What's HA about Jira Datacenter?

[Jonas]

Hi,

We were sold Jira DC under the pretence that it was a HA solution, but after working with it for half a year i have a problem calling it HA, or i simply have not found the functions that would make it call it High Available.

A few how-do-i's:

- How do i evict all user sessions established for a node to other nodes without them losing their session data, if i need to service a node? So far i can only shut down the node, but if a user was about to create a ticket and have filled out all forms but not not submitted the form, this data will be lost.

In lack of this very basic functionality, it would be great to know what users are on one node, so i can warn them manually to get off the node, but i can't find that functionality either? Is there a "List users on node X" functions i am missing?

- A user reports a internal server error 500 on Node1. How do i switch my session to that node, without open 10 incognito browsers and login until i by random hit the node?

IMO a rolling restart of a DC cluster will instead of losing a users data once, make them lose their data repeatedly because:

User A is on Node1. I restart node1. User loses their data, but the loadbalancer will soon move them on to node2. They start working there and then it's turn to restart node2. they once more lose their data and can now either be moved to node3, or node1 if already back up. How is this better than all users losing their sessions once? 

Beside if a server dies, users can be falled over to another node (and lose their data) but be back in a new session on another server, but that's basically the only "HA" thing about Atlassian DataCenter products. 

Kind regards

Jonas Andersson

[Jonas Andersson]

@atlassian Any feedback on my questions above?

[Carlos Garcia Navarro]

Hi @Jonas ,

I hear you. I don´t know if a specific configuration will address all the issues that you described but I was wondering if you had read this article: https://confluence.atlassian.com/enterprise/high-availability-guide-for-jira-288657149.html

https://jira.atlassian.com/browse/JRASERVER-61196

By doing rolling restarts you ensure that not all the servers will be down at the same time,so new clients can start transactions without having to wait.

[Jonas]

Carlos, This doesn't evict user sessions to other servers, and does not circumvent the problems i describe. It will indeed cause the user to lose it's sessions and filled out fields when user submitts, no matter if the server went down and came back up, as sessions are not retained through restarts. In worst case user will submit fields, loadbalancer will recognize node is down, haproxy will reroute it to another node, which will demand login and will land the user on the dashboard. Should i shutdown another node, user risks being on this node too, and might end up risking losing this session too, when that node is restarted. 

First link does not contain anything regarding my question, and i have seen it before.

The jira you linked is about database fallovers (on the backend) and once more not related to what i am asking atlassian here.

Lets see, maybe atlassian indeed have something for this, but i followed all best practices setting these clusters up, and all my testing shows exactly what i describe above.

Thanks for the response however.

Kind regards.

Jonas