CASSANDRA-18397: Unified Compaction Strategy

CHANGES.txt

@@ -1,4 +1,5 @@
 5.0
+ * Implementation of the Unified Compaction Strategy as described in CEP-26 (CASSANDRA-18397)
  * Upgrade commons cli to 1.5.0 (CASSANDRA-18659)
  * Disable the deprecated keyspace/table thresholds and convert them to guardrails (CASSANDRA-18617)
  * Deprecate CloudstackSnitch and remove duplicate code in snitches (CASSANDRA-18438)

NEWS.txt

@@ -71,6 +71,10 @@ using the provided 'sstableupgrade' tool.
 
 New features
 ------------
+    - Added a new "unified" compaction strategy that supports the use cases of the legacy compaction strategies, with
+      low space overhead, high parallelism and flexible configuration. Implemented by the UnifiedCompactionStrategy
+      class. Further details and documentation can be found in
+      src/java/org/apache/cassandra/db/compaction/UnifiedCompactionStrategy.md
     - New `VectorType` (cql `vector<element_type, dimension>`) which adds new fixed-length element arrays. See CASSANDRA-18504
     - Removed UDT type migration logic for 3.6+ clusters upgrading to 4.0.  If migration has been disabled, it must be
       enabled before upgrading to 5.0 if the cluster used UDTs. See CASSANDRA-18504

conf/cassandra.yaml

@@ -1002,6 +1002,20 @@ snapshot_links_per_second: 0
 # Min unit: KiB
 column_index_cache_size: 2KiB
 
+# Default compaction strategy, applied when a table's parameters do not
+# specify compaction.
+# The selected compaction strategy will also apply to system tables.
+#
+# The default is to use SizeTieredCompactionStrategy, with its default
+# compaction parameters.
+#
+# default_compaction:
+#   class_name: UnifiedCompactionStrategy
+#   parameters:
+#     scaling_parameters: T4
+#     target_sstable_size: 1GiB
+
+
 # Number of simultaneous compactions to allow, NOT including
 # validation "compactions" for anti-entropy repair.  Simultaneous
 # compactions can help preserve read performance in a mixed read/write

src/java/org/apache/cassandra/config/CassandraRelevantProperties.java

@@ -29,6 +29,7 @@
 import org.apache.cassandra.db.virtual.LogMessagesTable;
 import org.apache.cassandra.exceptions.ConfigurationException;
 import org.apache.cassandra.service.FileSystemOwnershipCheck;
+import org.apache.cassandra.utils.FBUtilities;
 import org.apache.cassandra.utils.StorageCompatibilityMode;
 
 // checkstyle: suppress below 'blockSystemPropertyUsage'
@@ -516,6 +517,12 @@ public enum CassandraRelevantProperties
     TRIGGERS_DIR("cassandra.triggers_dir"),
     TRUNCATE_BALLOT_METADATA("cassandra.truncate_ballot_metadata"),
     TYPE_UDT_CONFLICT_BEHAVIOR("cassandra.type.udt.conflict_behavior"),
+    // See org.apache.cassandra.db.compaction.unified.Controller for the definition of the UCS parameters
+    UCS_BASE_SHARD_COUNT("unified_compaction.base_shard_count", "4"),
+    UCS_OVERLAP_INCLUSION_METHOD("unified_compaction.overlap_inclusion_method"),
+    UCS_SCALING_PARAMETER("unified_compaction.scaling_parameters", "T4"),
+    UCS_SURVIVAL_FACTOR("unified_compaction.survival_factor", "1"),
+    UCS_TARGET_SSTABLE_SIZE("unified_compaction.target_sstable_size", "1GiB"),
     UDF_EXECUTOR_THREAD_KEEPALIVE_MS("cassandra.udf_executor_thread_keepalive_ms", "30000"),
     UNSAFE_SYSTEM("cassandra.unsafesystem"),
     /** User's home directory. */
@@ -725,6 +732,56 @@ public long getLong(long overrideDefaultValue)
         return LONG_CONVERTER.convert(value);
     }
 
+    /**
+     * Gets the value of a system property as a double.
+     * @return System property value if it exists, defaultValue otherwise. Throws an exception if no default value is set.
+     */
+    public double getDouble()
+    {
+        String value = System.getProperty(key);
+        if (value == null && defaultVal == null)
+            throw new ConfigurationException("Missing property value or default value is not set: " + key);
+        return DOUBLE_CONVERTER.convert(value == null ? defaultVal : value);
+    }
+
+    /**
+     * Gets the value of a system property as a double.
+     * @return system property value if it exists, defaultValue otherwise.
+     */
+    public double getDouble(double overrideDefaultValue)
+    {
+        String value = System.getProperty(key);
+        if (value == null)
+            return overrideDefaultValue;
+
+        return DOUBLE_CONVERTER.convert(value);
+    }
+
+    /**
+     * Gets the value of a system property, given as a human-readable size in bytes (e.g. 100MiB, 10GB, 500B).
+     * @return System property value if it exists, defaultValue otherwise. Throws an exception if no default value is set.
+     */
+    public long getSizeInBytes()
+    {
+        String value = System.getProperty(key);
+        if (value == null && defaultVal == null)
+            throw new ConfigurationException("Missing property value or default value is not set: " + key);
+        return SIZE_IN_BYTES_CONVERTER.convert(value == null ? defaultVal : value);
+    }
+
+    /**
+     * Gets the value of a system property, given as a human-readable size in bytes (e.g. 100MiB, 10GB, 500B).
+     * @return System property value if it exists, defaultValue otherwise.
+     */
+    public long getSizeInBytes(long overrideDefaultValue)
+    {
+        String value = System.getProperty(key);
+        if (value == null)
+            return overrideDefaultValue;
+
+        return SIZE_IN_BYTES_CONVERTER.convert(value);
+    }
+
     /**
      * Gets the value of a system property as an int.
      * @return system property int value if it exists, overrideDefaultValue otherwise.
@@ -847,6 +904,32 @@ public interface PropertyConverter<T>
         }
     };
 
+    private static final PropertyConverter<Long> SIZE_IN_BYTES_CONVERTER = value ->
+    {
+        try
+        {
+            return FBUtilities.parseHumanReadableBytes(value);
+        }
+        catch (ConfigurationException e)
+        {
+            throw new ConfigurationException(String.format("Invalid value for system property: " +
+                                                           "expected size in bytes with unit but got '%s'\n%s", value, e));
+        }
+    };
+
+    private static final PropertyConverter<Double> DOUBLE_CONVERTER = value ->
+    {
+        try
+        {
+            return Double.parseDouble(value);
+        }
+        catch (NumberFormatException e)
+        {
+            throw new ConfigurationException(String.format("Invalid value for system property: " +
+                                                           "expected floating point value but got '%s'", value));
+        }
+    };
+
     /**
      * @return whether a system property is present or not.
      */

src/java/org/apache/cassandra/config/Config.java

@@ -1098,6 +1098,11 @@ public enum PaxosOnLinearizabilityViolation
     public volatile long min_tracked_partition_tombstone_count = 5000;
     public volatile boolean top_partitions_enabled = true;
 
+    /**
+     * Default compaction configuration, used if a table does not specify any.
+     */
+    public ParameterizedClass default_compaction = null;
+
     public static Supplier<Config> getOverrideLoadConfig()
     {
         return overrideLoadConfig;

src/java/org/apache/cassandra/config/DatabaseDescriptor.java

@@ -4751,4 +4751,9 @@ public static StorageCompatibilityMode getStorageCompatibilityMode()
         else
             return conf.storage_compatibility_mode;
     }
+
+    public static ParameterizedClass getDefaultCompaction()
+    {
+        return conf != null ? conf.default_compaction : null;
+    }
 }

src/java/org/apache/cassandra/db/AbstractCompactionController.java

@@ -44,7 +44,7 @@ public AbstractCompactionController(final ColumnFamilyStore cfs, final long gcBe
 
     public String getKeyspace()
     {
-        return cfs.keyspace.getName();
+        return cfs.getKeyspaceName();
     }
 
     public String getColumnFamily()