我正在使用Beam SQL并尝试将整数转换为日期时间字段。
Schema resultSchema =
Schema.builder()
.addInt64Field("detectedCount")
.addStringField("sensor")
.addInt64Field("timestamp")
.build();
PCollection<Row> sensorRawUnboundedTimestampedSubset =
sensorRowUnbounded.apply(
SqlTransform.query(
"select PCOLLECTION.payload.`value`.`count` detectedCount, \n"
+ "PCOLLECTION.payload.`value`.`id` sensor, \n"
+ "PCOLLECTION.`timestamp` `timestamp` \n"
+ "from PCOLLECTION "))
.setRowSchema(resultSchema);
对于某些计算和窗口处理,我想将timestamp
转换/广播为Datetime
字段吗?请提供一些指针,以将timestamp
中的resultSchema
转换为DateTime
。数据类型。
答案 0 :(得分:1)
Beam(or in Calcite)中没有开箱即用的方法。简短版本-方解石或Beam无法知道如何实际存储日期或时间戳的整数形式。但是,假设您有纪元,这应该可以:
@Test
public void testBlah() throws Exception {
// input schema, has timestamps as epoch millis
Schema schema = Schema.builder().addInt64Field("ts").addStringField("st").build();
DateTime ts1 = new DateTime(2019, 8, 9, 10, 11, 12);
DateTime ts2 = new DateTime(2019, 8, 9, 10, 11, 12);
PCollection<Row> input =
pipeline
.apply(
"createRows",
Create.of(
Row.withSchema(schema).addValues(ts1.getMillis(), "two").build(),
Row.withSchema(schema).addValues(ts2.getMillis(), "twelve").build()))
.setRowSchema(schema);
PCollection<Row> result =
input.apply(
SqlTransform.query(
"SELECT \n"
+ "(TIMESTAMP '1970-01-01 00:00:00' + ts * INTERVAL '0.001' SECOND) as ts, \n"
+ "st \n"
+ "FROM \n"
+ "PCOLLECTION"));
// output schema, has timestamps as DateTime
Schema outSchema = Schema.builder().addDateTimeField("ts").addStringField("st").build();
PAssert.that(result)
.containsInAnyOrder(
Row.withSchema(outSchema).addValues(ts1, "two").build(),
Row.withSchema(outSchema).addValues(ts2, "twelve").build());
pipeline.run();
}
或者,您始终可以在Java中而不是在SQL中进行操作,只需将自定义ParDo
应用于SqlTransform
的输出即可。在该ParDo
中,从Row
对象中提取整数时间戳,将其转换为DateTime
,然后发出,例如作为具有不同架构的另一行的一部分。