Castled 源码解析 - container 模块说明

container 属于Castled api 后端服务，后端包含了任务调度，db 迁移，有几个服务是比较重要的
主要是pipelineservice，ExternalAppService，WarehouseService，而且官方还提供了一套基于events 的处理
主要包含PipelineEvent,CastledEvent,其他的主要是基于dropwizard 开发的rest api 了，整体代码并不难
pipelineservice 在其中比较核心，进行了app 与connector 的关联操作，pipelineservice 会使用到event ，task 处理
PipelineExecutor 对于数据的处理主要是在此task 执行的（时间数据拉取以及发送处理就是在这里边的）

pipelineservice核心方法

参考下图

 

 

PipelineExecutor 处理

 

 public String executeTask(Task task) {

        Long pipelineId = ((Number) task.getParams().get(CommonConstants.PIPELINE_ID)).longValue();

        Pipeline pipeline = this.pipelineService.getActivePipeline(pipelineId);

        if (pipeline == null) {

            return null;

        }

        // 挺重要的，进行状态统计的

        WarehouseSyncFailureListener warehouseSyncFailureListener = null;

        Warehouse warehouse = this.warehouseService.getWarehouse(pipeline.getWarehouseId());

        PipelineRun pipelineRun = getOrCreatePipelineRun(pipelineId);

        WarehousePollContext warehousePollContext = WarehousePollContext.builder()

                .primaryKeys(PipelineUtils.getWarehousePrimaryKeys(pipeline)).pipelineUUID(pipeline.getUuid())

                .pipelineRunId(pipelineRun.getId()).warehouseConfig(warehouse.getConfig())

                .dataEncryptionKey(encryptionManager.getEncryptionKey(warehouse.getTeamId()))

                .queryMode(pipeline.getQueryMode())

                .query(pipeline.getSourceQuery()).pipelineId(pipeline.getId()).build();

        try {

          // 调用warehouse connector，获取数据

            WarehouseExecutionContext warehouseExecutionContext = pollRecords(warehouse, pipelineRun, warehousePollContext);

 

            log.info("Poll records completed for pipeline {}", pipeline.getName());

            this.pipelineService.updatePipelineRunstage(pipelineRun.getId(), PipelineRunStage.RECORDS_POLLED);

 

            ExternalApp externalApp = externalAppService.getExternalApp(pipeline.getAppId());

            ExternalAppConnector externalAppConnector = this.externalAppConnectors.get(externalApp.getType());

            RecordSchema appSchema = externalAppConnector.getSchema(externalApp.getConfig(), pipeline.getAppSyncConfig())

                    .getAppSchema();

 

            log.info("App schema fetch completed for pipeline {}", pipeline.getName());

 

            warehousePollContext.setWarehouseSchema(warehouseExecutionContext.getWarehouseSchema());

            warehouseSyncFailureListener = warehouseConnectors.get(warehouse.getType())

                    .syncFailureListener(warehousePollContext);

 

            MysqlErrorTracker mysqlErrorTracker = new MysqlErrorTracker(warehousePollContext);

 

            ErrorOutputStream schemaMappingErrorOutputStream = new ErrorOutputStream(warehouseSyncFailureListener, mysqlErrorTracker);

 

            SchemaMappedMessageInputStream schemaMappedMessageInputStream = new SchemaMappedMessageInputStream(

                    appSchema, warehouseExecutionContext.getMessageInputStreamImpl(), pipeline.getDataMapping().appWarehouseMapping(),

                    pipeline.getDataMapping().warehouseAppMapping(), schemaMappingErrorOutputStream);

 

            SchemaMappedRecordOutputStream schemaMappedRecordOutputStream =

                    new SchemaMappedRecordOutputStream(SchemaUtils.filterSchema(warehousePollContext.getWarehouseSchema(),

                            PipelineUtils.getWarehousePrimaryKeys(pipeline)), warehouseSyncFailureListener,

                            pipeline.getDataMapping().warehouseAppMapping());

 

            ErrorOutputStream sinkErrorOutputStream = new ErrorOutputStream(schemaMappedRecordOutputStream,

                    new SchemaMappedErrorTracker(mysqlErrorTracker, warehouseExecutionContext.getWarehouseSchema(), pipeline.getDataMapping().warehouseAppMapping()));

 

            log.info("App Sync started for pipeline {}", pipeline.getName());

 

            List<String> mappedAppFields = pipeline.getDataMapping().getFieldMappings().stream().filter(mapping -> !mapping.isSkipped())

                    .map(FieldMapping::getAppField).collect(Collectors.toList());

 

            DataSinkRequest dataSinkRequest = DataSinkRequest.builder().externalApp(externalApp).errorOutputStream(sinkErrorOutputStream)

                    .appSyncConfig(pipeline.getAppSyncConfig()).mappedFields(mappedAppFields)

                    .objectSchema(appSchema).primaryKeys(pipeline.getDataMapping().getPrimaryKeys())

                    .messageInputStream(schemaMappedMessageInputStream)

                    .build();

 

           //  进行数据同步使用了MonitoredDataSink 对象，实现了一些统计信息

            PipelineSyncStats pipelineSyncStats = monitoredDataSink.syncRecords(externalAppConnector.getDataSink(),

                    pipelineRun.getPipelineSyncStats(), pipelineRun.getId(), dataSinkRequest);

 

            schemaMappedMessageInputStream.close();

 

            log.info("App Sync completed for pipeline {}", pipeline.getName());

            //flush output streams

            schemaMappingErrorOutputStream.flushFailedRecords();

            sinkErrorOutputStream.flushFailedRecords();

 

            warehouseConnectors.get(warehouse.getType()).getDataPoller().cleanupPipelineRunResources(warehousePollContext);

            // Also add the records that failed schema mapping phase to the final stats

            pipelineSyncStats.setRecordsFailed(schemaMappedMessageInputStream.getFailedRecords() + pipelineSyncStats.getRecordsFailed());

            this.pipelineService.markPipelineRunProcessed(pipelineRun.getId(), pipelineSyncStats);

 

        } catch (Exception e) {

            if (ObjectRegistry.getInstance(AppShutdownHandler.class).isShutdownTriggered()) {

                throw new PipelineInterruptedException();

            }

            this.pipelineService.markPipelineRunFailed(pipelineRun.getId(), Optional.ofNullable(e.getMessage()).orElse("Unknown Error"));

            log.error("Pipeline run failed for pipeline {} ", pipeline.getId(), e);

            this.warehouseConnectors.get(warehouse.getType()).getDataPoller().cleanupPipelineRunResources(warehousePollContext);

            Optional.ofNullable(warehouseSyncFailureListener).ifPresent(syncFailureListener ->

                    syncFailureListener.cleanupResources(pipeline.getUuid(), pipelineRun.getId(), warehouse.getConfig()));

 

            if (e instanceof PipelineExecutionException) {

                handlePipelineExecutionException(pipeline, (PipelineExecutionException) e);

            } else {

                log.error("Pipeline run failed for pipeline {} ", pipeline.getId(), e);

            }

        }

        return null;

    }

说明

目前从代码中可以看到每个创建的任务会发送消息到Castled的统计服务中，如果不需要的话，最好处理下，目前看配置定义，暂时没有开关可以禁用
尽管系统使用了kafka，但是感觉kafaka 的使用并不是很明显（更多是一个任务排队的处理），并不是基于kafka 的消息发送处理

参考资料

https://github.com/castledio/castled