Why is Tilt broken?
Tilt is broken. What should you do?
There are a lot of ways to get help, depending on how willing you are to get your hands dirty.
Where can I ask questions?
Or you can file an issue.
Why does Tilt crash on startup?
Tilt will print out status information:
- Tilt version
- Operating system
- Docker host and version
- Kubernetes version
- Type of Kubernetes cluster (Docker for Mac, Microk8s, etc)
- Container runtime
This can help you figure out what cluster Tilt thinks you’re using. It’s usually the first thing we ask for when people file issues with Tilt.
What info should I send when I need help?
A snapshot is a link that you can send to someone that will allow them to interactively explore your current Tilt state. This is useful for async debugging and adding context to bug reports. They look like pretty much just like Tilt, but frozen in time:
A snapshot is a frozen “moment-in-time” version of the Tilt UI. In a snapshot you can drill in to specific services, see alerts and Kubernetes events. Pretty much anything you can do in normal Tilt, you can do in a snapshot!
This is also helpful to send along with bug reports.
For more information, see “Share Errors and Cluster State with Snapshots”.
Why does my image build fail in Tilt when it succeeds with
tilt docker -- build ARGS
This will run Docker the same way that Tilt runs Docker.
Tilt automatically enables optimizations that you may not be using by default. For example, if you are using minikube and run:
tilt docker -- build -t image-name .
Tilt may print:
Running Docker command as: DOCKER_HOST=tcp://192.168.99.100:2376 DOCKER_CERT_PATH=/home/nick/.minikube/certs DOCKER_TLS_VERIFY=1 docker build -t image-name .
because it’s running against Minikube’s Docker instance, not your Docker instance.
Why is Tilt using so much CPU or memory?
Please file an issue if Tilt is being a resource hog! We’ve made it a lot better in the last few months but there may still be builds that cause problems.
As of v0.10.26, Tilt exposes the standard Go pprof hooks over HTTP.
To look at a 30-second CPU profile:
go tool pprof http://localhost:10350/debug/pprof/profile?seconds=30
To look at the heap profile:
go tool pprof http://localhost:10350/debug/pprof/heap
This opens a special REPL that lets you explore the data.
web in the REPL to see a CPU graph.
For more information on pprof, see the Go pprof guide.
What does Tilt think is happening right now?
The internal Tilt engine is implemented as a control loop, just like Kubernetes.
Tilt watches lots of different inputs (your Tiltfile, your local source files, and your Kubernetes cluster) for changes. The control loop records these changes in a central state store. Then, Tilt kicks off updates to your cluster based on the state store.
That means all the state Tilt knows about lives in a single place. And you can inspect it!
While Tilt is running in one terminal, open another terminal and run:
tilt dump engine
Tilt will print a JSON representation of everything it knows about your build state and your cluster state.
The Tilt UI has a similar control loop. Run:
tilt dump webview
to see the complete state of the Tilt web UI.
How can I keep track of Tilt usage on my team?
Tilt has an experimental Tiltfile function:
This command takes a string which is a command to run. Tilt will exec this command every minute and pass it on STDIN a series of OpenTelemetry spans in the form of newline-separated JSON objects. These spans representing all of the user’s activity in the last minute, and you can manipulate and ingest them as you will.
For example, you could write a script to send Tilt’s telemetry output to Honeycomb, and invoke it
experimental_telemetry_cmd like so:
The argument to this function is just a shell command, so there’s a lot of flexibility. If for example you wanted to send Tilt’s telemetry output to a script run via a Docker image, you could call:
experimental_telemetry_cmd("docker run --env USER -i my-telemetry-image")
The JSON that gets passed looks like this:
Here are some example scripts that report these spans as:
This is an experimental feature designed for larger companies. If this sounds interesting to you, please reach out.