Is there a way to enable high availability for jira server without data center?

[Sree Divya Bonagiri]

We would like to have failover for jira Server. Is there a way to do that without data center?

[Nic Brough -Adaptavist-]

No.

A very big part of high availability is effectively "clustering" - two or more servers doing the same work so that if one fails, the other can pick up the sessions people have with the dying one and carry on.

Another big part of it is having all nodes looking at a single source of data.

Jira Server does not cluster.  Jira server requires single exclusive access to the data source  (you can't run a second Jira with the same database)

[Sree Divya Bonagiri]

Can we have 2 instances and keep them in sync by any means?

[Nic Brough -Adaptavist-]

No.

The best you can do is an active/passive failover, which means downtime while you fully shut down the dying node and bring up the failover node.

[Sree Divya Bonagiri]

Can we have 2 instances with shared application data folder and database?

[Nic Brough -Adaptavist-]

Please read the last sentence of the original answer.

[Eric Seibert]

So you have a second node that is a complete clone and synchronized via rsync or some other method, and you have a second database that is synchronized by a tool. That's what it sounds like you are saying.

I've read some people have a node that is synchronized with confluence installed but not running.  Which has the same home and install directories as well as a shared home directory  ( similar to data_center), and  when the first one fails the load balancer shifts to the second one with HA proxy or some such tool and a CI/CD tool starts the second confluence instance, while the first instance is being repaired... yes there is downtime as the instance isn't spun up immediately, but it minimizes the downtime substantially while the first instance is worked on.

Then again I have heard this but never seen a passive failover work because once the first node is fixed it causes cluster panic and shuts both instances down...

the short answer is you need to get data center or have a second server that a load balancer and  nginx/httpd config need to be the same sharting certs with some sort of strange synchronization process

[Nic Brough -Adaptavist-]

>never seen a passive failover work because once the first node is fixed it causes cluster panic

The trick there is that you should not be clustering the nodes.  They are not a cluster, they are an active/passive failover.  Don't cluster them, you won't get a cluster panic.

[dekke046]

It is interesting to read the responses from Nic.

I think we all understand there is no motivation to facilitate Jira being able to work with loadbalancing as long as it is not a data center license, this despite the question of it is technically even possible or not, it would simply not contribute to the commercial business model. :-)

Maarten

[Nic Brough -Adaptavist-]

You're right about the commercial side.  Some vendors struggled for years to try to provide clustering in Jira and consistently ran into "it does not work" for four big consistent blocking reasons.   (I can only remember two reliably - sessions and shared storage being too slow for some of the non-database data stores), and no-one managed to solve the index inconsistencies or clashing writes adequately.

Atlassian came up with "Data Centre" to get past those problems.  The problem was not "do not cluster" in general, but very much "The platforms Atlassian stuff runs on do not support the functions that Atlassian stuff needs to be able to cluster (without sacrificing reliability, data integrity and/or speed)"

So, Atlassian poured a lot of money into new architecture that does cluster, and made it work for most of us.  But the tech is more complex, harder to support, and has significant impact on users and vendors (hence most vendors spending a LOT of time updating and then proving that their code deserved the "works on DC" sign-off from Atlassian).  

TLDR:  It's nowhere near as simple as just "load balancing".  DC is inherently more expensive than Server, so Atlassian ask for more money.