Hi to all, we wanted to do a group by on elements that can contains null values and we discovered that Table API support this while Dataset API does not. Is this documented somehwere on the Flink site? Best, Flavio ------------------------------------------------------- PS: you can test this with the following main: public static void main(String[] args) throws Exception { final ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment(); final BatchTableEnvironment btEnv = TableEnvironment.getTableEnvironment(env); final DataSet<String> testDs = env .fromElements("test", "test", "test2", "null", "null", "test3") .map(x -> "null".equals(x) ? null : x); boolean testDatasetApi = true; if (testDatasetApi) { testDs.groupBy(x -> x).reduceGroup(new GroupReduceFunction<String, Integer>() { @Override public void reduce(Iterable<String> values, Collector<Integer> out) throws Exception { int cnt = 0; for (String value : values) { cnt++; } out.collect(cnt); } }).print(); } btEnv.registerDataSet("TEST", testDs, "field1"); Table res = btEnv.sqlQuery("SELECT field1, count(*) as cnt FROM TEST GROUP BY field1"); DataSet<Row> result = btEnv.toDataSet(res, new RowTypeInfo(BasicTypeInfo.STRING_TYPE_INFO, BasicTypeInfo.LONG_TYPE_INFO)); result.print(); } |
Hi Flavio, Whether groupBy with null values works or not depends on the type of the key, or more specifically on the TypeComparator and TypeSerializer that are used to serialize, compare, and hash the key type. The processing engine supports null values If the comparator and serializer can handle null input values. Flink SQL wraps keys in the Row type and the corresponding serializer / comparator can handle null fields. If you use Row in DataSet / DataStream programs, null values are supported as well. I think it would be good to discuss the handling of null keys on the documentation about data types [1] and link to that from operators that require keys. Would you mind creating a Jira issue for that? Thank you, Fabian Am Mo., 19. Nov. 2018 um 12:31 Uhr schrieb Flavio Pompermaier <[hidden email]>:
|
Sure! The problem is that Dataset API does an implicit conversion to Tuples during chaining and I didn't found any documentation about this (actually I was pleasantly surprised by the fact that the Table API were supporting aggregates on null values..). Thanks for the reply, Flavio On Tue, Nov 20, 2018 at 11:33 AM Fabian Hueske <[hidden email]> wrote:
|
I assigned the issue to me. Because I
wanted to that for a very long time. I already did some
prerequisite work for the documentation in
`org.apache.flink.api.common.typeinfo.Types`.
Thanks,
Timo
Am 20.11.18 um 11:44 schrieb Flavio
Pompermaier:
|
Free forum by Nabble | Edit this page |