Prism Long Term Storage

prism-long-term-storage alerts and suggested responses/actions to take when this service alerts you.

Prism long term storage is meant to propagate data from our customers through the Prism architecture into a storage area for an infinite period of time.

How does prism-long-term-storage work?

Prism long term storage uses the AWS S3 connector for Kafka Connect to route data from our Kafka cluster into AWS S3. Customer data is ingested by a Kafka Producer and Kafka Connect functions as a Kafka Consumer.

Partitioning

We are using a custom partition scheme. The source is located in /java. This uses the metric_date field in each incoming Avro message to route these messages to a particular S3 bucket and file. Each S3 file contains a flushSize number of Avro records for that particular day. For instance, with a metric_date of 2018-01-04, each record will be routed to a file in the bucket: s3://prod-prism-lts/topics/ac-user-event/year=2018/month=01/day=04/. The flush size can be any number that is reasonable. We are starting with a flush size of 250 as a default value. This can be updated in prism-lts/values.yaml.

Local Development

Initial Setup

• minikube start
• Setup AWS secrets, with kubectl create secret generic aws-s3-creds --from-file ~/.aws/credentials
• Setup Quay credentials, go to https://quay.io/organization/stsatlas?tab=robots, select stsatlas+platform_deployer and follow the directions for Kubernetes Secret
  • Rename the secret (in the file downloaded) to be quay-sts
  • Run kubectl create -f /path/to/stsatlas-platform-deployer-secret.yml
• Install helm with brew install kubernetes-helm
• Run helm init
• Get dependencies for prism-lts by running helm dep update prism-lts && helm dep build prism-lts from the root of this project

Updating the local-charts [environment specific] values

If you update any environment specific values or the local kafka chart, you will need to update and build dependencies in prism-lts again.

• Run helm dep update
• Run helm dep build

Testing Prism-lts

• Run ./bin/kafka-connect-test
• NOTE: requires $QUAY_USERNAME and $QUAY_PASSWORD to be set OR the quay-sts secret.
• NOTE: requires $AWS_ACCESS_KEY_ID and $AWS_SECRET_ACCES_KEY to be set OR the aws-s3-creds secret.

Running Prism-lts

<release_name> is how you will refer to your installation of the helm chart in your local cluster.

• Your first run will use the command: helm install --replace --name=<release name> ./prism-lts --set tags.prism-lts-local-values=true
• After the first run, use the command: helm upgrade --install <release_name> ./prism-lts --set tags.prism-lts-local-values=true

Changing java code

If you change the java source code for partitioners, make sure to build a jar, new image and tag appropriately to use the new image.

1. Build the Jar under java directory using lein jar
2. Make sure your docker context is pointing at minikube. eval $(minikube docker-env)
3. Make sure you are logged into quay.io on minikube. If not already logged in, you can use docker login quay.io to login.
4. Build the docker image docker build -t quay.io/stsatlas/prism-lts:my-new-change . (No need to push the image up to quay.io)
5. Build the docker image docker build -t quay.io/stsatlas/bash-curl:my-new-change -f DockerfileCurl . if required (No need to push the image up to quay.io)
6. Upgrade the release - update the release_name and tag name helm upgrade --install <release_name> ./prism-lts --set tags.prism-lts-local-values=true --set kafkaConnect.tag=my-new-change
7. Then use ./bin/send_a_bunch_of_data <topic_name> to send messages to kafka and test the flow

To Send Data to the REST Proxy

• setup port forwarding for the local kafka rest proxy: kubectl port-forward $(kubectl get po -o name -l app=local-kafka-rest --sort-by='.metadata.creationTimestamp' | cut -d \/ -f 2 | tail -n 1) 8082:8082
• view the logs with: kubectl logs $(kubectl get po -o name -l app=prism-lts --sort-by='.metadata.creationTimestamp' | cut -d \/ -f 2 | tail -n 1) -c prism-lts
• finally, run: ./bin/send_a_bunch_of_data <topic name> <desired num messages> to send a bunch of messages onto the local kafka bus

Changing metadata labels or annotations

If you change metadata labels or annotations, helm does not know that the previous release running in your minikube cluster is the same app. Delete your old release by running helm delete <release_name> --purge The --purge flag removes references that helm has to track your release.