IKO Plus: Database Management, Migration, or Day 1 Seeding your Mirrored IrisCluster

Mirror Your Database Across the Galaxy with Seeding

Hello cpf fans!  This distraction I used the “seed” capability in IRIS to provision an entire IrisCluster mirror, 4 maps wide with compute starting from an IRIS.DAT in a galaxy far far away.  This is pretty powerful if you have had a great deal of success with a solution running on a monolithic implementation and want it to scale to the outer rim with Kubernetes and the InterSystems Kubernetes Operator.  Even though my midichlorian count is admittely low, I have seen some hardcore CACHE hackers shovel around DATS, compact and shrink and update their ZROUTINES, so this same approach could also be helpful shrinking and securing your containerized workload too.  If you squint and feel all living things around you,  you can see a glimpse of in place (logical) mirroring in the future as a function of the operator and a migration path to a fully operational mirrored Death Star as the workload matures.

Mission

Ill show you how I built an IrisCluster that pulls an IRIS.DAT file from an arbitrary location via an initContainer and uses it to “seed” a mirrored database throughout its topology.

I used a repo that I visit at least annually @Guillaume Rongier ‘s repo InstallSamples which features a committed IRIS.DAT, there is also another one that appropriately puts ENSDEMO back on your systems too, but I kept it simple adhering to my prescription of Focusyn.

https://media.githubusercontent.com/media/grongierisc/InstallSamples/ref…

🌱 This will be the seed for the IrisCluster, and I expect it to be production grade mirrored and like deployable on a Friday if you know what I mean.

Kubernetes Cluster (Kind)

So I provision a quick kind cluster, give it 5 worker nodes, install Cilium as the default CNI and install the InterSystems Kubernetes Operator.  Waited for it to complete, then when it did, I labelled the nodes, one per Star Wars planet, including one from the Outer Rim.

I now have a happy Kubernetes cluster, nodes are ready and labelled as planets=(tatooine coruscant hoth naboo endor)… warp speed.

And an InterSystems Kubernetes Operator  parked in orbit around the forest moon of Endor.

Then I had R2 declare the IrisCluster topology…

IrisCluster

Here is is in its entirety, but we will break it down, starting with the init container.

init-container

Our init-container has one job, and is going to retrieve our seed database from github, and appropriate it with a mount and hand it off to our IRIS pods so it can be referenced from a POSIX path.

...
          initContainers:
            - name: init-grongierisc-samples
              image: debian:bookworm-slim
              command:
                - sh
                - -c
                - |
                  echo "Installing getting Samples from @grongierisc's repository https://github.com/grongierisc/InstallSamples/tree/master ..."
                  #set -euo pipefail
                  apt-get update
                  DEBIAN_FRONTEND=noninteractive apt-get install -y curl
                  cd /airgapdir
                  curl -L https://media.githubusercontent.com/media/grongierisc/InstallSamples/refs/heads/master/samples/IRIS.DAT --output IRIS.DAT
                  #curl -L http://192.168.1.231:8080/IRIS.DAT --output IRIS.DAT
                  chmod -R 777 /airgapdir
                  chown -R 51773:51773 /airgapdir
              volumeMounts:
                - name: airgapdir
                  mountPath: /airgapdir
...

The database will show up at /airgapdir/IRIS.DAT in the IRIS pods.

Data

💫The seeding comes in at this block in the specification, this is telling the operator to weild the force

      irisDatabases:
        - name: SAMPLES
          directory: /irissys/data/IRIS/mgr/SAMPLES
          mirrored: true
          ecp: true
          seed: /airgapdir/
          logicalOnly: false
      irisNamespaces:
        - name: ENSDEMO
          routines: SAMPLES
          globals: SAMPLES
          interop: true

I enabled mirroring, turned ecp to true as we will be including a compute node, and pointed it at our seed directory of the database.  Under the hood, IKO is generating a cpf for merge in the Actions directive, which is really accountable for the heavy seeding capability.

[Actions]
ModifyService:Name=%service_ecp,Enabled=0
CreateDatabase:Name=iriscluster,Directory=/irissys/data/IRIS/mgr/iriscluster
CreateNamespace:Name=IRISCLUSTER,Globals=iriscluster,Routines=iriscluster
CreateDatabase:Name=SAMPLES,Directory=/irissys/data/IRIS/mgr,Seed=/airgapdir
CreateNamespace:Name=ENSDEMO,Globals=SAMPLES,Routines=SAMPLES,Interop=1

Compute

I added a single replica of an ecp node to the equation, a clone if you will.

    compute:
      image: containers.intersystems.com/intersystems/irishealth:2025.1
      ephemeral: true
      replicas: 1
      preferredZones:
        - endor

mirrorMap

The mirrormap includes the following roles:

mirrorMap: primary,backup,drasync,rwrasync
mirrored: true

Zones

To stay in step with the Star Wars theme Ill lock our mirror topolology to the labelled nodes.

      preferredZones:
        - tatooine
        - coruscant
        - hoth
        - naboo

Note that we already set a zone for the compute node on Endor.

Apply

kubectl apply -f iriscluster.yaml --kubeconfig ikoseed.kubeconfig
iriscluster.intersystems.com/ikoplus-seed-sweenx12 created
secret/license-key-secret unchanged
secret/webgateway-secret unchanged
secret/containers-pull-secret unchanged
secret/cert-secret configured
configmap/seed-before-script unchanged
secret/webgateway-secret unchanged
configmap/seed-after-script unchanged

I waited a few minutes and it the IrisCluster came to life…

The IrisCluster role view…

And a view from Hubble…

I pulled up Mirroring Monitor, to see if our seeded database “SAMPLES” in Namespace “ENSDEMO” made it to the mirror party

🎉 It DID!

Then I inspected the result of the clone operation on the pod itself to make sure the DAT is in all the right places (podspec, filesystem, DATADIR).

🎉 They ARE!

And just to check the topology, lets see if the single replica ECP client is connected…

🎉 Totally IS!

Attestation

I am a little weirded out by these force powers and hand waving, back in the day I seemed to have to square dance mounting/shuffling around the DAT with PVC swings, SFTP, ssh, whatever, and even wrote sscp with @Eduard Lebedyuk to do the task over the superserver.

So I write a global out in the on the Primary mirror on the IrisCluster…

Then loop over the topology to see if it it got mirrored appropriately across the galaxy.

Its a little sloppy, Wado did not give me the day off today, but the results are good…

The force seems to be with us on:

✅ primary (persisted)

✅ backup (persisted)

✅ drasync (persisted)

✅ rwrasync (persisted)

✅ compute (accessible)

🎉