DEV Community

Arseny Zinchenko
Arseny Zinchenko

Posted on • Originally published at rtfm.co.ua on

Kubernetes: running SQL migrations with Kubernetes Job and Helm hook

#kubernetes #tutorial #sql #devops

We have a project running in Kubernetes that needs to run SQL migrations during deployment.

To run migrations need to clone a Github repository and run actually migrations stored in it.

Currently, this is done with Kubernetes initContainers , and there are two of them — the first one with git clones the repository with migrations files into a Kubernetes Volume, and then another one with sql-migrate runs those migrations from this shared volume.

Still, there are a few issues with this approach:

  1. each time when we are running a new pod it runs its initContainers and migrations
  2. if a Deployment spin up a couple of pods — each will run migrations
  3. if migrations will run for w while and will not respond to Kubernetes readiness — it could be killed without finishing migrations

To avoid all the above let’s reconfigure the process by using a Kubernetes Job to run only one pod and by adding Helm Hooks to trigger migrations during deployment.

Note: the kk here is an alias for the kubectl.

Preparation

Docker image

At first, let’s create our own Docker image with blackjack and git and https://github.com/rubenv/sql-migrate.

Create a Dockerfile:

FROM golang:alpine AS builder
RUN apk add --no-cache git gcc g++
RUN go get -v github.com/rubenv/sql-migrate/sql-migrate
RUN mv /go/bin/sql-migrate /bin/sql-migrate
Enter fullscreen mode Exit fullscreen mode

Build and push it:

$ docker build -t projectname/sql-migrate-git .
$ docker push projectname/sql-migrate-git
Enter fullscreen mode Exit fullscreen mode

Git authentification

The second thing is the Github authentification.

At this moment our git-container authenticates via an RSA key which is held in a Kubernetes Secrets, then it passes to a pod via an environment variable from where it’s taken by a bash script /opt/git/git.sh which is used to create a key-file /root/.ssh/id_rsa inside of the container, and this key finally is used to authenticate.

The initContainers in our Deployment currently looks like the next:

...
      initContainers:
      - name: git-clone
        image: projectname/git-cloner
        env:
        - name: SSH_PRIVATE_KEY
          valueFrom:
            secretKeyRef:
              name: git-ssh-key
              key: id_rsa
        - name: REPOSITORY_URL
          value: {{ .Values.git.repo }}
        - name: GIT_BRANCH
          value: {{ .Values.git.branch }}
        command: ['sh', '-c', '/opt/git/git.sh']
        volumeMounts:
          - name: git-volume
            mountPath: "/git"
            readOnly: false
      - name: init-migration
        image: fufuhu/sql-migrate:latest
        command: ['sh', '-c', 'while [! -d /git/db/migrations]; do sleep 2; done && sleep 2; /bin/sql-migrate up -config=/config/config.yaml -env=main']
        volumeMounts:
          - name: migration-config
            mountPath: "/config/"
            readOnly: true
          - name: git-volume
            mountPath: "/git"
            readOnly: false
...
Enter fullscreen mode Exit fullscreen mode

A lot of steps, a lot of objects, complicated process.

Instead, let’s use a login and a Github token which will be passed into our container from environment variables, and then we can clone the repository via HTTPS.

Let’s test it:

~ # export GIT_AUTHUSER=backend-user
~ # export GIT_AUTHKEY=cdc***0fe
~ # git clone
[https://$GIT_AUTHUSER:$GIT_AUTHKEY@github.com/projectname-dev/backend-services.git](https://%24GIT_AUTHUSER:%24GIT_AUTHKEY@github.com/projectname-dev/backend-services.git)
Cloning into ‘backend-services’…
…
Receiving objects: 100% (5115/5115), 846.55 KiB | 1.30 MiB/s, done.
Resolving deltas: 100% (2826/2826), done.
Enter fullscreen mode Exit fullscreen mode

Nice, “It works” ©

SQL migrations in Kubernetes

Now, we can start writing a manifest file templates/appname-api-migrations.yaml to describe our Kubernetes Job which later will be triggered by a Helm hook.

Kubernetes Job

git clone

First, to make sure the Job is working — let’s write it without Helm variables and values, all pod’s environment variables will be set as plaintext values, and the action here will be the git clone for now:

apiVersion: batch/v1
kind: Job
metadata:
  name: "migration-job"
  labels:
  annotations:
spec: 
  backoffLimit: 0
  template:
    metadata:
      name: "migration-job-pod"
    spec:
      restartPolicy: Never
      containers:
      - name: db-migrations
        image: projectname/sql-migrate-git:latest
        command: ["/bin/sh", "-c"]
        args:
          - git clone --single-branch --branch develop [https://backend-user:cdc***0fe@github.com/projectname/backend-services.git](https://backend-user:cdc***0fe@github.com/projectname/backend-services.git) &&
            ls -l backend-services/db/migrations
Enter fullscreen mode Exit fullscreen mode

Here in the restartPolicy we've set to not to restart a container if it fails, as we'd like to see that migrations were failed, and the same in the backoffLimit=0 - to not to re-create a pod if it fails and just leave the job with the Failed status.

In the git clone a branch will be set from a Jenkins job, user and URL will be set in a values.yaml, and the authentication token will be kept with Helm secrets, later will move it to an environment variable.

Create the Job:

$ kk -n eks-dev-1-appname-api-ns apply -f appname-api-jobs.yaml
job.batch/migration-job created
Enter fullscreen mode Exit fullscreen mode

Check its logs:

$ kk -n eks-dev-1-appname-api-ns logs job/migration-job
Cloning into ‘backend-services’…
total 20
-rw-r — r — 1 root root 538 Oct 24 12:20 BS_1_init_schema.up.sql
-rw-r — r — 1 root root 180 Oct 24 12:20 BS_2_add_brand_field.up.sql
-rw-r — r — 1 root root 225 Oct 24 12:20 BS_3_alter_table.up.sql
-rw-r — r — 1 root root 194 Oct 24 12:20 BS_4_add_created_at_field.sql
-rw-r — r — 1 root root 272 Oct 24 12:20 BS_5_alter_table_nourishment_diet.up.sql
Enter fullscreen mode Exit fullscreen mode

The repository was cloned, migration fils now are accessible.

Check the pod’s status:

$ kk -n eks-dev-1-appname-api-ns get pod
NAME READY STATUS RESTARTS AGE
migration-job-f72vs 0/1 Completed 0 9s
Enter fullscreen mode Exit fullscreen mode

And the Job’s status:

$ kk -n eks-dev-1-appname-api-ns get job
NAME COMPLETIONS DURATION AGE
migration-job 1/1 2s 5s
Enter fullscreen mode Exit fullscreen mode

Now can proceed with the exact migrations process.

Secret

To run migrations we need to create a config file that will be stored in a Kubernetes ConfigMap, but in this file, a database’s password must be set.

It’s not a good idea to store it in plaintext in a ConfigMap and the file but sql-migrate allows us to use an environment variable in the file, check its documentation - https://github.com/rubenv/sql-migrate#as-a-standalone-tool

So, we will create a variable for the pod called $DB_PASSWORD, and will keep the actual password in a Kubernetes Secrets, and later in Helm, we will use the Helm secrets to store it encrypted in a chart's values.

Also in this Secrets, we will store a value for a $GIT_TOKEN environment variable to be used in the git clone command.

Still in the templates/appname-api-migrations.yaml add a Secret:

...
---
apiVersion: v1
kind: Secret
metadata:
  name: backend-db-password
type: Opaque
stringData:
  db_password: password
  git_token: cdc***0fe
Enter fullscreen mode Exit fullscreen mode

In the spec.containers.env of the Job add variables and update the git clone to use the $GIT_TOKEN variable:

...
      containers:
      - name: db-migrations
        image: projectname/sql-migrate-git:latest
        command: ["/bin/sh", "-c"]
        args:
          - git clone --single-branch --branch develop [https://backend-user:$GIT_TOKEN@github.com/projectnamev/backend-services.git](https://backend-user:%24GIT_TOKEN@github.com/projectnamev/backend-services.git) &&
            ls -l backend-services/db/migrations;
            cat /config/config.yaml
        env:
        - name: GIT_TOKEN
          valueFrom:
            secretKeyRef:
              name: backend-db-password
              key: git_token
        - name: DB_PASSWORD
          valueFrom:
            secretKeyRef:
              name: backend-db-password
              key: db_password
...
Enter fullscreen mode Exit fullscreen mode

ConfigMap

Next, create a ConfigMap to keep the /config/config.yaml content for the sql-migrate and use the $DB_PASSWORD variable in it:

...
---
apiVersion: v1
kind: ConfigMap
metadata:
  name: migration-config
data:
  config.yaml: |
    main:
      dialect: mysql
      datasource: backend-user:${DB_PASSWORD}@tcp(stage.backend-db3-master.example.com:3306)/dbname?parseTime=true
      dir: backend-services/db/migrations
      table: backend_services_migrations
Enter fullscreen mode Exit fullscreen mode

In the pod’s template of the Job add volumes, and in the spec.containers via volumeMounts mount the volume with the ConfigMap as the /config/config.yaml file.

The Job’s full manifest now is the next:

apiVersion: batch/v1
kind: Job
metadata:
  name: "migration-job"
  labels:
  annotations:
spec: 
  backoffLimit: 0
  template: 
    metadata:
      name: "migration-job-pod"
    spec:
      restartPolicy: Never
      containers:
      - name: db-migrations
        image: projectname/sql-migrate-git:latest
        command: ["/bin/sh", "-c"]
        args:
          - git clone --single-branch --branch develop [https://backend-user:$GIT_TOKEN@github.com/projectname/backend-services.git](https://backend-user:%24GIT_TOKEN@github.com/projectname/backend-services.git) &&
            ls -l backend-services/db/migrations;
            cat /config/config.yaml
        env:
        - name: GIT_TOKEN
          valueFrom:
            secretKeyRef:
              name: backend-db-password
              key: git_token
        - name: DB_PASSWORD
          valueFrom:
            secretKeyRef:
              name: backend-db-password
              key: db_password
        volumeMounts:
        - name: migration-config
          mountPath: "/config/config.yaml"
          subPath: "config.yaml"
          readOnly: true
      volumes:
        - name: migration-config
          configMap: 
            name: migration-config
            items:
            - key: "config.yaml"
              path: "config.yaml"
...
Enter fullscreen mode Exit fullscreen mode

Run it:

$ kk -n eks-dev-1-appname-api-ns apply -f appname-api-jobs.yaml
job.batch/migration-job created
secret/backend-db-password created
configmap/migration-config created
Enter fullscreen mode Exit fullscreen mode

Check:

$ kk -n eks-dev-1-appname-api-ns logs job/migration-job
Cloning into ‘backend-services’…
total 20
-rw-r — r — 1 root root 538 Oct 24 13:41 BS_1_init_schema.up.sql
-rw-r — r — 1 root root 180 Oct 24 13:41 BS_2_add_brand_field.up.sql
-rw-r — r — 1 root root 225 Oct 24 13:41 BS_3_alter_table.up.sql
-rw-r — r — 1 root root 194 Oct 24 13:41 BS_4_add_created_at_field.sql
-rw-r — r — 1 root root 272 Oct 24 13:41 BS_5_alter_table_nourishment_diet.up.sql
main:
dialect: mysql
datasource: backend-user:${DB_PASSWORD}@tcp(stage.backend-db3-master.example.com:3306)/dbname?parseTime=true
dir: backend-services/db/migrations
table: backend_services_migrations
Enter fullscreen mode Exit fullscreen mode

Good — the repository was cloned, the config file was created.

Running migrations

And now we can describe the migrations process, for now with the -dryrun option, and with the second command - check its status:

...
        args:
          - git clone --single-branch --branch develop [https://backend-user:$GIT_TOKEN@github.com/projectname/backend-services.git](https://backend-user:%24GIT_TOKEN@github.com/projectname/backend-services.git) &&
            ls -l backend-services/db/migrations &&
            cat /config/config.yaml &&
            /bin/sql-migrate up -config=/config/config.yaml -env=main -dryrun &&
            /bin/sql-migrate status -config=/config/config.yaml -env=main
...
Enter fullscreen mode Exit fullscreen mode

Run, and check its logs:

$ kk -n eks-dev-1-appname-test-migrations-ns logs job/migration-job
Cloning into ‘backend-services’…
total 20
-rw-r — r — 1 root root 538 Oct 24 14:02 BS_1_init_schema.up.sql
-rw-r — r — 1 root root 180 Oct 24 14:02 BS_2_add_brand_field.up.sql
-rw-r — r — 1 root root 225 Oct 24 14:02 BS_3_alter_table.up.sql
-rw-r — r — 1 root root 194 Oct 24 14:02 BS_4_add_created_at_field.sql
-rw-r — r — 1 root root 272 Oct 24 14:02 BS_5_alter_table_nourishment_diet.up.sql
main:
dialect: mysql
datasource: backnd-user:${DB_PASSWORD}@tcp(stage.backend-db3-master.example.com:3306)/dbname?parseTime=true
dir: backend-services/db/migrations
table: backend_services_migrations
+ — — — — — — — — — — — — — — — — — — — — — + — — — — — — — — — — — — — — — -+
| MIGRATION | APPLIED |
+ — — — — — — — — — — — — — — — — — — — — — + — — — — — — — — — — — — — — — -+
| BS_1_init_schema.up.sql | 2020–05–07 12:21:25 +0000 UTC |
| BS_2_add_brand_field.up.sql | 2020–05–12 14:31:17 +0000 UTC |
| BS_3_alter_table.up.sql | 2020–05–13 06:17:25 +0000 UTC |
| BS_4_add_created_at_field.sql | 2020–07–21 09:55:49 +0000 UTC |
| BS_5_alter_table_nourishment_diet.up.sql | 2020–07–21 09:55:49 +0000 UTC |
+ — — — — — — — — — — — — — — — — — — — — — + — — — — — — — — — — — — — — — -+
Enter fullscreen mode Exit fullscreen mode

Can go to the Helm chart now.

Helm template

What do we need to do in the old chart?

  1. remove initContainers
  2. remove old Secrets
  3. move values for our new variables in the values.yaml
  4. move the token and database’s password into the secrets.yaml

And the main part here is to add annotations to trigger the migrations process during Helm deploy

Add the annotations to the Job:

apiVersion: batch/v1
kind: Job
metadata:
  name: {{ .Chart.Name }}-migration-job
  labels:
  annotations:
    "helm.sh/hook": pre-install,pre-upgrade
    "helm.sh/hook-weight": "-1"
    "helm.sh/hook-delete-policy": before-hook-creation
spec: 
  backoffLimit: 0
...
Enter fullscreen mode Exit fullscreen mode

Here:

  • "helm.sh/hook": pre-install,pre-upgrade: run the Job before helm install or upgrade (in our Jenkins pipeline it's started with the helm secrets upgrade --install)
  • "helm.sh/hook-weight": "-1": priority of the creation of the resources, as first we need to create the ConfigMap and Secret that will be used by our Job, so set their weight less than for the Job
  • "helm.sh/hook-delete-policy": the default value is the before-hook-creation (check the documentation), set if for testing purpose, and then it can be changed to the hook-succeeded (but in that case you'll not be able to check the logs if the migration will fail)

Add the annotations block to the ConfigMap and Secrets with the hook-weight less than in the Job.

The ConfigMap manifest now full content:

---
apiVersion: v1
kind: ConfigMap
metadata:
  name: migration-config
  annotations:
    "helm.sh/hook": pre-install,pre-upgrade
    "helm.sh/hook-weight": "-5"
    "helm.sh/hook-delete-policy": before-hook-creation
data:         
  config.yaml: |
    main:
      dialect: {{ .Values.backendConfig.db.driver }}
      datasource: {{ .Values.backendConfig.db.user }}:${DB_PASSWORD}@tcp({{ .Values.backendConfig.db.host }}:{{ .Values.backendConfig.db.port }})/{{ .Values.backendConfig.db.database }}?parseTime=true
      dir: backend-services/db/migrations
      table: {{ .Values.backendConfig.db.migrationsTable }}
Enter fullscreen mode Exit fullscreen mode

The Secret one:

---     
apiVersion: v1
kind: Secret
metadata:
  name: {{ .Chart.Name }}-migration-secrets
  annotations: 
    "helm.sh/hook": pre-install,pre-upgrade
    "helm.sh/hook-weight": "-10"
    "helm.sh/hook-delete-policy": before-hook-creation
type: Opaque
stringData:
  backend-db-password: {{ .Values.backendConfig.db.password }}
  git_token: {{ .Values.git.token }}
Enter fullscreen mode Exit fullscreen mode

And the Job:

apiVersion: batch/v1
kind: Job
metadata:
  name: {{ .Chart.Name }}-migration-job
  labels:
  annotations:
    "helm.sh/hook": pre-install,pre-upgrade
    "helm.sh/hook-weight": "-1"
    "helm.sh/hook-delete-policy": before-hook-creation
spec:
  backoffLimit: 0
  template:
    metadata:
      name: {{ .Chart.Name }}-migration-job-pod
    spec:
      restartPolicy: Never
      containers:
      - name: {{ .Chart.Name }}-db-migrations
        image: projectname/sql-migrate-git:latest
        command: ["/bin/sh", "-c"]
        args: 
          - git clone --single-branch --branch {{ .Values.git.branch }} [https://{{](https://%7B%7B) .Values.git.user }}:$GIT_TOKEN@{{ .Values.git.repo }} &&
            ls -l backend-services/db/migrations &&
            cat /config/config.yaml &&
            /bin/sql-migrate up -config=/config/config.yaml -env=main || exit 1;
            /bin/sql-migrate status -config=/config/config.yaml -env=main
        env:
        - name: GIT_TOKEN
          valueFrom:
            secretKeyRef:
              name: {{ .Chart.Name }}-migration-secrets
              key: git_token
        - name: DB_PASSWORD
          valueFrom:
            secretKeyRef:
              name: {{ .Chart.Name }}-migration-secrets
              key: backend-db-password
        volumeMounts:
        - name: migration-config
          mountPath: "/config/config.yaml"
          subPath: "config.yaml"
          readOnly: true
      volumes:
        - name: migration-config
          configMap: 
            name: migration-config
            items:
            - key: "config.yaml"
              path: "config.yaml"
Enter fullscreen mode Exit fullscreen mode

Here I’ve added the exit 1 to the /bin/sql-migrate u so the Job will Fail in case of errors during migrations and thus will not start the deployment process.

Run it in the Jenkins:

In the HOOKS we can see that first the Secret was created as it has hook-weight": "-10", then the ConfigMap, and the Job at last.

And the deployment process looks now like the next:

At first, the Secret, ConfigMap and Job resources are removed (as per the "helm.sh/hook-delete-policy": before-hook-creation), and then are created.

Check the Job’s status:

$ kk -n eks-stage-1-appname-api-ns get job
NAME COMPLETIONS DURATION AGE
appname-api-migration-job 1/1 3s 6m21s
Enter fullscreen mode Exit fullscreen mode

Its logs:

$ kk -n eks-stage-1-appname-api-ns logs job/appname-api-migration-job
Cloning into ‘backend-services’…
total 20
-rw-r — r — 1 root root 538 Oct 26 11:32 BS_1_init_schema.up.sql
-rw-r — r — 1 root root 180 Oct 26 11:32 BS_2_add_brand_field.up.sql
-rw-r — r — 1 root root 225 Oct 26 11:32 BS_3_alter_table.up.sql
-rw-r — r — 1 root root 194 Oct 26 11:32 BS_4_add_created_at_field.sql
-rw-r — r — 1 root root 272 Oct 26 11:32 BS_5_alter_table_nourishment_diet.up.sql
main:
dialect: mysql
datasource: backend-user:${DB_PASSWORD}@tcp(stage.backend-db3-master.example.com:3306)/dbname?parseTime=true
dir: backend-services/db/migrations
table: backend_services_migrations
Applied 0 migrations
+ — — — — — — — — — — — — — — — — — — — — — + — — — — — — — — — — — — — — — -+
| MIGRATION | APPLIED |
+ — — — — — — — — — — — — — — — — — — — — — + — — — — — — — — — — — — — — — -+
| BS_1_init_schema.up.sql | 2020–05–07 12:21:25 +0000 UTC |
| BS_2_add_brand_field.up.sql | 2020–05–12 14:31:17 +0000 UTC |
| BS_3_alter_table.up.sql | 2020–05–13 06:17:25 +0000 UTC |
| BS_4_add_created_at_field.sql | 2020–07–21 09:55:49 +0000 UTC |
| BS_5_alter_table_nourishment_diet.up.sql | 2020–07–21 09:55:49 +0000 UTC |
+ — — — — — — — — — — — — — — — — — — — — — + — — — — — — — — — — — — — — — -+
Enter fullscreen mode Exit fullscreen mode

Applied 0 migrations, as there are no changes in the migrations files after the last APPLIED.

All done.

Originally published at RTFM: Linux, DevOps, and system administration.

Top comments (0)

Read next

monarene profile image

Designing a fault-tolerant etcd cluster

Michael Mekuleyi -

574n13y profile image

Advanced Networking Concepts

Vivesh -

amjadmh73 profile image

Install a NATS server on Debian/Ubuntu

Amjad Abujamous -

jancizmar profile image

I have localized the best short-linking platform in the world

Jan Cizmar -