Intersystems Mirroring of code databases

Discussion

Erik Svensson · Sep 18, 2020

Hello!

First of all, let me state that I am no senior InterSystems expert.

In my organization, we have a HealthShare Health Connect setup where each namespace has one code database and one data database, which are both actively mirrored. We have two nodes in the mirror.

We had a controlled failover last night to make sure that the backup node works as intended, which it didn't. It turned out that we had only deployed code onto the primary node in several namespaces causing errors with missing classes after the failover. So it seems that each time you deploy productions, you have to manually deploy it to both instances (the primary and failover). That makes me wonder:

What is actually mirrored when you mirror a code database
1. Obviously not new classes
2. changes to existing classes?
3. settings on the production adapters?
4. something else?
How do you guys go about deploying new code?
1. Are you utilizing some kind of automation tool to keep the mirrored nodes consistent regarding code and versions of code?
2. Are you just manually deploying to each node and have good routines doing it?
Or do we have some kind of faulty setup which makes this not work as intended?

I don't think our setup is faulty, I think we just missed this a bunch of times which makes me want to abstract this to a way where you deploy to something that deploys the same code to both nodes.

An example: We have 3 environments (production, QA and test), for each of QA and prod, we receive webservice requests from 2 different networks, an internal network and an external one. For each network, we have a pair of web servers running httpd with web gateway. This makes 4 web server hosts for each production and qa environment and in the test environment, we have slimmed this to only have the one pair. Totally 10 web servers. This is bound to be time consuming to maintain and create inconsistency and details that is not done exactly the same between the hosts if you are not extremely thorough, if you would do it manually. So we use Ansible. I have made a playbook and a set of configs for each environment and each network type so each pair is treated exactly the same and the playbook is always used to deploy changes and keep consistency.

I would like to achieve something similar with deploying code to our HeathConnect mirrored instances.

How do you guys do it?

How do you deploy code to each node in a mirrored intersystems setup?

Robert Cemper · Sep 18, 2020

#1) check that ALL code Dbs are part of your Mirror.
There is a fair chance that not all code you use is in a single code DB but is mapped to other DBs.
I'm not talking about implicitly mapped pieces like all System and %* Utilities.
#2) If you use code mapping It is highly important that Package mapping AND routine mapping go hand in hand
#3) Whatever Mirror is synchronizing is based on Global Journal. So also all code DBs require Journalling.
Since every routine or class whether deployed or not is stored in some global.

But my personal preference is not to apply Mirror to code DBs.
Mainly to control the point in time when a Change/Update happens.
I'm a fan of the Red Fire Button and like to control the moment of essential changes

2 0

Dan Pahnke · Sep 18, 2020

Hi Robert,

Seeing as how I'm fighting the same issue (keeping mirror members synched, not code DBs though), what does the "Red Fire Button" refer to?

Thanks,

Dan

0 0

Hi @Dan Pahnke
The "Red Fire Button" is a synonym I used over the years with my various teams for an action/decision
that should not be taken by a single person but follows (at least) the 4-eyes-principle.

Inspired by an endless number of Airforce fighting movies from Hollywood and
the old but still incredible song from The Dubliners.
And its best cover version

Dmitry Maslennikov · Sep 18, 2020

Definitely, something wrong in the configuration. Code in InterSystems in fact is no different from any other data stored there. So, you may have some wrong mappings, or store some of your code in %SYS.

I have a configuration with mirroring + ECP, and it works perfectly, I don't even care which of the nodes is primary, and can switch it any time, with no issues. And I have even more than one Code databases, and more than 20 Data databases. Mirroring Nodes works on 2018.1 while ECP Application Servers on 2012.2, with no issues.

If you have some doubts about your configuration, you can ask for help through WRC, or we can help you with it, we can review your settings, and say what actually happened and how to solve it

3 0

Erik Svensson · Oct 19, 2020

The issue was that the general package mappings we had in %ALL namespace was not mirrored and they were not mapped in the backup node.