Skip to main content

Autoscale Zilla pods with Prometheus metrics

Autoscale Zilla pods with Prometheus metrics

In this guide, run zilla in kubernetes and autoscale the number of pods based on prometheus metrics. The Kubernetes horizontal pod autoscaler is set up to enable the zilla deployment to scale from 1 to 5 pods with the goal of an average load of 10 active connections per pod.

Tl;Dr

Download and run the Zilla kubernetes.prometheus.autoscale cookbook using this install script. It will start Zilla and everything you need for this guide.

wget -qO- https://raw.githubusercontent.com/aklivity/zilla-examples/main/startup.sh | sh -s -- kubernetes.prometheus.autoscale

Note

Alternatively, download kubernetes.prometheus.autoscale and follow the README yourself.

Prerequisites

Autoscaling Zilla

Run the commands from this guide from a shell in the kubernetes.prometheus.autoscale directory that you downloaded.

curl -d "Hello, world" -X "POST" http://localhost:7114

output:

Hello, world

The initial status is:

  • no open connections
  • the value of the stream_active_received metric should be 0
  • there should be 1 zilla pod in the deployment

If the kubernetes custom metrics API response does not appear correctly please wait a few seconds and try again before proceeding further.

./check_metric.sh

output:

The value of stream_active_received metric
------------------------------------------

Prometheus API:
{
...
        "metric": {
          "__name__": "stream_active_received",
        },
        "value": [
          1683013504.619, # timestamp
          "0" # value
...
}

Kubernetes custom metrics API:
{
...
      "metricName": "stream_active_received",
      "value": "0",
...
}

The zilla deployment should consist of 1 pod.

./check_hpa.sh

output:

The status of horizontal pod autoscaling
----------------------------------------

HorizontalPodAutoscaler:
NAME    REFERENCE          TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
zilla   Deployment/zilla   0/10      1         5         1          4m24s

Deployment:
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
zilla   1/1     1            1           4m25s

Pods:
NAME                     READY   STATUS    RESTARTS   AGE
zilla-6db8d879f5-2wxgw   1/1     Running   0          4m25s

Open 21 connections to zilla as instances of netcat in the background.

for i in `seq 1 21`; do nc localhost 7114 &; done

output:

[42] 88886
[43] 88887
[44] 88888
...

There should be 21 open connections in the background now.

ps auxw | grep "nc localhost 7114" | grep -v grep | wc -l

output:

21

Wait for a few seconds so the metrics get updated. The value of stream_active_received metric should be 21 for one of the pods.

./check_metric.sh

output:

The value of stream_active_received metric
------------------------------------------

Prometheus API:
{
...
        "metric": {
          "__name__": "stream_active_received",
        },
        "value": [
          1683013504.619, # timestamp
          "21" # value
...
}

Kubernetes custom metrics API:
{
...
      "metricName": "stream_active_received",
      "value": "21",
...
}

Wait for a minute so the autoscaler can catch up. The zilla deployment should be soon scaled up to 3 pods.

./check_hpa.sh

output:

The status of horizontal pod autoscaling
----------------------------------------

HorizontalPodAutoscaler:
NAME    REFERENCE          TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
zilla   Deployment/zilla   7/10      1         5         3          7m14s

Deployment:
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
zilla   3/3     3            3           7m15s

Pods:
NAME                     READY   STATUS    RESTARTS   AGE
zilla-6db8d879f5-2wxgw   1/1     Running   0          7m15s
zilla-6db8d879f5-9bnkh   1/1     Running   0          75s
zilla-6db8d879f5-fmgqx   1/1     Running   0          75s

Open 21 connections to zilla as instances of netcat in the background.

for i in `seq 1 21`; do nc localhost 7114 &; done

output:

[77] 77775
[78] 77776
[79] 77777
...

There should be 42 open connections in the background now.

```bash
ps auxw | grep "nc localhost 7114" | grep -v grep | wc -l

output:

42

Wait for a few seconds so the metrics get updated. The value of stream_active_received metric should be 42 for one of the pods.

./check_metric.sh

output:

The value of stream_active_received metric
------------------------------------------

Prometheus API:
{
...
        "metric": {
          "__name__": "stream_active_received",
        },
        "value": [
          1683013504.619, # timestamp
          "42" # value
...
}

Kubernetes custom metrics API:
{
...
      "metricName": "stream_active_received",
      "value": "42",
...
}

Wait for a minute so the autoscaler can catch up. The zilla deployment should be soon scaled up to 5 pods.

./check_hpa.sh

output:

The status of horizontal pod autoscaling
----------------------------------------

HorizontalPodAutoscaler:
NAME    REFERENCE          TARGETS    MINPODS   MAXPODS   REPLICAS   AGE
zilla   Deployment/zilla   8400m/10   1         5         5          12m

Deployment:
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
zilla   5/5     5            5           12m

Pods:
NAME                     READY   STATUS    RESTARTS   AGE
zilla-6db8d879f5-2wxgw   1/1     Running   0          12m
zilla-6db8d879f5-9bnkh   1/1     Running   0          6m3s
zilla-6db8d879f5-fmgqx   1/1     Running   0          6m3s
zilla-6db8d879f5-g74hl   1/1     Running   0          63s
zilla-6db8d879f5-q5fmm   1/1     Running   0          63s

Shut down all running netcat instances.

ps auxw | grep "nc localhost 7114" | grep -v grep | awk '{print $2}' | xargs kill

output:

[23]  + 55555 terminated  nc localhost 7114
[22]  + 55554 terminated  nc localhost 7114
[21]  + 55553 terminated  nc localhost 7114
...

There should be no open connections in the background now.

```bash
ps auxw | grep "nc localhost 7114" | grep -v grep | wc -l

output:

0

Wait for a few seconds so the metrics get updated. The value of stream_active_received metric should be 0 for all pods.

./check_metric.sh

output:

The value of stream_active_received metric
------------------------------------------

Prometheus API:
{
...
        "metric": {
          "__name__": "stream_active_received",
        },
        "value": [
          1683013504.619, # timestamp
          "0" # value
...
}

Kubernetes custom metrics API:
{
...
      "metricName": "stream_active_received",
      "value": "0",
...
}

Wait for a a minute so the autoscaler can catch up. The zilla deployment should be soon scaled down to 1 pod.

```bash
./check_hpa.sh

output:

The status of horizontal pod autoscaling
----------------------------------------

HorizontalPodAutoscaler:
NAME    REFERENCE          TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
zilla   Deployment/zilla   0/10      1         5         1          14m

Deployment:
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
zilla   1/1     1            1           14m

Pods:
NAME                     READY   STATUS    RESTARTS   AGE
zilla-6db8d879f5-2wxgw   1/1     Running   0          14m

Remove the running namespace

Find the path to the teardown.sh script(s) in the use the teardown script(s) to clean up section of the example output and run it.

© aklivity, inc. 2023-2024