You are viewing a plain text version of this content. The canonical link for it is here.
Posted to notifications@skywalking.apache.org by GitBox <gi...@apache.org> on 2018/06/17 19:47:36 UTC

[GitHub] gamefundas opened a new issue #1364: Recommended architecture for micro services

gamefundas opened a new issue #1364: Recommended architecture for micro services
URL: https://github.com/apache/incubator-skywalking/issues/1364
 
 
   Please answer these questions before submitting your issue.
   
   - Why do you submit this issue?
   - [X] Question or discussion
   - [ ] Bug
   - [ ] Requirement
   - [ ] Feature or performance improvement
   
   ___
   ### Question
   - What do you want to know?
   
   What is the recommended production configuration for micro service architecture? 
   I have about 100 services hosted on a VPC where in we are looking to bring an APM to monitor and manage performance for all services. When I look at Skywalking cluster setup, my understanding is the the collectors serve as backend for all the metrics and data working in conjunction with zookeeper for stage and ES for persistence. However I have a problem understanding the overall approach with java agents when it comes to decoupling the application instances from the collector themselves. I am not sure if the java agents themselves should talk to the central collector/s as then we need to know their IP's before configuring the agents (agent.conf) 
   
   This becomes a major roadblock as at any point in time the APM gets refreshed or redeployed I don't intent to restart all my applications to pick the new collector. We don't allow any static ip's configuration in our hosted AWS environment. Also DNS CNAMES have proved to be a problem as the clients typically cache them. This restricts a lot on the client/application side as this architecture is not flexible to constantly changing environments. I can think of launching a collector in every instance but not sure if the whole dependency of zookeeper and ES on every application node makes sense.
   
   Unless I misunderstand any part of Skywalking it seems we need a light weight client/agent collector configured on every application instance that can simply allow local connections from any application (Java, SB, Nodejs, Go, Python etc). These will be responsible to buffer and send data to central collectors. In case of a central infra outage only the client/agent collector needs to be restarted to pick new IP's. The application in itself is not impacted from this outage and doesn't require a restart.
   
   - Which company or project?
   Researching APM solution for various projects - Currently POC
   
   ___
   ### Requirement or improvement
   - Please describe about your requirements or improvement suggestions.
   As suggested above an agent collector will help loosely couple the application with the central APM infra. Tools like consul have a similar architecture and has worked well in my experience.
   
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services