This is the custom magda config used for running search.data.gov.au. It shows an example of how you can run a high-availability configuration of Magda under GKE.
Keep in mind that Data.gov.au uses a number of services that exist outside Kubernetes and are a bit fiddly to set up - in particularly it uses Let's Encrypt for HTTPS through its domain hosted on Route53 in AWS, a GCE Ingress (which should be automatically provisioned by Google Cloud but often needs some nudging via the console) and Google Cloud SQL postgres (which requires the right secrets to be set up for both a service account and a database user).
helm repo update
# download helm charts dependencies
helm dep up ./chart
helm upgrade -n dga-prod magda ./chart --timeout 30000s --install -f config.yaml
Magda version should be specified in deploy chart Chart.yaml dependencies section.
Upgrades with improvements & bug fixes
Minor upgrade with improvements & bug fixes
This upgrade includes dataset indice version bump, which will trigger a full reindex.
To prevent users from partial result & minimise the downtime, you can set .Values.datasetsIndexVersion to previous version 49 until reindex is completed.
also upgrade csw-connector to v2.0.2
Minor upgrade
This upgrade includes dataset indice version bump, which will trigger a full reindex.
To prevent users from partial result & minimise the downtime, you can set .Values.datasetsIndexVersion to previous version 48 until reindex is completed.
v1.2.0 comes with Elasticsearch upgrade.
More upgrade / migration info can be found from: here
v1.1.0 requires upgrading Google Cloud SQL to postgreSQL 13. We used Google DMS for this job. Steps of upgrading Google Cloud SQL to postgreSQL 13 using Google DMS can be found from here.
v0.0.60 update elasticsearch dataset indice version to 47, which will trigger a full reindex.
To prevent users from partial result & minimise the downtime, you can set .Values.datasetsIndexVersion to previous version 46 until reindex is completed.
In order to minimise the downtime during the migration, we skipped the version 0.0.57 and jumped to RC release 0.0.58-rc.3.
Since 0.0.57, we started to use local deploy chart based approach to deploy Magda. We defined Magda components as dependecies in file Chart.yaml. You can specify the version required in the same file as well.
As connectors are moved out from Magda main repo and serves as external extensions now, to add a new connector, you will need:
- add the required connector type as a dependecy in Chart.yaml and assign an alias to it
- add connector config in config.yaml
During the upgrade, we also upgraded cert-manager to the latest version 1.02.
This is a complicated one. Add
image:
tag: 0.0.55-RC1... to the config for search-api AND registry-api, and under tags:, change indexer to:
indexer: false
before running helm upgrade. This is because both the registry db and search indices have changed enough not to be backwards compatible, so we want to keep running the old versions of both APIs while the helm upgrade takes place. If the indexer is allowed to run during this process, it'll create new empty indices and then quit because it can't access the registry anymore, so turn it off before the upgrade.
The DB migration for the registry in this version is incredibly heavy - on staging it takes 5-10 minutes, during which the registry will be mostly unavailable.
Once helm finishes, immediately remove the image lines from registry-api, turn indexer back to true and run the helm upgrade again, this time with --no-hooks. This will change the registry over to the new version and allow the indexer to start.
Monitor the indexer logs - once it has finished creating new indices for both regions and datasets, remove the image lines from search-api also.
No elasticsearch upgrades this time, although there has been a slight tweak to the default email templates that you might want to incorporate - the question.html email now has a {{message.note}} field that explains why the default email address is getting a message in the event that an email can't be extracted from either the dataset or publisher contact point.
This also involves another elasticsearch index upgrade, so add...
image:
tag: 0.0.53-RC1... to the config for search-api before running helm upgrade. Look at the indexer logs, when it's finished indexing then remove those lines and run helm upgrade again.
Run sql-migrations/add-favicon.sql in the content-db first, then helm upgrade as usual.
This also involves an elasticsearch index upgrade, so add...
image:
tag: 0.0.52-RC2... to the config for search-api before running helm upgrade. Look at the indexer logs, when it's finished indexing then remove those lines and run helm upgrade again.
Nothing special this time, just helm upgrade as usual
This moves the readiness check for gateway, which doesn't propagate properly into the GKE health check. Before you start, turn the GKE health check up to 300 second intervals and 10 unhealthy threshold to give yourself some time.
This version includes a pretty serious regions index upgrade, which requires a new elasticsearch upgrade before the search-api OR indexer can be upgraded. So...
- Add to the config for
search-apiANDindexer:
image:
tag: 0.0.50-RC3-
Open the GCE health check page ready to quickly change it! To give yourself more time, change the failure threshold and timeout to as long as you can make it.
-
Run
helm upgradeas above. -
(while helm upgrade is running) Change the health check URL in GCE over to
/status/live. -
(while helm upgrade is running) Run
kubectl edit deployment search-apiand manually edit the readiness and liveness check urls to/v0/datasets. -
Wait for helm upgrade to finish.
-
Use
kubectl port-forward <kibana pod id> 5601to connect to kibana andGET /_cluster/healthto make sure everything's finished initializing. -
Remove the image tag from the
indexerconfig, runhelm upgradeagain with--no-hooksand wait for it to finish indexing the new regions index -
Remove the
image.tagdeclaration from the config fromsearch-apiandhelm upgradewith--no-hooksagain -
Use
kubectl editto change the readiness/liveness check for search-api back to their usual values in search-api/deployment.yaml -
Restore the normal values in the health check for failure threshold and timeout (should be 70 seconds and 10 failures).
Before helm upgrade: In the content db:
- Run https://github.com/magda-io/magda-config-data.gov.au/blob/master/sql-migrations/add-email-templates.sql in order to create the email templates
- Run https://github.com/magda-io/magda-config-data.gov.au/blob/master/sql-migrations/add-navigation-items.sql in order to create navigation menu items
- Run https://github.com/magda-io/magda-config-data.gov.au/blob/master/sql-migrations/add-static-pages.sql to add static pages
- Run https://github.com/magda-io/magda-config-data.gov.au/blob/master/sql-migrations/add-home-content.sql to add home content items
- Run https://github.com/magda-io/magda-config-data.gov.au/blob/master/sql-migrations/add-third-party-js.sql to add the old third party js code.
- Run https://github.com/magda-io/magda-config-data.gov.au/blob/master/sql-migrations/add-initial-translations.sql to make sure the content of the suggest dataset email doesn't change and the app name stays the same
- Run https://github.com/magda-io/magda-config-data.gov.au/blob/master/sql-migrations/add-footer-config.sql to change up footer items
Run kubectl apply -f role-binding.yaml to make the admin api work.
This version includes a datasets index upgrade, so add...
image:
tag: 0.0.49-RC3... to the config for search-api before running helm upgrade. Look at the indexer logs, when it's finished indexing then remove those lines and run helm upgrade again.
This also makes the change from the web client getting content from the CMS, so you'll need to do that for web-server too.
Because the CSS now comes from the content-api, we need to insert the correct SQL before upgrading or the page will be unstyled until the upgrade is finished. You can find a script to insert it in sql-migrations/add-stylesheet-to-content-api.sql. Run this against the content-db.
- Delete stuff that wasn't managed by helm that we'll need to replace with stuff that now is.
kubectl delete cronjobs --all
kubectl delete configmap connector-config- Add secrets for config-db
kubectl edit secret db-passwords3 Add the magda chart repo to helm
helm repo add magda-io https://charts.magda.io
# "magda-io" has been added to your repositories- Install magda. The
--develflag allows us to install the latest RC version
cd terraform/magdagcloud iam service-accounts keys create key.json --iam-account=$SERVICE_ACCOUNT_EMAILYou will now have a key.json file in terraform/magda, containing a private key. We suggest you put this somewhere safe like a password manager.
DO NOT CHECK IT INTO SOURCE CONTROL.
-
No need to delete es-data pods manually, helm now does that 🎉.
-
No need to do anything with the cronjobs or cron config, it's all in helm now! 🎉