Refactor the code around metadata

[brfrn169]

I wanted to add a DistributedStorageAdmin interface that gets table metadata and creates/drops tables. But, before that, we need to do some refactoring for the code around metadata. This PR is for this.

In this change, I removed the existing TableMetadata interface and added the new TableMetadata class with the builder class. Because of this, we can remove CassandraTableMetadata and DynamoTableMetadata and JdbcTableMetadata (CosmosTableMetadata is still needed though).

After it's merged, I want to add DistributedStorageAdmin like the following:

public interface DistributedStorageAdmin {
  TableMetadata getTableMetadata(String namespace, String table);
  void createTable(String namespace, String table, TableMetadata metadata, Map<String, String> options);
  void dropTable(String namespace, String table);
}

Please take a look at this!

[feeblefakie]

Overall looking good! Thank you.
Left 1 question and 1 minor comment.

[feeblefakie]

Just a question. Why this cannot be unified to TableMetadata?

[brfrn169]

Thank you for reviewing this!

We still need CosmosTableMetadata because we need a POJO (that means a class with some fields and their getters and setters maybe) to serialize and deserialize data in Cosmos DB.

For example, the following code writes the metadata into Cosmos DB with CosmosTableMetadata:

scalardb/src/integration-test/java/com/scalar/db/storage/cosmos/CosmosIntegrationTest.java

Lines 85 to 100 in 2ff0ff3

	CosmosTableMetadata metadata = new CosmosTableMetadata();
	metadata.setId(table(NAMESPACE, TABLE));
	metadata.setPartitionKeyNames(Collections.singletonList(COL_NAME1));
	metadata.setClusteringKeyNames(Collections.singletonList(COL_NAME4));
	metadata.setSecondaryIndexNames(new HashSet<>(Collections.singletonList(COL_NAME3)));
	Map<String, String> columns = new HashMap<>();
	columns.put(COL_NAME1, "int");
	columns.put(COL_NAME2, "text");
	columns.put(COL_NAME3, "int");
	columns.put(COL_NAME4, "int");
	columns.put(COL_NAME5, "boolean");
	metadata.setColumns(columns);
	client
	.getDatabase(database(METADATA_DATABASE))
	.getContainer(METADATA_CONTAINER)
	.createItem(metadata);

And the following code retrieves the metadata from Cosmos DB with CosmosTableMetadata:

scalardb/src/main/java/com/scalar/db/storage/cosmos/CosmosTableMetadataManager.java

Lines 49 to 51 in 2ff0ff3

	return container
	.readItem(fullName, new PartitionKey(fullName), CosmosTableMetadata.class)
	.getItem();

As the new TableMetadata is not a POJO, we can't use it for that.

[feeblefakie]

Thank you! I see.
Since CosmosTableMetadata seems only used in CosmosTableMetadataManager, can it be an inner class of the manager class?

[brfrn169]

CosmosTableMetadata is used only in CosmosTableMetadataManager in the main code, but it's actually used in the integration test code as well. And we might use this class another part when we introduce DistributedStorageAdmin. So I want to keep it an outer class for now. If we realize it doesn't need to be an outer class in the future, we can do that at that time. Thanks.

[feeblefakie]

OK, I understand.

[feeblefakie]

It's not in the google style guide, but I think it would be a bit better and consistent if a comment for a class starts with a noun.

[feeblefakie]

LGTM! Thank you!

[feeblefakie]

@yito88 I merged it but since it's a bit big PR so please take a look when you get a chance.