简述：
目前通过源码得知camunda流程引擎，在系统启动时，会根据bpmn文件部署流程，所以我们来看下整个bpmn文件是如何加载存储并使用的。

源码入口
SpringProcessEngineConfiguration

  @Override
  public ProcessEngine buildProcessEngine() {
    ProcessEngine processEngine = super.buildProcessEngine();
    //自动部署
    autoDeployResources(processEngine);
    return processEngine;
  }

autoDeployResources会通过RepositoryServiceImpl.deploy方法执行部署。内部利用了命令模式。最终调用DeployCmd.doExecute(CommandContext commandContext)
下面我们介绍下这个方法的执行逻辑。

  //解析源信息待部署
  Map<String, ResourceEntity> resourcesToDeploy = resolveResourcesToDeploy(commandContext, deployment);

resolveResourcesToDeploy方法主体分两步

step1：读取库中存有的最近一个部署版本和部署版本对应的流程图

库为:ACT_GE_BYTEARRAY 和 ACT_RE_DEPLOYMENT

  Map<String, ResourceEntity> existingResources = commandContext
          .getResourceManager()
          .findLatestResourcesByDeploymentName(deployment.getName(), containedResources.keySet(), source, deployment.getTenantId());

对应的mybatis sql 为

 <sql id="resourcesFromLastDeploymentWithName">
    <bind name="processApplicationSource" value="'process application'"/>

    select B.*
    from ${prefix}ACT_GE_BYTEARRAY B
    inner join
      (select B.NAME_, MAX(D.DEPLOY_TIME_) DEPLOY_TIME_
      from ${prefix}ACT_GE_BYTEARRAY B
        inner join ${prefix}ACT_RE_DEPLOYMENT D
        on B.DEPLOYMENT_ID_ = D.ID_
      where D.NAME_ = #{parameter.deploymentName}
        and
        <if test="processApplicationSource.equals(parameter.source)">
          (D.SOURCE_ is null or
        </if>
          D.SOURCE_ = #{parameter.source}
        <if test="processApplicationSource.equals(parameter.source)">
          )
        </if>
      <if test="parameter.resourcesToFind != null &amp;&amp; parameter.resourcesToFind.size > 0">
        and B.NAME_ in 
        <foreach item="resourceName" index="index" collection="parameter.resourcesToFind"
                 open="(" separator="," close=")">
          #{resourceName}
        </foreach>
      </if>
      group by B.NAME_) LAST_RESOURCE
      
    on B.NAME_ = LAST_RESOURCE.NAME_
    inner join ${prefix}ACT_RE_DEPLOYMENT D
    on B.DEPLOYMENT_ID_ = D.ID_
    and D.DEPLOY_TIME_ = LAST_RESOURCE.DEPLOY_TIME_ 
    and D.NAME_ = #{parameter.deploymentName}
    and
    <if test="processApplicationSource.equals(parameter.source)">
      (D.SOURCE_ is null or
    </if>
      D.SOURCE_ = #{parameter.source}
    <if test="processApplicationSource.equals(parameter.source)">
      )
    </if>
    <if test="parameter.tenantId != null">
      and D.TENANT_ID_ = #{ parameter.tenantId }
    </if>
    <if test="parameter.tenantId == null">
      and D.TENANT_ID_ is null
    </if>
  </sql>

step2：比较传入的流程图版本是否和库中的一致，如果不一致则使用当前resource中读取到的流程图

 if (existingResource == null
            || existingResource.isGenerated()
            || resourcesDiffer(deployedResource, existingResource))

resourcesDiffer方法

protected boolean resourcesDiffer(ResourceEntity resource, ResourceEntity existing) {
    byte[] bytes = resource.getBytes();
    byte[] savedBytes = existing.getBytes();
    return !Arrays.equals(bytes, savedBytes);
  }

拿到需要部署的流程图后,如果resourcesToDeploy不为空，那么需要将最新版本的流程落库存储并存入缓存DeploymentCache。

 if (!resourcesToDeploy.isEmpty()) {
          //落库并部署
          LOG.debugCreatingNewDeployment();
          deployment.setResources(resourcesToDeploy);
          deploy(deployment);
        } else {
          LOG.usingExistingDeployment();
          deployment = getExistingDeployment(commandContext, deployment.getName());
        }

 protected void deploy(DeploymentEntity deployment) {
    deployment.setNew(true);
    Context
      .getCommandContext()
      .getDeploymentManager()
      .insertDeployment(deployment);
  }

  public void insertDeployment(DeploymentEntity deployment) {
    getDbEntityManager().insert(deployment);
    createDefaultAuthorizations(deployment);

    for (ResourceEntity resource : deployment.getResources().values()) {
      resource.setDeploymentId(deployment.getId());
      getResourceManager().insertResource(resource);
    }

    Context
      .getProcessEngineConfiguration()
      .getDeploymentCache()
      .deploy(deployment);
  }

接着执行的是deploy方法实际执行的是子类BpmnDeployer.transformDefinitions 进行数据解析

  protected List<ProcessDefinitionEntity> transformDefinitions(DeploymentEntity deployment, ResourceEntity resource, Properties properties) {
    byte[] bytes = resource.getBytes();
    ByteArrayInputStream inputStream = new ByteArrayInputStream(bytes);

    BpmnParse bpmnParse = bpmnParser
        .createParse()
        .sourceInputStream(inputStream)
        .deployment(deployment)
        .name(resource.getName());

    if (!deployment.isValidatingSchema()) {
      bpmnParse.setSchemaResource(null);
    }

    bpmnParse.execute();

    if (!properties.contains(JOB_DECLARATIONS_PROPERTY)) {
      properties.set(JOB_DECLARATIONS_PROPERTY, new HashMap<String, List<JobDeclaration<?, ?>>>());
    }
    properties.get(JOB_DECLARATIONS_PROPERTY).putAll(bpmnParse.getJobDeclarations());

    return bpmnParse.getProcessDefinitions();
  }

最终会走到

protected void parseRootElement() {
    collectElementIds();
    parseDefinitionsAttributes();
    parseImports();
    parseMessages();
    parseSignals();
    parseErrors();
    parseEscalations();
    parseProcessDefinitions();
    parseCollaboration();

    // Diagram interchange parsing must be after parseProcessDefinitions,
    // since it depends and sets values on existing process definition objects
    parseDiagramInterchangeElements();

    for (BpmnParseListener parseListener : parseListeners) {
      parseListener.parseRootElement(rootElement, getProcessDefinitions());
    }
  }