Kubernetes archivos - CloudArch

OpenSearch for SREs: The Open-Source Observability Powerhouse on Kubernetes

joaquinjimenezgarcia — Sun, 08 Feb 2026 18:54:11 +0000

The Rise of the Open-Source Alternative: OpenSearch for the Modern SRE

In the dynamic world of cloud-native infrastructure, robust observability is not just a nice-to-have; it’s a foundational requirement for any Site Reliability Engineer (SRE). For years, Elasticsearch dominated the landscape for log aggregation and search. However, a significant licensing change by Elastic in early 2021 created a void, prompting AWS to fork the last Apache 2.0 licensed version and launch OpenSearch.

What is OpenSearch? At its core, OpenSearch is a distributed, RESTful search and analytics engine built on Apache Lucene. It’s designed for high-volume data ingestion and rapid querying across massive datasets. Think of it as a specialized database for semi-structured data like logs, metrics, and traces.

Why is its open-source nature important for SREs? The “open” in OpenSearch isn’t just a marketing buzzword; it’s critical for SREs:

No Vendor Lock-in: You’re free from proprietary licenses and sudden feature restrictions. This gives you long-term stability and predictability in your tech stack.
Community-Driven Development: Bugs are squashed faster, features are added based on broad community needs, and you can even contribute fixes yourself.
Auditability & Transparency: You can inspect the source code, understand exactly how your data is being handled, and ensure there are no hidden surprises—crucial for security-conscious environments.
Cost Predictability: While managed services exist, you always have the option to self-host without escalating license costs as your data grows.

For an SRE, OpenSearch provides the backbone for what’s often called the “OSD Stack” (OpenSearch, OpenSearch Dashboards, and Data Prepper/Fluent Bit), empowering proactive monitoring, rapid incident response, and deep operational insights.

Core Concepts for the SRE Toolkit

To wield OpenSearch effectively, you need to understand its fundamental building blocks:

Cluster & Nodes:
- An OpenSearch Cluster is a group of interconnected servers (nodes) that work together to store and search your data.
- Nodes specialize: some are cluster_manager (formerly “master”) nodes (brains for cluster state), others are data nodes (muscles for storing and searching data), and some can be ingest nodes (for pre-processing data). SREs design clusters with dedicated nodes for stability and scale.
Indices & Documents:
- An Index is like a logical database table, a collection of related JSON documents. For logs, you typically create time-based indices (e.g., application-logs-2026.02.08).
- A Document is a single JSON record within an index—e.g., one log line, one metric data point, one trace span.
Shards & Replicas:
- To handle massive data, OpenSearch horizontally scales using Shards. An index is split into primary shards, distributed across data nodes.
- Replicas are copies of primary shards. They provide high availability (if a node fails, a replica becomes primary) and scale read operations. SREs fine-tune shard/replica counts for performance and resilience.
OpenSearch Dashboards:
- The browser-based UI for visualizing, analyzing, and managing your OpenSearch data. It’s where you build dashboards, run queries, and configure alerts.
Index State Management (ISM):
- An automated policy engine. SREs use ISM to define lifecycle rules for indices, like moving old data to cheaper storage, taking snapshots, or deleting it after a set period. This prevents disk exhaustion and controls costs.

Getting Hands-On: Your Production-Lite OpenSearch Lab on Kubernetes

The best way to understand OpenSearch is to run it. We’ll set up a mini-stack on Minikube (or any Kubernetes cluster) that mirrors a production setup for logs:

OpenSearch Cluster: The data store.
OpenSearch Dashboards: The UI.
Nginx: A sample application generating logs.
Fluent Bit: The lightweight log shipper.

All configurations will be declarative, using Helm charts and Kubernetes Jobs, making it fully automated and version-controllable—true SRE style.

Prerequisites:

Kubernetes cluster (Minikube recommended for local dev)
kubectl installed and configured
helm installed

1. Initialize Minikube

Ensure your Minikube has enough resources for OpenSearch:

minikube start --cpus 4 --memory 8192 --driver docker

2. Prepare Helm Repositories

helm repo add opensearch https://opensearch-project.github.io/helm-charts/
helm repo add fluent https://fluent.github.io/helm-charts
helm repo update

3. Deploy OpenSearch & Dashboards (with a Password)

We’ll use a values.yaml file to set up a single-node OpenSearch cluster with a secure admin password and OpenSearch Dashboards.

Save the following as opensearch-values.yaml:

# opensearch-values.yaml
singleNode: true
persistence:
  enabled: false # For a lab, we'll keep it stateless. Set to true for production with PVs.
extraEnvs:
  - name: OPENSEARCH_INITIAL_ADMIN_PASSWORD
    value: YourStrongPassword123! # <<< CHANGE THIS TO A STRONG PASSWORD

Now deploy:

helm install my-os opensearch/opensearch -f opensearch-values.yaml
helm install my-dashboards opensearch/opensearch-dashboards

Note: OpenSearch will take a few minutes to start as it initializes its JVM and security plugin. Keep an eye on kubectl get pods -w.

4. Deploy Sample Nginx App

This is our log source:

kubectl create deployment nginx-server --image=nginx
kubectl expose deployment nginx-server --port=80

5. Deploy Fluent Bit (The Log Shipper)

Fluent Bit will scrape Nginx logs and send them to OpenSearch. We’ll use a fluent-bit-values.yaml to handle the configuration declaratively and resolve common SRE headaches (DNS, auth, mapping issues).

Save the following as fluent-bit-values.yaml:

# fluent-bit-values.yaml
config:
  service: |
    [SERVICE]
        Daemon          Off
        Flush           1
        Log_Level       info
        Parsers_File    parsers.conf
        HTTP_Server     On
        HTTP_Listen     0.0.0.0
        HTTP_Port       2020
        Health_Check    On

  inputs: |
    [INPUT]
        Name           tail
        Path           /var/log/containers/*.log
        multiline.parser docker, cri
        Tag            kube.*
        Mem_Buf_Limit  5MB
        Skip_Long_Lines On

  filters: |
    [FILTER]
        Name                kubernetes
        Match               kube.*
        Kube_URL            https://kubernetes.default.svc:443
        Kube_CA_File        /var/run/secrets/kubernetes.io/serviceaccount/ca.crt
        Kube_Token_File     /var/run/secrets/kubernetes.io/serviceaccount/token
        Kube_Tag_Prefix     kube.var.log.containers.
        Merge_Log           On
        Merge_Log_Key       log_processed
        Keep_Log            Off
        # These help prevent mapping conflicts from inconsistent pod labels
        Labels              Off
        Annotations         Off

  outputs: |
    [OUTPUT]
        Name            es
        Match           *
        Host            my-os-opensearch # This is the Kubernetes Service name for OpenSearch
        Port            9200
        HTTP_User       admin
        HTTP_Passwd     YourStrongPassword123! # <<< USE THE SAME PASSWORD AS ABOVE
        Logstash_Format On
        Logstash_Prefix nginx-logs
        tls             On
        tls.verify      Off
        Suppress_Type_Name On # Crucial for OpenSearch 2.x
        Trace_Error        On # For better debugging in Fluent Bit logs

Deploy Fluent Bit:

helm install fluent-bit fluent/fluent-bit -f fluent-bit-values.yaml

The Log Shipper: Why Fluent Bit?

In a Kubernetes ecosystem, logs are ephemeral—when a pod dies, its logs go with it. To prevent this, we need a DaemonSet that acts as a “log vacuum.” We chose Fluent Bit because it’s the lightweight, high-performance cousin of Fluentd. It has a tiny memory footprint (crucial when you’re running it on every node in a cluster) and handles the “Log Pipeline” in three distinct stages:

Input (Tail): It watches the raw .log files created by the Kubernetes container engine on the node’s disk.
Filters (Kubernetes): This is the “SRE secret sauce.” Fluent Bit talks to the Kubernetes API to enrich your logs with metadata like the Pod Name, Namespace, and Labels. We’ve also added logic here to “clean” the logs to prevent mapping conflicts.
Output (OpenSearch): It batches the logs and ships them securely via HTTPS to our OpenSearch cluster.

The Anatomy of a Fluent Bit Pipeline

When you look at a Fluent Bit configuration, it’s organized into distinct sections. Each one has a specific job in the “Ingestion Lifecycle.” Understanding these is the key to onboarding any new service into OpenSearch.

[SERVICE] (The Global Brain): This section defines the engine’s behavior. It controls how often data is “flushed” (sent) to the destination, where the internal logs go, and whether to enable a health-check server. For SREs, this is where we tune performance and monitoring for the log shipper itself.
[INPUT] (The Collector): This is the “vacuum cleaner.” It tells Fluent Bit where to get data. In Kubernetes, we usually use the tail input to follow the log files generated by the container runtime (CRI/Docker). You can have multiple inputs—one for system logs, one for app logs, and even one for metrics.
[FILTER] (The Processor): This is where the magic happens. Filters allow you to modify data in flight.
- The Kubernetes Filter is the most popular; it reaches out to the K8s API to tag your logs with pod names and namespaces.
- You can also use filters to drop sensitive data (PII), parse strings into JSON, or “flatten” complex labels to avoid the mapping conflicts we saw earlier.
[OUTPUT] (The Destination): This defines the “Exit” for your data. In our case, it’s the es (Elasticsearch/OpenSearch) plugin. This section handles the connection details, authentication, and index naming conventions.

6. Automate ISM Policy & Index Pattern with a Kubernetes Job

This is where the SRE magic happens. We’ll use a Job to apply our ISM policy (7-day retention) and create the nginx-logs-* index pattern in Dashboards, all via API calls.

Save the following as observability-setup-job.yaml:

# observability-setup-job.yaml
apiVersion: v1
kind: ConfigMap
metadata:
  name: opensearch-sre-setup-script
data:
  setup.sh: |
    #!/bin/bash
    set -euo pipefail

    OPENSEARCH_USER="admin"
    OPENSEARCH_PASSWORD="YourStrongPassword123!" # <<< USE THE SAME PASSWORD
    DASHBOARDS_URL="http://my-dashboards-opensearch-dashboards.default.svc.cluster.local:5601"
    OPENSEARCH_URL="https://my-os-opensearch.default.svc.cluster.local:9200"

    echo "Waiting for OpenSearch Dashboards to be available..."
    until curl -s -u "$OPENSEARCH_USER:$OPENSEARCH_PASSWORD" -k "$DASHBOARDS_URL/api/status" | grep -q "available"; do
      sleep 5
    done
    echo "OpenSearch Dashboards is available."

    echo "Creating ISM policy: nginx_retention..."
    curl -X PUT "$OPENSEARCH_URL/_plugins/_ism/policies/nginx_retention" \
      -u "$OPENSEARCH_USER:$OPENSEARCH_PASSWORD" -k -H "Content-Type: application/json" \
      -d '{
            "policy": {
              "description": "Delete logs after 7 days",
              "default_state": "hot",
              "states": [
                {
                  "name": "hot",
                  "actions": [],
                  "transitions": [
                    {
                      "state_name": "delete",
                      "conditions": { "min_index_age": "7d" }
                    }
                  ]
                },
                {
                  "name": "delete",
                  "actions": [ { "delete": {} } ],
                  "transitions": []
                }
              ],
              "ism_template": [
                {
                  "index_patterns": ["nginx-logs-*"],
                  "priority": 100
                }
              ]
            }
          }'
    echo "ISM policy created."

    echo "Creating OpenSearch Dashboards Index Pattern: nginx-logs-pattern..."
    curl -X POST "$DASHBOARDS_URL/api/saved_objects/index-pattern/nginx-logs-pattern" \
      -u "$OPENSEARCH_USER:$OPENSEARCH_PASSWORD" -H "osd-xsrf: true" -H "Content-Type: application/json" \
      -d '{
            "attributes": {
              "title": "nginx-logs-*",
              "timeFieldName": "@timestamp"
            }
          }'
    echo "Index pattern created."

---
apiVersion: batch/v1
kind: Job
metadata:
  name: opensearch-observability-setup
spec:
  template:
    spec:
      containers:
      - name: setup-runner
        image: curlimages/curl:latest # A lightweight image with curl and bash
        command: ["bash", "/scripts/setup.sh"]
        volumeMounts:
        - name: setup-script
          mountPath: /scripts
      volumes:
      - name: setup-script
        configMap:
          name: opensearch-sre-setup-script
          defaultMode: 0744 # Make the script executable
      restartPolicy: OnFailure

Apply the setup Job:

kubectl apply -f observability-setup-job.yaml

Understanding the Automation: ISM & Index Patterns

Before we fire off the job, let’s talk about what’s happening under the hood. We aren’t just pushing config; we are defining the lifecycle and visibility of our data.

ISM (Index State Management): In a production cluster, logs are a “growing fire.” If you don’t manage them, they will eventually eat your disk and crash your nodes. By defining an ISM Policy, we automate the “Hot-to-Delete” lifecycle. In this setup, we’re telling OpenSearch: “Keep these logs fresh for 7 days, then delete them automatically.” No manual cleanup, no 3 AM disk-space alerts.
Index Patterns: While the Index is where the data lives, the Index Pattern is the “Lens” used by OpenSearch Dashboards to see it. By creating this via code, we ensure that as soon as the stack is up, your Discover tab is ready to go. We’re essentially “gluing” all those daily daily nginx-logs-* into one continuous timeline so you can query across multiple days without lifting a finger.

7. Generate Some Nginx Traffic

To see logs, your Nginx server needs visitors:

kubectl run load-gen --image=busybox --restart=Never -- /bin/sh -c "while true; do wget -qO- http://nginx-server; sleep 2; done"

8. Access OpenSearch Dashboards

Port-forward Dashboards to your local machine:

kubectl port-forward svc/my-dashboards-opensearch-dashboards 5601:5601

Then, open https://localhost:5601 in your browser. Log in with admin and your chosen password. Go to the “Discover” tab, select nginx-logs-* in the dropdown, set your time range (e.g., “Last 15 minutes”), and watch your logs flow in!

Conclusion

By following this automated approach, you’ve built a robust, observable, and easily reproducible log aggregation stack with OpenSearch. You’ve tackled critical SRE challenges like security, data ingestion, and lifecycle management—all as code. This hands-on experience forms a solid foundation for further exploration into metrics, traces, and advanced alerting in your cloud-native environments.

What other OpenSearch challenges will you automate next?

See the whole code at my repository

La entrada OpenSearch for SREs: The Open-Source Observability Powerhouse on Kubernetes se publicó primero en CloudArch.

Introduction to KEDA: Event-Driven Autoscaling for Kubernetes

joaquinjimenezgarcia — Sun, 07 Dec 2025 20:12:58 +0000

When building applications in Kubernetes, one of the most important challenges is efficiently managing workloads. You want your services to scale up when demand spikes, and scale down (even to zero) when idle, to save resources and reduce costs. This is especially critical in production environments where traffic can be unpredictable.

This is where KEDA (Kubernetes Event-Driven Autoscaling) comes in. KEDA is a lightweight, open-source component that integrates with Kubernetes to provide event-driven autoscaling. Unlike traditional Horizontal Pod Autoscalers (HPAs) that only scale based on CPU or memory usage, KEDA can scale your workloads based on external metrics or events, such as:

Messages in a queue (Redis, RabbitMQ, Azure Service Bus, etc.)
Jobs waiting in Kafka topics
Custom metrics from Prometheus
Database triggers or cloud events

With KEDA, your applications can react instantly to real-world workloads without over-provisioning resources. It allows you to run microservices cost-effectively while maintaining responsiveness.

What we are building

In this guide, we’ll build a realistic event-driven microservice that processes jobs from a Redis queue. The scenario mirrors what many production systems face:

A backend service receives tasks (e.g., image processing, notifications, or data ingestion) and pushes them into a Redis queue.
Worker pods consume tasks from the queue.
KEDA monitors the queue and scales the number of worker pods up or down depending on the number of pending tasks.

By the end of this guide, you will have:

A working Minikube cluster with KEDA installed
A Redis-backed job queue
A worker deployment that automatically scales according to queue length
Complete YAML manifests and a test workflow to simulate real production traffic

This project is valuable because it lets you experience a real-world autoscaling scenario. Most production systems have unpredictable workloads, and learning how KEDA responds to events gives you a deep understanding of:

Event-driven design patterns
How Kubernetes interacts with external triggers
Autoscaling strategies beyond CPU/memory metrics

Even if your real application is more complex (with multiple queues or different event sources), this example provides a solid foundation to implement production-ready, cost-efficient, and resilient services.

What you’ll learn:

Start a 3‑node Minikube cluster (if not started)
Install Helm (if needed)
Install KEDA with Helm
Deploy Redis (simple single‑replica) and a worker Deployment that consumes jobs
Create a ScaledObject that uses the Redis List scaler
Create a producer CronJob that pushes items into the Redis list to simulate load
Observe scaling behavior and clean up

Prerequisites

Make sure you have the following locally installed and working on your machine:

kubectl (compatible with your Minikube Kubernetes version)
minikube (we will use it to run a 3‑node local cluster)
helm (Helm 3)
docker (or your container runtime used by Minikube driver)

If any of those are missing, install them first. For example on Debian/Ubuntu: sudo snap install kubectl –classic (or use your distro’s package manager). I will not assume any particular OS beyond these tools being available.

Start a 3-node Minikube cluster

# Start minikube with 3 nodes (one control-plane + 2 workers), adjust memory/
CPUs if needed
$ minikube start --nodes=3 --driver=docker --memory=4096 --cpus=2

# Verify nodes come up
$ kubectl get nodes

Notes: –nodes=3 creates 1 control-plane and 2 worker nodes by default. If you already have the cluster running, skip the minikube start step. Make sure kubectl context points to your minikube cluster: kubectl config current-context .

Install Helm

This command will install Helm for you in your local machine. Please, refer to this other post if you wanna learn more about it.

# Add KEDA Helm repo and update
helm repo add kedacore https://kedacore.github.io/charts
helm repo update

# Install KEDA into namespace 'keda'
helm install keda kedacore/keda --namespace keda --create-namespace

# Wait for KEDA pods to become ready
kubectl -n keda get pods

Why Helm? Helm is the easiest official way to install KEDA and its CRDs. KEDA installs CRDs that are required for ScaledObject and ScaledJob resources.

What we’re going to deploy (high level)

redis-deployment.yaml — a simple Redis single‑pod deployment + service to be able to access it from other pods
worker-deployment.yaml — a small Python worker Deployment that polls job-queue (a Redis list) and processes items, initially it will have 0 replics since it will imitate a job that doesn’t need to be running always
producer-cronjob.yaml — a CronJob that pushes messages into Redis periodically to generate load
scaledobject-redis.yaml — KEDA ScaledObject definition that scales worker based on Redis list length

I’ll explain each file before showing it. All manifests are designed to run in the default namespace for simplicity. You can change namespace fields if you prefer.

Redis Manifest

# redis-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: redis
  labels:
    app: redis
spec:
  replicas: 1
  selector:
    matchLabels:
      app: redis
  template:
    metadata:
      labels:
        app: redis
    spec:
      containers:
      - name: redis
        image: redis:7.0-alpine
        ports:
        - containerPort: 6379
        resources:
          requests:
            cpu: "100m"
            memory: "128Mi"
          limits:
            cpu: "250m"
            memory: "256Mi"
---
apiVersion: v1
kind: Service
metadata:
  name: redis
  labels:
    app: redis
spec:
  ports:
  - port: 6379
    targetPort: 6379
    protocol: TCP
  selector:
    app: redis
  type: ClusterIP

As we should know already, we are gonna deploy this in our cluster with:

kubectl apply -f redis-deployment.yaml

To be sure the redis service is client, we can test the connectivity with a redis client:

kubectl run -i --tty redis-client --image=redis:7.0-alpine --restart=Never
-- sh
# inside pod shell run: redis-cli -h redis ping
# should reply PONG

Worker code + Deployment

We’ll create a tiny Python worker that continuously polls a Redis list (named job-queue ) with BRPOP and “processes” messages (here, just sleep and echo ). Just to replicate some functionality of reading from Redis. The purpouse of this guide is not the logic of the code itself but it’s realibility on production environments.

In real life your worker would do meaningful job processing.

# worker.py
import time
import os
import redis


REDIS_HOST = os.getenv('REDIS_HOST', 'redis')
REDIS_PORT = int(os.getenv('REDIS_PORT', '6379'))
LIST_NAME = os.getenv('LIST_NAME', 'job-queue')


r = redis.Redis(host=REDIS_HOST, port=REDIS_PORT, decode_responses=True)
print('Worker started, connecting to', REDIS_HOST)


while True:
    try:
        # BRPOP blocks until an item is available
        item = r.brpop(LIST_NAME, timeout=5)
        if item:
            # item is (list_name, value)
            value = item[1]
            print('Processing', value)
            # simulate processing
            time.sleep(2)
            print('Done', value)
        else:
            # nothing to do, sleep to avoid tight loop
            time.sleep(1)
    except Exception as e:
        print('Worker error:', e)
        time.sleep(2)

# Dockerfile.worker
FROM python:3.11-slim
WORKDIR /app
COPY worker.py .
RUN pip install --no-cache-dir redis
CMD ["python","/app/worker.py"]

By default if we create the image, it will be in our local registry and Minikube won’t see it. To solve this problem we can enable an addon called registry which will allow us to create a registry in Minikube.

minikube addons enable registry

We need to do a port-forward to forward the data to the registry:

kubectl port-forward -n kube-system service/registry 5000:80

And now, in other terminal to not terminate our tunnel, we need to build and push the image so it can be used within the cluster:

docker build -t keda-worker:latest -f Dockerfile.worker .
docker tag keda-worker:latest localhost:5000/keda-worker:latest
docker push localhost:5000/keda-worker:latest

And now we need to create the deployment of our service using the brand new image:

# worker-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: keda-worker
  labels:
    app: keda-worker
spec:
  replicas: 0 # start with 0 so we can see KEDA scale up from zero
  selector:
    matchLabels:
      app: keda-worker
  template:
    metadata:
      labels:
        app: keda-worker
    spec:
      containers:
      - name: worker
        image: localhost:5000/keda-worker:latest
        imagePullPolicy: Always
        env:
        - name: REDIS_HOST
          value: "redis"
        - name: REDIS_PORT
          value: "6379"
        - name: LIST_NAME
          value: "job-queue"
        resources:
          requests:
            cpu: "50m"
            memory: "64Mi"
          limits:
            cpu: "200m"
            memory: "256Mi"

Please, pay attention that we stated that we want a total of 0 replicas since the design of this is to only run when Keda allows it, saving us computing time and resources in our cluster.

kubectl apply -f worker-deployment.yaml

Create the KEDA ScaledObject for Redis List

Now that we have Redis ready to have message and our service ready to start reading those messages when it’s needed, we need to create a ScaledObject which will tell our service when it’s time to work, create instances and do its job.

# scaledobject-redis.yaml
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: keda-worker-scaledobject
  labels:
    app: keda-worker
spec:
  scaleTargetRef:
    name: keda-worker
  pollingInterval: 5 # how often KEDA checks Redis (seconds)
  cooldownPeriod: 30 # how long to wait after scale down before next check
  minReplicaCount: 0
  maxReplicaCount: 10
  triggers:
  - type: redis
    metadata:
      address: "redis.default.svc:6379"
      listName: "job-queue"
      listLength: "5" # scale target: number of items -> triggers scale
      activationListLength: "1" # minimum backlog to activate scaling

Fields explained:

scaleTargetRef.name — the Deployment the ScaledObject controls (kedaworker).
pollingInterval — how often KEDA will query Redis.
minReplicaCount /maxReplicaCount — limits for scaling.
triggers — an array of trigger definitions. Here we use the redis scaler.
listLength is the target backlog size that will cause KEDA to adjust replicas (KEDA converts this into HPA metrics internally).
activationListLength prevents scaling up until backlog is at least that value.

Please, pay special attention at triggers.type since that value is unique for the kind of job we are doing. Keda provides us with an extensive list of different triggers we can use depending on what we want to observe in order to scale our services. You can have a complete list at their official site.

Keda and Redis are on different namespaces on our cluster, when specifying the address make sure you are pointing correctly to Redis. If you don’t use default namespace as I did, it will be different than redis.default.svc:6379

kubectl apply -f scaledobject-redis.yaml

Producer Cronjon to generate load

Finally, we are going to create a cronjob that will just generate some load to replicate a real user, and will just push a message to redis every minute so we can emulate the whole flow automatically and don’t hit any button neither waiting for an user.

# producer-cronjob.yaml
apiVersion: batch/v1
kind: CronJob
metadata:
  name: job-producer
spec:
  schedule: "*/1 * * * *"
  jobTemplate:
    spec:
      template:
        spec:
          restartPolicy: Never
          containers:
          - name: producer
            image: redis:7.0-alpine
            command: [ "sh", "-c" ]
            args:
            - |
              now=$(date +%s)
              echo "Producing job-$now"
              redis-cli -h redis rpush job-queue job-$now
              sleep 1

kubectl apply -f producer-cronjob.yaml

Observe KEDA scaling in action

For this purpose I have splited my terminal into 3 sections, so I can see everything at one glance. First I have to sections of the same size: right and left.

On the right size I will be running the logs of the Scaled Object to see how it’s being triggered everytime it sees there is 1 or more messages in the queue

kubectl logs -f -n keda keda-operator-f948b6c4-ln9ch

And then the left side I will have it splitted into two parts again. In the upper part I will be checking how many messages do I have on the list

watch 'kubectl exec -it deploy/redis -- redis-cli llen job-queue'

And at the botton I will be watching the number of pods from my deplotments

watch 'kubectl get deploy'

There we will be able to see how every minute the list has a new message, a pod of the worker is being created and the message is deleted. Reading all the logs in the right side.

By this way, we could have a service that only will be running when it’s needed. But also, this can be replicated and configured to just scale up or down services based on some triggers to ensure we are always giving the desired availability and reliability.

Clean up resources

kubectl delete -f producer-cronjob.yaml
kubectl delete -f scaledobject-redis.yaml
kubectl delete -f worker-deployment.yaml
kubectl delete -f redis-deployment.yaml
helm uninstall keda -n keda
kubectl delete namespace keda

You can see and download all the code on GitHub https://github.com/JoaquinJimenezGarcia/LearningKeda

La entrada Introduction to KEDA: Event-Driven Autoscaling for Kubernetes se publicó primero en CloudArch.

Helm Repositories: Your Gateway to Kubernetes Apps 🗂️⛵

joaquinjimenezgarcia — Tue, 08 Apr 2025 20:46:25 +0000

So you’ve installed Helm, taken your first breath in this world of Kubernetes package management, and you’re wondering… Now what? What makes Helm truly magical isn’t just that it simplifies your deployments — it’s where it gets its powers from: the Helm repository.

If Helm is the Kubernetes version of a package manager, then Helm repos are its treasure chests. And just like any good explorer, you’re going to need to know where to find the maps.

So… what is a Helm repo, really?

Let’s imagine you’re building a house in Kubernetes. Instead of crafting each brick by hand (writing all those YAMLs), you can order a ready-to-assemble kit. That kit? It’s called a Helm chart. And where do you get it from? A Helm repository.

A Helm repo is basically a collection of Helm charts — pre-packaged Kubernetes applications that are ready for deployment. These charts are stored in an online repository, which you can connect to and browse from your terminal.

Think of it like a Play Store or App Store, but for Kubernetes.

The Default Chart Repo: Bitnami & Friends

The Helm team maintains a default, stable repo full of widely-used applications: NGINX, MySQL, Prometheus, Grafana, WordPress… you name it. One of the most popular ones is from Bitnami, which packages and maintains a huge collection of up-to-date and secure charts.

But the real beauty is this: you can add any Helm repo to your local setup — official, third-party, or even your own private one.

Let’s get our hands dirty

Adding a Helm repo is as simple as a single command. For example, to add Bitnami’s public repository:

helm repo add bitnami https://charts.bitnami.com/bitnami

That’s it. You’ve just connected your Helm CLI to a vast world of applications.

Want to make sure everything’s in sync?

helm repo update

This command refreshes the list of available charts and their latest versions from all the repos you’ve added.

To check which repos you currently have on your machine:

helm repo list

It’ll show you a tidy little table with all your configured repositories. Nice and clean.

Exploring the Chart Library with `helm search`

Once you’ve added a repository — say Bitnami — and updated it, you’re probably thinking: “Cool… but what can I actually install?”

That’s where the magic of searching comes in.

You can use:

helm search repo wordpress

And Helm will scan through all your added repositories and list any charts that match the name “wordpress”. It’ll even show you the latest version and a short description — kind of like browsing a menu at a restaurant, but for apps in your Kubernetes cluster .

Want to see everything available from Bitnami? Just do:

helm search repo bitnami

And boom — a full buffet of deployable goodness.

This command is especially useful when you’re not quite sure what you’re looking for, or when you’re exploring new tools to add to your stack.

Installing Charts from Repos

Now comes the fun part. Let’s say you want to install WordPress in your Kubernetes cluster (because why not launch a blog while you’re launching containers? ).

Installing any chart is as easy as using the helm install command following with some parameters:

helm install

Once you’ve added Bitnami’s repo, you can run:

helm install my-wordpress bitnami/wordpress

Boom. Helm fetches the chart from the Bitnami repo, installs it into your cluster, and gives you all the necessary info to access it.

You don’t even need to download the chart locally unless you want to customize it (more on that in a future post ).

Behind the Scenes

When you install a chart from a repo, Helm does a few things:

It looks up the chart metadata in the repo index.
It pulls down a .tgz (tar.gz) package that contains all the YAML templates, values, and configuration.
It renders those templates with your values and sends the final manifests to Kubernetes.

The best part? You didn’t touch a single YAML file.

Final Thoughts

Getting familiar with Helm repos is a crucial step in mastering Helm — and honestly, it’s where the fun begins. You now have access to hundreds of production-grade Kubernetes apps, just a command away.

But don’t stop here.

Next, we’re going to explore how to customize these charts to suit your environment, your values, and your style (yes, Kubernetes can have style ).

Until then, try playing with a few different charts. Install Redis, launch a full monitoring stack with Prometheus + Grafana, or spin up your own Jenkins CI system — all in just a few keystrokes.

Your cluster will thank you.

Enjoying the series?
Make sure to follow, subscribe, or share this post with someone who still thinks Helm is just something you wear on a boat.
#Helm #Kubernetes #DevOps #CloudNative #K8s #InfrastructureAsCode #HelmCharts

La entrada Helm Repositories: Your Gateway to Kubernetes Apps 🗂️⛵ se publicó primero en CloudArch.

What is Helm? The Kubernetes Package Manager Explained for Beginners 🚀

joaquinjimenezgarcia — Tue, 08 Apr 2025 10:30:27 +0000

Welcome to the first post in our Helm series! Whether you’re a DevOps engineer, cloud enthusiast, or just getting started with Kubernetes, you’re about to discover a powerful tool that can make your life way easier: Helm.

In this post, we’ll dive into:

What is Helm?
What problems does it solve?
Key benefits of using Helm
How to install Helm (in just a few steps!)

Let’s jump right in!

What is Helm?

Imagine managing a complex application in Kubernetes — dozens of YAML files, multiple services, deployments, secrets, configs, and everything in between. It’s like assembling IKEA furniture without the manual.

Helm is here to be your manual.

It’s the package manager for Kubernetes — just like apt is for Ubuntu or yum is for CentOS, but made for the cloud-native world.

With Helm, you can:

Package your Kubernetes YAMLs into reusable templates (called Charts)
Deploy complex applications with a single command
Easily manage upgrades, rollbacks, and configurations

In short: Helm makes deploying to Kubernetes faster, simpler, and less error-prone.

What Problems Does Helm Solve?

Kubernetes is powerful, but managing it manually is like juggling flaming swords. Here are some real headaches Helm helps with:

1. Too many YAML files

Applications often require multiple resources: Deployments, Services, ConfigMaps, Ingresses, etc. Helm bundles them into one chart.

2. Hard-coded values

Editing the same YAML files over and over to change values (like image tags or environment settings)? Helm allows dynamic values with templates and variables.

3. No easy way to upgrade or rollback

With Helm, you can upgrade an app version and roll back instantly if something goes wrong.

4. App reuse across environments

Want the same app in dev, staging, and prod with just different configs? Helm makes it effortless.

Benefits of Using Helm

Helm isn’t just a nice-to-have — it’s a must-have for scalable, maintainable Kubernetes environments.

Here’s why:

• Saves Time

Automate and simplify deployments with one-liners.

• Improves Consistency

Avoid human error and deploy the same app across environments with confidence.

• Version Control Friendly

Helm charts can live in Git, making your infrastructure-as-code even more powerful.

• Easily Shareable

Share your charts with teammates or open source them for the community.

• Built-in Rollbacks

One bad deployment? Roll it back like it never happened.

How to Install Helm

Ready to get started? Installing Helm is quick and painless.

Step 1: Download the Helm Binary

For macOS:

brew install helm

For Linux:

curl https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash

For Windows:

Use Chocolatey or Scoop:

choco install kubernetes-helm

Step 2: Verify the Installation

helm version

You should see something like:

version.BuildInfo{Version:"v3.x.x", GitCommit:"...", ...}

Boom — you’re ready to helm your ship!

What’s Next?

In the next post, we’ll break down what a Helm Chart really is, how it’s structured, and how to create your very first one.

If you’re enjoying this series, don’t forget to:

Subscribe to the blog
Share this post with your DevOps squad
Leave a comment if you have questions or want a topic covered!

Until next time — keep it cloud-native!

La entrada What is Helm? The Kubernetes Package Manager Explained for Beginners 🚀 se publicó primero en CloudArch.

How to deploy your Kubernetes app – Part II

joaquinjimenezgarcia — Sun, 04 Aug 2024 22:42:02 +0000

In a previous post we were learning how to install and configure ArgoCD to learn how to deploy your Kubernetes app.

In this new post we will be seeing how to integrate our first Kubernetes deployment by default and learn the basics to start automating our services and workloads.

Before starting we must have something to deploy. For this lab I have prepared a default deployment that will create 3 pods with a Nginx container in each. You can take a look or using it by this link: https://github.com/JoaquinJimenezGarcia/argocd-deploy-test but in case you want to create your own repo basically it looks like this:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 3
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: nginx:1.14.2
        ports:
        - containerPort: 80

Now that we have our ArgoCD up and running, we will see that it’s empty. However in the upper left corner we would see a button labeled “New App”. We must click there and a new window will pop up with some info we must complete.

However these are the most important fields we need to fill:

Application Name: this is the name that will return our ArgoCD to refeer to our resources. We can assign here the value that we want, but following the name of the deployment I gave it the same one: argocd-deploy-test
Project Name: the project we are going to use in ArgoCD, since we only have “default” this is the value it must have, but we can create multiple projects with multiple resources
Sync Policy: if we want to sync manually or automatically, we would mark “automatically” since we want to be updated if we update the YAML files on Github
Repository URL: where the YAML files live, if you are using directly the one I have created, it’s https://github.com/JoaquinJimenezGarcia/argocd-deploy-test
Path: the path to deploy our pods, by default I left it as “.”
Cluster URL: in which cluster we want to deploy, since we only have the cluster where ArgoCD is installed, we can leave it as https://kubernetes.default.svc
Namespace: the namespace where the pods would be deployed, as a good practice we should have multiple namespaces, but as per this lab we are going to be uing only “default” namespace

Now we have all these fields ready, we need to click on create app and ArgoCD autimatically will go to our repo, read the yaml files and execute them.

Because this is only a small YAML file, we should see just a few seconds later our app completeley deployed

If we want to double check, we can go to our cluster and examine the pods looking to see if the pod names match with the ones are deployed there:

And as we can see, the results matches, so our app is perfectly deployed.

So as we marked the auto sync, now it will be reading our Git each 3 minutes to detect changes and apply them. So if for example we push a PR changing the replicas from 3 to 5, it would be updated automatically without more human interaction.

La entrada How to deploy your Kubernetes app – Part II se publicó primero en CloudArch.

How to deploy your Kubernetes app – Part I

joaquinjimenezgarcia — Sun, 28 Jul 2024 18:52:00 +0000

Learn how to deploy your Kubernetes app! When we are working with Kubernetes, we might face multiple yaml files for a single application: deployment file, secrets files, service account files, etc. It may be overwhelming in some cases to maintain all those files and keep the last versions on our infrastructure. That’s why here we are gonna learn how to deploy your Kubernetes app using Github and ArgoCD.

In this first part we will be covering the ArgoCD installation using minikube. However, the exactly same steps can be followed in another Kubernetes environment. In the cloud or on-premises.

What is ArgoCD

Argo CD is a declarative, GitOps continuous delivery tool for Kubernetes. Argo CD follows the GitOps pattern of using Git repositories as the source of truth for defining the desired application state.

You can read more info in their official website

Installing ArgoCD

Taking into consideration we already have minikube installed and running with kubectl correctly configured, we just need to create a namespace and deploy a Kubernetes yaml file from the official ArgoCD repository.

kubectl create namespace argocd
kubectl apply -n argocd -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml

It’s important to keep the namespace as the one in the examnple because the manifest we are downloading includes references to it such as some for Service Accounts. If we want or we need to change it, we would need to download the repository before applying the manifest and change the references to the namespace.

Since we are gonna be working in the namespace we just created, a good practice would be setting up the defualt namespace to that one to avoid confusion.

kubectl config set-context --current --namespace=argocd

Right after that we would be ready to install the argo command line tool to interact with our ArgoCD via the console. We can just download it from https://github.com/argoproj/argo-cd/releases/latest but if you are using Mac, it can be easier to install it using brew.

brew install argocd

We are able to check that ArgoCD is sucessfully deployed by checking the containers in the current namespace.

k get po
NAME                                               READY   STATUS    RESTARTS   AGE
argocd-application-controller-0                    1/1     Running   0          30m
argocd-applicationset-controller-65bb5ff89-wrfw9   1/1     Running   0          30m
argocd-dex-server-6f898cbd9-4f52l                  1/1     Running   0          30m
argocd-notifications-controller-64bc7c9f7-rpfxk    1/1     Running   0          30m
argocd-redis-5df55f45b7-x8snc                      1/1     Running   0          30m
argocd-repo-server-74d5f58dc5-96zmf                1/1     Running   0          30m
argocd-server-5b86767ddb-lxjl5                     1/1     Running   0          30m

Making our Argo accesible

By default ArgoCD is not exposed to an external IP, so if we need to acces it from outside for example to add new apps or to check the status or our current apps, we must create a Load Balancer, an Ingress or do a port-fordwarding.

Since we are working on minikube, with the port-forwarding would be more than enough. But for cloud environments Ingress may be the right fit. We would be covering now only port-forwarding, but in a later part of this serie we would be working on customizating our ArgoCD by adding other clusters to deploy our apps and exposing it under other kind of network Services.

kubectl port-forward svc/argocd-server -n argocd 8080:443

After forwarding we would be able to see our Argo in our localhost.

Retrieving your ArgoCD credentials

Your credentials were stored in a Secret in your Kubernetes cluster when you were deploying it, but also you can use the argo command-line tool to retrieve the password. The user by default is “admin”.

argocd admin initial-password -n argocd

However a good practice is always changing the default password of your service, you can do it also using the argocd command-line tool

argocd account update-password

Once we have logged in using our credentials, we should see a site like the following.

Farewell

If you enjoyed reading this first part, stay tune to get to know in the next one how to configure the deployment on ArgoCD of our apps.

Also, if you need to know more about Kubernetes, visit our latest posts.

La entrada How to deploy your Kubernetes app – Part I se publicó primero en CloudArch.

Solving CKA exercises – Part 4

joaquinjimenezgarcia — Sun, 17 Mar 2024 21:13:20 +0000

When we are talking about security in Kubernetes, one of the first things we must work on are the Service Accounts and roles attached to those. Solving these CKA exercises you will lear about this topic easily.

Service Accounts are essential to impersonate and consume resources in our clusters, they can be used by services, applications or even users. We can create roles based on Kubernetes’ permissions and binding those to Service Accounts, so we can get granted those permissions.

We can choose between Roles and ClusterRoles, where the first ones are created in the context of a namespace and the second one are created at Cluster level, so it can be reused across different namespaces and contexts.

Also, when creating the binding to add that role to a Service Account we can choose between RoleBinding and ClusterRoleBinding, where the first one does the binding at context level and the second one at the cluster lever, making it available for all the resources from the cluster specified on the permissions attached to it.

If you like it, don’t forget to see other exercises on KillerCoda and read more posts about Kubernetes.

Used Commands:

kubectl -n create sa
To create a Service Account in a specific Namespace
kubectl describe clusterrole
To see permissions attached to a ClusterRole
kubectl create clusterrolebinding –clusterrole –serviceaccount :
To create a ClusterRoleBinding between a ClusterRole an a Service Account (similar can be run for RoleBinding and Roles, but in the case of normal RoleBinding we must specify the Namespace)
kubectl create clusterrole –verb –resource
Create a ClusterRole doing X actions on Y resources. Similar can be executed for normal Role.
kubectl auth can-i –as system:serviceaccount:: -n
Checks if the Service Account has permissions to do the specified action on the declared resources from the written Namespace

La entrada Solving CKA exercises – Part 4 se publicó primero en CloudArch.

Solving CKA exercises – Part 3

joaquinjimenezgarcia — Sun, 17 Mar 2024 16:08:14 +0000

One of the most common uses when troubleshooting an application is to redirect the logs somewhere else to not lose them and being able to analyze those later to look for errors or patterns.

Solving these CKA exercises you will learn how to store in a safer place logs from pods even if they have multiple containers and to read them to be able to fix a Deployment which is not working.

Don’t forget to visit killercoda to get more exercises like this one and visit our other posts from Kuberentes

Used commands:

kubectl config set-context –namespace= –current
Set the desired namespace to the current context so we don’t need to declare it in each execution
kubectl logs –all-containers deploy/
Get the logs from each container in the desired Deployment
>
Redirect the output of “first command” to “second command”
kubectl edit deploy/
Edit the Deployment yaml and apply changes when exit

La entrada Solving CKA exercises – Part 3 se publicó primero en CloudArch.

Solving CKA exercises – Part 2

joaquinjimenezgarcia — Wed, 13 Mar 2024 20:47:41 +0000

Let’s continue solving exercises about application troubleshooting!

La entrada Solving CKA exercises – Part 2 se publicó primero en CloudArch.

Solving CKA exercises – Part 1

joaquinjimenezgarcia — Fri, 08 Mar 2024 20:13:52 +0000

Have you recently considered taking the Kubernetes CKA exam or are you just looking for some practice exercises to improve your skills? Then stay tuned because we are solving CKA exercises!

In this new series we will be watching video by video different exercises of the CKA exam and we will be solving them together with their explanation to be able to manage Kubernetes without problems!

You can find this exercise and more labs regarding other exams at https://killercoda.com/ but if you want to learn more about cloud computing, visiting our entries about Kubernetes

La entrada Solving CKA exercises – Part 1 se publicó primero en CloudArch.