Monitoring

An operator can watch the Lightsail container service and the MSK brokers. Every command on this page only reads; the runbook says what to do when something looks wrong.

The Lightsail service

The service runs on Lightsail's micro power (0.25 vCPU and 1 GB of memory), and Lightsail records its CPU and memory use as percentages of that. Each open consumer instance holds a Kafka consumer in the service's memory. The Lightsail console shows these metrics on the service's page. From the CLI, memory use in 5-minute periods:

aws lightsail get-container-service-metric-data --region us-east-1 \
    --service-name diaspora-web-service --metric-name MemoryUtilization \
    --start-time <start-time> --end-time <end-time> \
    --period 300 --statistics Average Maximum

The times are in ISO 8601, such as 2026-10-26T06:00:00Z. --metric-name CPUUtilization gives the CPU use. The history starts at the current deployment: AWS says that a new deployment makes the existing utilization metrics disappear, so every deploy, rollback or credential rotation clears them.

The container's log holds gunicorn's and the service's output for the last three days (Read the logs). aws lightsail get-container-services shows whether the service is running and which deployment is active (Check a deploy).

On GitHub Actions, the deployment workflow's runs show each deploy, and the tests workflow's runs show the live tests, which run nightly and on every push to main. The live tests run against the AWS account and the MSK cluster, not against the deployed service.

The MSK brokers

Each broker's CPUCreditBalance metric in CloudWatch (namespace AWS/Kafka, dimensions Cluster Name and Broker ID) is the number of CPU credits it has left. AWS says that running out "can have a negative impact on your cluster's performance" (Broker CPU credits).

The Kafka console

The Diaspora Admin Console is a Kafbat UI instance that shows the cluster's topics, consumer groups and consumer lag. Its statistics can be up to 5 minutes old.