Test latency and throughput in Redpanda using Redpanda CLI (rpk):

To test latency and throughput in Redpanda, follow these steps. You'll use tools like rpk (Redpanda’s CLI) or existing Kafka benchmarking tools (since Redpanda is Kafka API-compatible). Below are approaches to measure both metrics effectively:

Prerequisites

1. Install Redpanda:
• Follow the installation instructions from Redpanda’s official website to set up a Redpanda cluster (single-node or multi-node).
2. Install Redpanda CLI (rpk):
• rpk (Redpanda's CLI) is essential for running benchmarks and managing the cluster. Install it as per the official instructions:

            Install RedPanda 

            apt install redpanda

3. Kafka-compatible Tools:
• Since Redpanda is compatible with Kafka, tools like Kafka Producer Performance (kafka-producer-perf-test.sh) and Kafka Consumer Performance (kafka-consumer-perf-test.sh) can be used.

---

1. Testing Throughput

Throughput measures the rate of data transfer in terms of messages per second or megabytes per second.

a) Using rpk to Measure Throughput

• rpk has built-in benchmarking capabilities to test the producer and consumer throughput.
• Producer Throughput Test: You can generate test data and measure the throughput of producing messages to a Redpanda topic.

bash

 rpk topic produce --brokers localhost:9092 --key test-key --value test-value -n 10000 --rate 500

Here:

• --brokers: The address of your Redpanda broker.
• -n 10000: Number of messages to send.
• --rate 500: Send messages at a rate of 500 messages per second.
  
  
• Consumer Throughput Test: Consume messages from a topic to measure how fast consumers can process them.

bash

 rpk topic consume test-topic --offset oldest --num 10000

This will consume 10,000 messages and show you the processing speed.

b) Using Kafka Performance Test Scripts

If you want to simulate heavy traffic and measure throughput:

• Producer Throughput (Kafka):

bash

 

kafka-producer-perf-test.sh \

    --topic test-topic \

    --num-records 100000 \

    --record-size 1024 \

    --throughput -1 \

    --producer-props bootstrap.servers=localhost:9092

Here:

• --num-records 100000: Sends 100,000 messages.
• --record-size 1024: Each message is 1024 bytes.
• --throughput -1: No limit on throughput (send as fast as possible).
• --producer-props: Kafka producer properties, including the Redpanda broker address.
  
  
• Consumer Throughput (Kafka):

bash

 

kafka-consumer-perf-test.sh \

    --broker-list localhost:9092 \

    --topic test-topic \

    --messages 100000

This will consume 100,000 messages from the topic and provide throughput results.

---

2. Testing Latency

Latency measures the time taken to deliver a message from producer to consumer.

a) Using rpk to Measure Latency

To test the latency of messages, you can produce and consume messages while observing the latency of message delivery.

• Producer Latency Test: Measure the time it takes for each message to be produced:

bash

 rpk topic produce --brokers localhost:9092 --key test-key --value test-value -n 10000 --latency

This command will measure the time each message takes to be delivered to the broker.

• End-to-End Latency Test: You can measure end-to-end latency by producing and consuming messages in real-time. This is done by observing the time when a message is produced and when it's consumed.
• Produce messages to a topic:

bash

 rpk topic produce test-topic -n 10000 --rate 100 --value "Message with latency test"

• At the same time, start a consumer:

bash

 rpk topic consume test-topic --offset oldest

• Compare the timestamps of when messages were produced and when they were consumed.

b) Using Kafka Tools

To perform a detailed latency test using Kafka’s producer performance tool, you can look at how long it takes to acknowledge a sent message.

• Producer Latency (Kafka):

bash

 

kafka-producer-perf-test.sh \

    --topic test-topic \

    --num-records 10000 \

    --record-size 1024 \

    --throughput 500 \

    --producer-props bootstrap.servers=localhost:9092 \

    --print-metrics

• --print-metrics: This will print out detailed producer metrics, including message send latency.

---

3. Benchmarking with Multiple Brokers

If you're using a multi-node Redpanda cluster, you can stress-test the system by producing/consuming from multiple nodes.

• Modify the --brokers argument to list all the brokers in your Redpanda cluster:

bash

 

rpk topic produce --brokers broker1:9092,broker2:9092 --topic test-topic -n 100000

This helps to measure latency and throughput across multiple brokers in a real-world distributed setup.

---

4. Monitoring Performance Metrics

• rpk metrics: Use rpk to observe performance and resource usage metrics in real-time.

bash

 

rpk cluster info

rpk metrics stream

This gives you detailed statistics like message throughput, disk usage, and network metrics.

---

5. Cloud-Based Testing

If you're testing Redpanda in a cloud environment, consider using monitoring solutions like Prometheus and Grafana to track latency, throughput, and system metrics (CPU, memory, disk I/O) during the test.

Conclusion:

• Throughput can be measured using rpk or Kafka’s producer/consumer performance scripts by stressing the cluster with a high volume of messages and measuring message rates.
• Latency can be measured using tools like rpk to observe end-to-end message delivery times or producer acknowledgment times.

Make sure to run tests in a production-like environment to get accurate insights into how Redpanda performs under load.