Replying to this old thread just to notice that the issue was finally solved:
https://issues.apache.org/jira/browse/SYNCOPE-1628 Fix will be available with upcoming Syncope 2.1.10. Regards. On 04/04/18 07:32, Elena Hong wrote:
Thanks to your help :D 2018년 4월 3일 (화) 오전 1:29, Francesco Chicchiriccò <ilgro...@apache.org <mailto:ilgro...@apache.org>>님이 작성: Hi Elena, my personal congrats, it seems you've got most of the picture :-) If your HA setup is correct, in fact, writing any data via REST on node A and then read the same data from node B is perfectly fine. When such data are in fact related to connector configuration (or even resource configuration, if you are using override), things are effectively a bit different, because the Spring bean associated to each connector gets automatically refreshed only on the node where the REST create / update was sent to. So, if connector update was sent to node A, provisioning tasks occurring on node B will still have the old configuration, you're right. Workarounds: * invoke POST /connectors/reload on node B - this will make all connector Spring beans to refresh their configuration from the underlying db * restart the Java EE container on node B * disable Quartz jobs execution on node B [1] The actual fix would be to implement a custom RemoteCommitListener [2] which triggers the connector's Spring bean refresh on node B. Let me finally add that normally this issue is not very important because the connector configuration is rather stable as soon as the deployment reaches out HA environments (as it was fine-tuned in lower environments). Regards. [1] http://syncope.apache.org/docs/reference-guide.html#quartz <http://syncope.apache.org/docs/reference-guide.html#quartz> [2] http://openjpa.apache.org/builds/2.4.2/apidocs/org/apache/openjpa/event/RemoteCommitListener.html <http://openjpa.apache.org/builds/2.4.2/apidocs/org/apache/openjpa/event/RemoteCommitListener.html> On 02/04/2018 11:05, Elena Hong wrote:oops, I have mistake in my question. sorry. I have trouble when do provisioning not call read API. As you say, read connector API is working well. It loads data from DB(Of course My A and B server using same DB). But during provisioning, Syncope loads data from spring bean factory is managed in instance's InMemory(I guessed). When run provisioning task, call doExecute method from AbstractProvisioningJobDelegate.java And it loads connector data via ConnectorFactory.java 's getConnector method. I can see getConnector method in ConnectorManager.java. It loads data from beanFactory. @Override public Connector getConnector(final ExternalResource resource) { // Try to re-create connector bean from underlying resource (useful for managing failover scenarios) if (!ApplicationContextProvider.getBeanFactory().containsBean(getBeanName(resource))) { registerConnector(resource); } return (Connector) ApplicationContextProvider.getBeanFactory().getBean(getBeanName(resource)); } So, After I update connector via Syncope server A , It updates DB and own spring bean. Then read connector API works well and I can see the updated connector data from management console. But server B's bean is not updated yet. In this case, If provisioning run in Server B, B has 'old' connector data. 2018-04-02 11:03 GMT+09:00 Elena Hong <aheer...@gmail.com <mailto:aheer...@gmail.com>>: How can each syncope servers in high available environment share connector which saved as spring bean at inmemory? * My environment. I set high available with two syncope servers called A, B and nginx. * My problem 1. I call connector update api to nginx. 2. nginx call syncope server A, and update connector 'new' data in DB and spring bean. 3. I call connector read api to nginx. 4. nginx call syncope server B, then B returned 'old' data at spring bean. How can I solved it..? give me a tip please..
-- Francesco Chicchiriccò Tirasa - Open Source Excellence http://www.tirasa.net/ Member at The Apache Software Foundation Syncope, Cocoon, Olingo, CXF, OpenJPA, PonyMail http://home.apache.org/~ilgrosso/
