[CASSANDRA-14336] sstableloader fails if sstables contains removed columns

• Type: Bug
• Status: Open
• Priority: Normal
• Resolution: Unresolved
• Component/s: Legacy/Tools
• Severity: Normal
• Assignee: Jaydeepkumar Chovatia
• Reporter: Hannu Kröger

Authors

Jaydeepkumar Chovatia

• Created: 22/Mar/18 10:43
• Updated: 31/Aug/20 06:41

Description

If I copy the schema and try to load in sstables with sstableloader, loading sometimes fails with

Exception in thread "main" org.apache.cassandra.tools.BulkLoadException: java.lang.RuntimeException: Failed to list files in /tmp/test/bug3_dest-acdc
    at org.apache.cassandra.tools.BulkLoader.load(BulkLoader.java:93)
    at org.apache.cassandra.tools.BulkLoader.main(BulkLoader.java:48)
Caused by: java.lang.RuntimeException: Failed to list files in /tmp/test/bug3_dest-acdc
    at org.apache.cassandra.db.lifecycle.LogAwareFileLister.list(LogAwareFileLister.java:77)
    at org.apache.cassandra.db.lifecycle.LifecycleTransaction.getFiles(LifecycleTransaction.java:561)
    at org.apache.cassandra.io.sstable.SSTableLoader.openSSTables(SSTableLoader.java:76)
    at org.apache.cassandra.io.sstable.SSTableLoader.stream(SSTableLoader.java:165)
    at org.apache.cassandra.tools.BulkLoader.load(BulkLoader.java:80)
    ... 1 more
Caused by: java.lang.RuntimeException: Unknown column d during deserialization
    at org.apache.cassandra.db.SerializationHeader$Component.toHeader(SerializationHeader.java:321)
    at org.apache.cassandra.io.sstable.format.SSTableReader.openForBatch(SSTableReader.java:440)
    at org.apache.cassandra.io.sstable.SSTableLoader.lambda$openSSTables$0(SSTableLoader.java:121)
    at org.apache.cassandra.db.lifecycle.LogAwareFileLister.lambda$innerList$2(LogAwareFileLister.java:99)
    at java.util.stream.ReferencePipeline$2$1.accept(ReferencePipeline.java:174)
    at java.util.TreeMap$EntrySpliterator.forEachRemaining(TreeMap.java:2969)
    at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:481)
    at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:471)
    at java.util.stream.ReduceOps$ReduceOp.evaluateSequential(ReduceOps.java:708)
    at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234)
    at java.util.stream.ReferencePipeline.collect(ReferencePipeline.java:499)
    at org.apache.cassandra.db.lifecycle.LogAwareFileLister.innerList(LogAwareFileLister.java:101)
    at org.apache.cassandra.db.lifecycle.LogAwareFileLister.list(LogAwareFileLister.java:73)
    ... 5 more

This requires that we have dropped columns in the source table and sstables exist from the "old schema" time.

This can be very easily reproduced. I used following script:

KS=test
SRCTABLE=bug3_source
DESTTABLE=bug3_dest
DATADIR=/usr/local/var/lib/cassandra/data
TMPDIR=/tmp

cqlsh -e "CREATE TABLE $KS.$SRCTABLE(a int primary key, b int, c int, d int);"
cqlsh -e "CREATE TABLE $KS.$DESTTABLE(a int primary key, b int, c int);"
cqlsh -e "INSERT INTO $KS.$SRCTABLE(a,b,c,d) values(1,2,3,4);"
nodetool flush $KS $SRCTABLE
cqlsh -e "ALTER TABLE $KS.$SRCTABLE DROP d;"
nodetool flush $KS $SRCTABLE

mkdir -p $TMPDIR/$KS/$DESTTABLE-acdc
cp $DATADIR/$KS/$SRCTABLE-*/* $TMPDIR/$KS/$DESTTABLE-acdc
sstableloader -d 127.0.0.1 $TMPDIR/$KS/$DESTTABLE-acdc

Issue Links

• is duplicated by: CASSANDRA-11416 No longer able to load backups into new cluster if there was a dropped column