package parallelai.spyglass.jdbc.example import org.scalatest.junit.JUnitRunner import org.junit.runner.RunWith import org.scalatest.FunSpec import com.twitter.scalding.{Tsv, JobTest, TupleConversions} import org.slf4j.LoggerFactory import cascading.tuple.{Tuple, Fields} import parallelai.spyglass.jdbc.JDBCSource /** * Simple example of JDBCSource testing */ @RunWith(classOf[JUnitRunner]) class JdbcSourceExampleTest extends FunSpec with TupleConversions { val output = "src/test/resources/outputs/testOutput" val log = LoggerFactory.getLogger(this.getClass.getName) val sampleData = List( (1, "c11", "c12", 11), (2, "c21", "c22", 22) ) JobTest("parallelai.spyglass.jdbc.example.JdbcSourceExample") .arg("test", "") .arg("app.conf.path", "app.conf") .arg("output", output) .arg("debug", "true") .source( new JDBCSource( "db_name", "com.mysql.jdbc.Driver", "jdbc:mysql://:/?zeroDateTimeBehavior=convertToNull", "user", "password", List("KEY_ID", "COL1", "COL2", "COL3"), List("bigint(20)", "varchar(45)", "varchar(45)", "bigint(20)"), List("key_id"), new Fields("key_id", "col1", "col2", "col3")), sampleData) .sink[Tuple](Tsv(output.format("get_list"))) { outputBuffer => log.debug("Output => " + outputBuffer) it("should return the mock data provided.") { assert(outputBuffer.size === 2) } } .run .finish }