[CASSANDRA-18956] Metadata Loss caused by unchecked rename column when upgrading from 2.x to 3.0.29

• Type: Bug
• Status: Needs Committer
• Priority: Normal
• Resolution: Unresolved
• Fix Version/s: 3.0.x
• Component/s: Legacy/Distributed Metadata
• Bug Category: Correctness - Unrecoverable Corruption / Loss
• Severity: Normal
• Complexity: Low Hanging Fruit
• Discovered By: Adhoc Test
• Platform: All

Test and Documentation Plan

CI

• Assignee: Ke Han
• Reporter: Ke Han

Authors

Ke Han

Reviewers

Stefan Miklosovic

• Created: 24/Oct/23 17:14
• Updated: 06/Nov/23 23:37

Description

After migrating data from 2.1.22 (released 2022-06-17) to 3.0.29 (released 2023-05-05), if user performs a rename operation to rename the primary key to an existing column, it will succeed, and the original column will get lost permanently.

Reproduce

Start up single cassandra node 2.1.22

Start up cassandra-cli

bin/cassandra-cli 

Execute the following commands using cassandra-cli interface

create keyspace test with strategy_options = {replication_factor:1} and placement_strategy = 'org.apache.cassandra.locator.SimpleStrategy';

use test;

create column family foos 
with column_type = 'Standard' 
and comparator = 'BytesType'
and key_validation_class = 'UTF8Type' 
and column_metadata = [{column_name: '2f', validation_class: 'UTF8Type'}]; 

Stop the daemon

bin/nodetool -h ::FFFF:127.0.0.1 stopdaemon 

Upgrade, start up new version, start up cqlsh with bin/cqlsh

cqlsh> desc test.foos ;
CREATE TABLE test.foos (
    key text PRIMARY KEY,
    "2f" text
) WITH COMPACT STORAGE
    AND bloom_filter_fp_chance = 0.01
    AND caching = {'keys': 'ALL', 'rows_per_partition': 'NONE'}
    AND comment = ''
    AND compaction = {'class': 'org.apache.cassandra.db.compaction.SizeTieredCompactionStrategy', 'max_threshold': '32', 'min_threshold': '4'}
    AND compression = {'chunk_length_in_kb': '64', 'class': 'org.apache.cassandra.io.compress.LZ4Compressor'}
    AND crc_check_chance = 1.0
    AND dclocal_read_repair_chance = 0.1
    AND default_time_to_live = 0
    AND gc_grace_seconds = 864000
    AND max_index_interval = 2048
    AND memtable_flush_period_in_ms = 0
    AND min_index_interval = 128
    AND read_repair_chance = 0.0
    AND speculative_retry = 'NONE';

cqlsh> ALTER TABLE test.foos RENAME key TO "2f";
cqlsh> desc test.foos ;
CREATE TABLE test.foos (
    "2f" text PRIMARY KEY
) WITH COMPACT STORAGE
    AND bloom_filter_fp_chance = 0.01
    AND caching = {'keys': 'ALL', 'rows_per_partition': 'NONE'}
    AND comment = ''
    AND compaction = {'class': 'org.apache.cassandra.db.compaction.SizeTieredCompactionStrategy', 'max_threshold': '32', 'min_threshold': '4'}
    AND compression = {'chunk_length_in_kb': '64', 'class': 'org.apache.cassandra.io.compress.LZ4Compressor'}
    AND crc_check_chance = 1.0
    AND dclocal_read_repair_chance = 0.1
    AND default_time_to_live = 0
    AND gc_grace_seconds = 864000
    AND max_index_interval = 2048
    AND memtable_flush_period_in_ms = 0
    AND min_index_interval = 128
    AND read_repair_chance = 0.0
    AND speculative_retry = 'NONE';

cqlsh> SELECT * FROM test.foos ;
 2f
----

(0 rows) 

Root Cause

The root cause is a missing check for the existing column when the table comparator from the old version is BytesType.

I have uploaded the data dir, use it to start up 3.0.29 and execute the commands will reproduce this bug.

Issue Links

• links to: GitHub Pull Request #2840
• links to: GitHub Pull Request #2861