The Traffic Police 🚨 - Controlling outgoing traffic with mirrord

Posted August 29, 2023 by Alexandre Cavalcante - 7 Min Read

So, you’ve been using mirrord to simplify your development process (if you haven’t, go here! ). Naturally, you want the traffic from the app you’re debugging to go through the cluster environment, so your app can communicate with its clustery pals. There is a problem though: your latest change adds some new columns to the database, and you don’t want to modify the database in the cluster and affect everyone else working on it. You do have a local instance of the database that you can modify, so your app can use that, but you still want it to talk to all the other components in the cluster. So what now? The new outgoing traffic filter feature is here to solve exactly this type of problem!

With the new filter, you can specify whether your app’s outgoing traffic should be sent locally or remotely based on its destination. If we take the example above, with the database running in the cluster as a service app-db, and locally with the same hostname, you can select which database your app will be talking to. Here is a sample config.json for this use case:

  feature": { 
    "network": { 
      "outgoing": { "filter": { "local": ["app-db"] } } 

In this case, when your app tries to resolve the app-db hostname it’ll do so locally (instead of in-cluster), meaning that traffic which would normally go to the cluster’s app-db will be sent to your local database, while all other traffic (that doesn’t match the filter) will keep flowing as you expect, in the cluster. Isn’t this neat? Your cluster’s database remains unmodified, and you can keep working with the other services in your cluster. The filter supports multiple options, listed here .

Let’s look at a more detailed example.

The Setup 📋

The recipe for our example is as follows:

  • 🥺 The uwu-service running in the cluster, which takes normal text and performs uwuification
  • 😏 The uwu-app in-cluster, that sends normal messages to uwu-service, then writes the returned uwu’d versions to a database. This is our target pod and the app we’re working on locally
  • 😀 A messages-db database in the cluster, which the uwu-app normally writes to, but in our debugging session, we’re going to be using a local version of this database
  • 😼 The local uwu-app we are currently working on
  • 😺 A local messages-db, so our schema changes won’t affect devs that rely on the cluster’s message-db

Diagram with the services and traffic flow

Outgoing traffic filter diagram

Our flow here is as follows:

  1. The uwu-app pod receives an HTTP request from the outside world;
  2. mirrord steals this request, forwarding it to the local uwu-app;
  3. Next, the local uwu-app makes a POST HTTP request to the uwu-service in our cluster;
  4. This request goes through the agent pod, before reaching the uwu-service;
  5. uwu-service sends back an HTTP response with the uwu’d version of the message;
  6. The response comes through the agent back to our local uwu-app;
  7. And finally, we save the new uwu message to our local messages-db.

Under normal circumstances, the response from uwu-service would be written to the cluster’s message-db database, but the cluster database doesn’t comply with our new schema, so the write operation would just fail.

The Example 💾

Our main issue is that we want our local app’s outgoing traffic to be sent both locally and remotely, depending on its destination. Let’s see how mirrord helps us solve this.

We start out by running this command, which starts the uwu-app with mirrord:

mirrord exec -f config.json ./uwu-app

Where config.json1 is just:

  "target": “pod/uwu-app”,
  "feature": { 
    "network": { 
      "incoming": "steal", 
      "outgoing": { "filter": { "local": [":8888"] } }

This is enough to start mirrord and the uwu-app that we want to test. You should see something like this:

⠧ mirrord cli starting...
 ⠧ exec
   ✓ ready to launch process
   ✓ layer extracted
   ✓ no operator detected
   ✓ agent pod created
   ✓ pod is ready

Our app is ready, and so is mirrord. Now we can make a request to the uwu-app pod and have it stolen by mirrord (traffic will be stolen from the cluster app to our local app). First let’s get the uwu-app service port:

$ kubectl describe service uwu-app

Name:                     uwu-app
Port:                     <unset>  9999/TCP
TargetPort:               9999/TCP
NodePort:                 <unset>  30032/TCP

Now we can make a request using curl:

$ curl -d \
"Hey, are you enjoying mirrord? Why not star us on github? It would be very cool of you." \

Added new message
hey, 🥺 awe you enjoying miwwowd? 
why n-nyot staw us on g-github? 
it wouwd b-be vewy coow o-of you.
with id 0

And here are the logs from the local messages-db:

[DEBUG messages_db] store; 
new_message="hey, 🥺 awe you enjoying miwwowd? 
             why n-nyot staw us on g-github? 
             it wouwd b-be vewy coow o-of you."
[DEBUG messages_db]
return="Added new message hey, 🥺 awe you enjoying miwwowd?
                          why n-nyot staw us on g-github? 
                          it wouwd b-be vewy coow o-of you. 
        with id 0"
[INFO  actix_web::middleware::logger] "POST /store/0 HTTP/1.1" 200 129 "-" "-" 0.000243

The cluster’s messages-db doesn’t even know that any of this ever happened! The logs there are so clean you could see your own reflection, meaning that no traffic ever hits the remote messages-db. To recap what happened: We captured incoming traffic from the remote uwu-app and sent it to our local app instead Our local uwu-app sent a request to the cluster’s uwu-service and got back its uwufied version. Finally, our local app wrote the uwufied message to a local messages-db.

Without mirrord, you would either need to run everything locally, including the uwu-service that turns normal messages into uwu’d versions, or deploy your uwu-app changes to the cluster, and potentially mess it up for the other devs.

With mirrord you don’t have to choose. You can have youw cake and eat it t-too!!

The Feature 🔃

With the new outgoing traffic filter feature (say it fast 3 times), you can pick and choose from where outgoing traffic is sent from. Paired with other mirrord features, such as the incoming HTTP traffic filter, you get unprecedented control over how your app communicates with your cluster.

The example we just saw was really small, but you can imagine if the app we’re working on had to talk with a bunch of services, how painful it would be to set up everything locally, or how messy it could get if everyone had to take turns using the shared environment, just because the code writes to the same database. The outgoing traffic filter makes it easy to test changes on your shared environment without impacting the rest of your team. No more going around asking your colleagues if anyone needs to use the database in the next couple of hours..

Get mirrord 🪞

Ready to improve your workflow? Want to start using the outgoing traffic filter? Check out the new configuration and start using it today!

What? You’re not using mirrord?! You have to deploy every small change, just so you can test it in cloud conditions? Let me stop you right there! Just download and install mirrord, and test your code in the context of your Kubernetes environment, right from the comfort of your own local machine. You can get it from here , what are you waiting for?

Here is a quick start guide link. Want to talk to an actual human? Then hop in our Discord (no chatbots here, I promise). Prefer more asynchronous communication? Open up an issue or a discussion on GitHub . We love feedback, don’t be shy!

  1. We’re setting incoming traffic to steal, but mirrord also supports traffic mirroring, and traffic stealing with a filter. You can see all the options here↩︎

Your subscription could not be saved. Please try again.
Your subscription has been successful.

sign up to our blog post for updates and news


Alexandre Cavalcante

Software Engineer @ MetalBear.

You may also like...

mirrord 3.0 is out!