java - 光束 : writing per window element count with window boundaries

Question

为了进行简单的概念验证，我尝试在两分钟的窗口中显示点击数据。我想要做的就是将每个窗口的计数以及窗口的边界打印到 BigQuery。在运行我的管道时，我不断收到以下错误:

org.apache.beam.sdk.Pipeline$PipelineExecutionException: java.lang.RuntimeException: java.io.IOException: Insert failed: [{"errors":[{"debugInfo":"","location":"windowend","message":"This field is not a record.","reason":"invalid"}],"index":0}]

管道看起来像这样:

// Creating the pipeline
Pipeline p = Pipeline.create(options);

// Window items
PCollection<TableRow> counts = p.apply("ReadFromPubSub", PubsubIO.readStrings().fromTopic(options.getTopic()))
.apply("AddEventTimestamps", WithTimestamps.of(TotalCountPipeline::ExtractTimeStamp).withAllowedTimestampSkew(Duration.standardDays(10000)))
        .apply("Window", Window.<String>into(
                FixedWindows.of(Duration.standardHours(options.getWindowSize())))
                .triggering(
                        AfterWatermark.pastEndOfWindow()
                                .withLateFirings(AfterPane.elementCountAtLeast(1)))
                .withAllowedLateness(Duration.standardDays(10000))
                .accumulatingFiredPanes())
        .apply("CalculateSum", Combine.globally(Count.<String>combineFn()).withoutDefaults())
        .apply("BigQueryFormat", ParDo.of(new FormatCountsFn()));

// Writing to BigQuery
counts.apply("WriteToBigQuery",BigQueryIO.writeTableRows()
                .to(options.getOutputTable())
                .withSchema(getSchema())
.withWriteDisposition(BigQueryIO.Write.WriteDisposition.WRITE_APPEND));

// Execute pipeline
p.run().waitUntilFinish();

我猜测这与 BigQuery 格式化函数有关，其实现如下:

static class FormatCountsFn extends DoFn<Long, TableRow> {
    @ProcessElement
    public void processElement(ProcessContext c, BoundedWindow window) {
        TableRow row =
                new TableRow()
                        .set("windowStart", window.maxTimestamp().toDateTime())
                        .set("count", c.element().intValue());
        c.output(row);
    }
}

受到 this post 的启发。有人能解释一下吗？我似乎无法理解它。

Answer 1

显然这个问题的答案与波束窗口无关，仅与 BigQuery 有关。将 DateTime 对象写入 BigQuery 行需要采用正确 yyyy-MM-dd HH:mm:ss 格式的字符串，这与我提供的 DateTime 对象形成对比。