Naming write performace

[nanamikon]

Hi, I want to share some information of pressure measurement

• nacos version 0.3
• 3 peers (cpu 16, mem 64G, ssd 2T)
• screen
  Changing the ip list of one service repeatedly

    CountDownLatch countDownLatch = new CountDownLatch(count);
    for (int i=0; i<count; i++) {
        executor.execute(() -> {
            try {
                long startTime = System.currentTimeMillis();
                String serviceName = "test.mysql";
                Service service = new Service(serviceName);
                service.setHealthCheckMode("none");
                Instance instance = new Instance();
                instance.setIp("127.0.0.1");
                instance.setPort(8080);
                namingService.registerInstance(serviceName, "127.0.0.1", 8080);
                System.out.println("cost time: " + (System.currentTimeMillis() - startTime));
            } catch (NacosException e) {
                System.out.println("error! " + e.getErrMsg());
            } finally {
                countDownLatch.countDown();
            }
        });
    }

• summary
  tps 300+ ， increase concurrency has no effect， I think it is not not suitable for large cluster

• key problem
  1、Two file (meta.properties， com.alibaba.nacos.naming.domains.meta.test.mysql) had to be wrote sync one by one. This step could cost 3ms ~ 4ms at lease.
  

2、One global lock， leader had to handle the request one by one

• suggestion
  refer to https://ramcloud.stanford.edu/~ongaro/thesis.pdf 10.1 ~ 10.3
  What about batch request or request process pipline？