[ https://issues.apache.org/jira/browse/CASSANDRA-14803?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Blake Eggleston updated CASSANDRA-14803: ---------------------------------------- Status: Patch Available (was: Open) |[3.0|https://github.com/bdeggleston/cassandra/tree/14803-3.0]|[circle|https://circleci.com/gh/bdeggleston/workflows/cassandra/tree/cci%2F14803-3.0]| |[3.11|https://github.com/bdeggleston/cassandra/tree/14803-3.11]|[circle|https://circleci.com/gh/bdeggleston/workflows/cassandra/tree/cci%2F14803-3.11]| The sstable used for the test was generated from [here|https://github.com/bdeggleston/cassandra/tree/14803-2.1] Since this is testing a specific problem upgrading from 2.x-3.x, it didn't seem like LegacySSTableTest was the right place for this > Rows that cross index block boundaries can cause incomplete reverse reads in > some cases. > ---------------------------------------------------------------------------------------- > > Key: CASSANDRA-14803 > URL: https://issues.apache.org/jira/browse/CASSANDRA-14803 > Project: Cassandra > Issue Type: Bug > Reporter: Blake Eggleston > Assignee: Blake Eggleston > Priority: Major > Fix For: 3.0.x, 3.11.x > > > When we're reading 2.1 sstables in reverse, we skip the first row of an index > block if it's split across index boundaries. The entire row will be read at > the end of the next block. In some cases though, the only thing in this index > block is the partial row, so we return an empty iterator. The empty iterator > is then interpreted as the end of the row further down the call stack, so we > return early without reading the rest of the data. This only affects 3.x > during upgrades from 2.1 -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org