Create
cancel
Showing results for 
Search instead for 
Did you mean: 
Sign up Log in

BitBucket Data Center on AWS - Why the nodes are placed only in one AZ ?

Karthikeyan Kaliappan August 6, 2020

In this architecture diagram - 

https://aws.amazon.com/quickstart/architecture/bitbucket/ 

I see that the BitBucket nodes are placed only on one AZ.

Ideally in a HA setup, the nodes should be installed in both Availability Zones ... Am I right?

Can someone help to clarify this?

1 answer

0 votes
Christian Glockner
Atlassian Team
Atlassian Team members are employees working across the company in a wide variety of roles.
August 11, 2020

Hi Karthikeyan,

The Hazelcast component of Bitbucket Server, which handles node-to-node communication, depends on low latency, and since only having all nodes in the same AZ guarantees low latencies, that's why all nodes are in the same AZ.

Cheers,

Christian

Premier Support Engineer

Atlassian

Karthikeyan Kaliappan August 17, 2020

Thanks Chris. Appreciate your help.

Please help me understand ... what role Hazelcast component play in BitBucket server, and why there is a need for an inter-node communication, within a cluster?  

Regards.

Christian Glockner
Atlassian Team
Atlassian Team members are employees working across the company in a wide variety of roles.
August 17, 2020

Hi Karthikeyan,

 

As I said, Hazelcast is used for inter-node communication. That communication is needed for various things, such as caches, synchronization of settings, identifying which node is the master, etc.

Cheers,

Christian

Premier Support Engineer

Atlassian

Karthikeyan Kaliappan August 17, 2020

I got it. It now makes sense.  

We are actually interested to host all 3 products - BitBucket, Jira and Cloud Data centers on AWS ... and roll it out enterprise wide.

Wondering whether we can place all the 3 clusters in the same AZ and build it out together. Does it make sense?

Regards.

Karthikeyan Kaliappan August 17, 2020

I just got a response from our AWS architect. 

Since HA is our prime goal, with zero-down-time, he suggest this:

"Multi AZ have a little impact on the performance as compared to single AZ but it will be very minor (sub-millisecond). However, only extreme-low-latency use cases fit the single-AZ model. Thus, you can consider spanning the nodes across AZs to build a HA solution"

So I would like to double-check with Atlassian, if there is any "risk" involved in splitting the BitBucket nodes across multi-AZ in AWS, to accomplish HA?

Would this cause any Git transaction failure / mis-behavior?

Regards,

Karthik

Christian Glockner
Atlassian Team
Atlassian Team members are employees working across the company in a wide variety of roles.
August 17, 2020

Hi Karthik,

Bitbucket Server will only work in environments where you can guarantee very low latency, which is why we only support environments where all nodes are in the same physical location - and different availability zones typically aren't in the same location.

You can of course try it - chances are it will work, but be prepared to move everything to same AZ after all in case you are experiencing problems.

Cheers,

Christian

Premier Support Engineer

Atlassian

Suggest an answer

Log in or Sign up to answer
TAGS
AUG Leaders

Atlassian Community Events