We have ~3 ways to build.
This has changed slightly from when the book is published so I'd view this as authoritative.
Insert Binary
This aligns with what is in the book.
You need to build the binary to run somehow and then insert it into a Docker image.
The easiest way to do this is to use the fully automated make system to build the binary and then create a Dockerfile for creating an image.
Create the binary by typing make at the command line. This'll build a docker image and then run it to compile the binary.
Now create a minimal Dockerfile to contain that binary:
FROM alpine
COPY bin/blue/amd64/kuard /kuard
ENTRYPOINT [ "/kuard" ]
Overwrite Dockerfile with this and then run docker build -t kuard-amd64:blue ..
Run with docker run --rm -ti --name kuard --publish 8080:8080 kuard-amd64:blue.
To upload to a registry you'll have to tag it and push to your registry. Refer to your registry documentation for details.
Multi-stage Dockerfile
A new feature of Docker, since the book was published, is a "multi-stage" build.
This is a way to run build multiple images and then copy files between them.
The Dockerfile at the root of this repo is an example of that.
It creates one image to build kuard and then another image for running kuard.
You can easily build an image with docker build -t kuard-amd64:blue ..
Run with docker run --rm -ti --name kuard --publish 8080:8080 kuard-amd64:blue.
To upload to a registry you'll have to tag it and push to your registry. Refer to your registry documentation for details.
Fancy Makefile for automated build and push
This will build and push container images to a registry.
This builds a set of images with "fake versions" (see below) to be able to play with upgrades.
make all-push REGISTRY=<my-gcr-registry>
If you are having trouble, try issuing a make clean to reset stuff.
KeyGen Workload
To help simulate batch workers, we have a synthetic workload of generating 4096 bit RSA keys. This can be configured through the UI or the command line.
--keygen-enable Enable KeyGen workload
--keygen-exit-code int Exit code when workload complete
--keygen-exit-on-complete Exit after workload is complete
--keygen-memq-queue string The MemQ server queue to use. If MemQ is used, other limits are ignored.
--keygen-memq-server string The MemQ server to draw work items from. If MemQ is used, other limits are ignored.
--keygen-num-to-gen int The number of keys to generate. Set to 0 for infinite
--keygen-time-to-run int The target run time in seconds. Set to 0 for infinite
The API is as follows with URLs being relative to <server addr>/memq/server. See pkg/memq/types.go for the data structures returned.
Method
Url
Desc
GET
/stats
Get stats on all queues
PUT
/queues/:queue
Create a queue
DELETE
/queues/:queue
Delete a queue
POST
/queues/:queue/drain
Discard all items in queue
POST
/queues/:queue/enqueue
Add item to queue. Body is plain text. Response is message object.
POST
/queues/:queue/dequeue
Grab an item off the queue and return it. Returns a 204 "No Content" if queue is empty.
Versions
Images built will automatically have the git version (based on tag) applied. In addition, there is an idea of a "fake version". This is used so that we can use the same basic server to demonstrate upgrade scenarios.
Originally (and in the Kubernetes Up & Running book) we had 1, 2, and 3. This confused people so going forward we will be using colors instead: blue, green and purple. This translates into the following container images:
For documentation where you want to demonstrate using versions but use the latest version of this server, you can simply reference gcr.io/kuar-demo/kuard-amd64:blue. You can then demonstrate an upgrade with gcr.io/kuar-demo/kuard-amd64:green.
(Another way to think about it is that :blue is essentially :latest-blue)
We also build versions for arm, arm64, and ppc64le. Just substitute the appropriate architecture in the image name. These aren't as well tested as the amd64 version but seem to work okay.
Development
If you just want to do Go server development, you can build the client as part of a build make. It'll drop the result in to sitedata/built/.
If you want to do both Go server and React.js client dev, you need to do the following:
Have Node installed
In one terminal
cd client
npm install
npm run start
This will start a debug node server on localhost:8081. It'll proxy all unhandled requests to localhost:8080
In another terminal
Ensure that $GOPATH is set to the directory with your go source code and binaries + ensure that $GOPATH is part of $PATH.
This should support live reload of any changes to the client. The Go server will need to be exited and restarted to see changes.
TODO
Make file system browser better. Show size, permissions, etc. Might be able to do this by faking out an index.html as part of the http.FileSystem stuff.
Clean up form for keygen workload. It is too big and the form build doesn't have enough flexibility to really shrink it down.
请发表评论