CASSANALYTICS-42 Add S3-backed Cassandra batch reader

build.gradle

@@ -461,3 +461,14 @@ sourcesDistTar.finalizedBy generateDistributionChecksums
 sourcesDistZip.finalizedBy generateDistributionChecksums
 assembleDist.finalizedBy generateDistributionChecksums
 assembleSourcesDist.finalizedBy generateDistributionChecksums
+
+// Wire shadowJar to run as part of the core subproject's assemble (and therefore the
+// root aggregate assemble), without creating a root-level finalizedBy on shadowJar.
+// The latter chains through distTar/distZip (which `dependsOn` every subproject Jar via
+// the `distributions { from subprojects.collect { it.tasks.withType(Jar) } }` block) and
+// loops back to :assemble, causing a Gradle 8 circular-dependency failure.
+project(':cassandra-analytics-core') {
+  afterEvaluate {
+    tasks.named('assemble').configure { dependsOn tasks.named('shadowJar') }
+  }
+}

cassandra-analytics-common/src/main/java/org/apache/cassandra/analytics/stats/Stats.java

@@ -25,6 +25,7 @@
 import java.util.Set;
 
 import org.apache.cassandra.spark.data.CqlField;
+import org.apache.cassandra.spark.data.FileType;
 import org.apache.cassandra.spark.data.SSTable;
 import org.apache.cassandra.spark.data.SSTablesSupplier;
 import org.apache.cassandra.spark.reader.IndexEntry;
@@ -498,4 +499,50 @@ public void indexFileSkipped()
     {
 
     }
+
+    /**
+     * S3 headObject operation performed for existence check.
+     *
+     * @param timeNanos time taken in nanoseconds for the S3 headObject operation
+     */
+    public void s3HeadObjectOperation(long timeNanos)
+    {
+
+    }
+
+    /**
+     * S3 getObject operation performed for data retrieval.
+     *
+     * @param timeNanos time taken in nanoseconds for the S3 getObject operation
+     */
+    public void s3GetObjectOperation(long timeNanos)
+    {
+
+    }
+
+    /**
+     * Mutable SSTable metadata objects (Summary.db, Filter.db, Statistics.db) can be rewritten in place
+     * on Cassandra live data directories. Backup manifests can therefore carry a stale size for the
+     * same S3 key. This event records when the actual object size differs from the manifest-provided
+     * size.
+     *
+     * @param fileType     mutable SSTable component type
+     * @param manifestSize size recorded in the autosnap manifest
+     * @param actualSize   current S3 object size
+     */
+    public void s3MutableMetadataDriftDetected(FileType fileType, long manifestSize, long actualSize)
+    {
+
+    }
+
+    /**
+     * Mutable metadata object was large enough that the reader used HEAD + exact ranged GET instead
+     * of a single open-ended GET.
+     *
+     * @param fileType mutable SSTable component type
+     */
+    public void s3MutableMetadataHeadFallback(FileType fileType)
+    {
+
+    }
 }

cassandra-analytics-common/src/main/java/org/apache/cassandra/spark/data/FileType.java

@@ -83,4 +83,15 @@ public String getFileSuffix()
     {
         return fileSuffix;
     }
+
+    /**
+     * Whether the on-disk size of this component can drift from the value recorded in a backup
+     * manifest. Cassandra rewrites Summary/Filter/Statistics in place during compaction, so a
+     * stale manifest size for these components can produce truncated ranged-GETs against the
+     * backing store. The data layer treats these components specially when issuing reads.
+     */
+    public boolean isMutableMetadata()
+    {
+        return this == SUMMARY || this == FILTER || this == STATISTICS;
+    }
 }