASSANDRA-10214 -- This message was sent by Atlassian JIRA (v6.3.4#6332)