new driver: kubernetes

Tested with `kind` and GKE.

Note: "nodes" shown in `docker buildx ls` are unrelated to Kubernetes "nodes".
Probably buildx should come up with an alternative term.

Usage:

  $ kind create cluster
  $ export KUBECONFIG="$(kind get kubeconfig-path --name="kind")"

  $ docker buildx create --driver kubernetes --driver-opt replicas=3 --use
  $ docker buildx build -t foo --load .

`--load` loads the image into the local Docker.

Driver opts:

  - `image=IMAGE` - Sets the container image to be used for running buildkit.
  - `namespace=NS` - Sets the Kubernetes namespace. Defaults to the current namespace.
  - `replicas=N` - Sets the number of `Pod` replicas. Defaults to 1.
  - `rootless=(true|false)` - Run the container as a non-root user without `securityContext.privileged`. Defaults to false.
  - `loadbalance=(sticky|random)` - Load-balancing strategy. If set to "sticky", the pod is chosen using the hash of the context path. Defaults to "sticky"

Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>

vendor/k8s.io/client-go/tools/remotecommand/v1.go

@@ -0,0 +1,160 @@
+/*
+Copyright 2015 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package remotecommand
+
+import (
+	"fmt"
+	"io"
+	"io/ioutil"
+	"net/http"
+
+	"github.com/golang/glog"
+	"k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/util/httpstream"
+)
+
+// streamProtocolV1 implements the first version of the streaming exec & attach
+// protocol. This version has some bugs, such as not being able to detect when
+// non-interactive stdin data has ended. See http://issues.k8s.io/13394 and
+// http://issues.k8s.io/13395 for more details.
+type streamProtocolV1 struct {
+	StreamOptions
+
+	errorStream  httpstream.Stream
+	remoteStdin  httpstream.Stream
+	remoteStdout httpstream.Stream
+	remoteStderr httpstream.Stream
+}
+
+var _ streamProtocolHandler = &streamProtocolV1{}
+
+func newStreamProtocolV1(options StreamOptions) streamProtocolHandler {
+	return &streamProtocolV1{
+		StreamOptions: options,
+	}
+}
+
+func (p *streamProtocolV1) stream(conn streamCreator) error {
+	doneChan := make(chan struct{}, 2)
+	errorChan := make(chan error)
+
+	cp := func(s string, dst io.Writer, src io.Reader) {
+		glog.V(6).Infof("Copying %s", s)
+		defer glog.V(6).Infof("Done copying %s", s)
+		if _, err := io.Copy(dst, src); err != nil && err != io.EOF {
+			glog.Errorf("Error copying %s: %v", s, err)
+		}
+		if s == v1.StreamTypeStdout || s == v1.StreamTypeStderr {
+			doneChan <- struct{}{}
+		}
+	}
+
+	// set up all the streams first
+	var err error
+	headers := http.Header{}
+	headers.Set(v1.StreamType, v1.StreamTypeError)
+	p.errorStream, err = conn.CreateStream(headers)
+	if err != nil {
+		return err
+	}
+	defer p.errorStream.Reset()
+
+	// Create all the streams first, then start the copy goroutines. The server doesn't start its copy
+	// goroutines until it's received all of the streams. If the client creates the stdin stream and
+	// immediately begins copying stdin data to the server, it's possible to overwhelm and wedge the
+	// spdy frame handler in the server so that it is full of unprocessed frames. The frames aren't
+	// getting processed because the server hasn't started its copying, and it won't do that until it
+	// gets all the streams. By creating all the streams first, we ensure that the server is ready to
+	// process data before the client starts sending any. See https://issues.k8s.io/16373 for more info.
+	if p.Stdin != nil {
+		headers.Set(v1.StreamType, v1.StreamTypeStdin)
+		p.remoteStdin, err = conn.CreateStream(headers)
+		if err != nil {
+			return err
+		}
+		defer p.remoteStdin.Reset()
+	}
+
+	if p.Stdout != nil {
+		headers.Set(v1.StreamType, v1.StreamTypeStdout)
+		p.remoteStdout, err = conn.CreateStream(headers)
+		if err != nil {
+			return err
+		}
+		defer p.remoteStdout.Reset()
+	}
+
+	if p.Stderr != nil && !p.Tty {
+		headers.Set(v1.StreamType, v1.StreamTypeStderr)
+		p.remoteStderr, err = conn.CreateStream(headers)
+		if err != nil {
+			return err
+		}
+		defer p.remoteStderr.Reset()
+	}
+
+	// now that all the streams have been created, proceed with reading & copying
+
+	// always read from errorStream
+	go func() {
+		message, err := ioutil.ReadAll(p.errorStream)
+		if err != nil && err != io.EOF {
+			errorChan <- fmt.Errorf("Error reading from error stream: %s", err)
+			return
+		}
+		if len(message) > 0 {
+			errorChan <- fmt.Errorf("Error executing remote command: %s", message)
+			return
+		}
+	}()
+
+	if p.Stdin != nil {
+		// TODO this goroutine will never exit cleanly (the io.Copy never unblocks)
+		// because stdin is not closed until the process exits. If we try to call
+		// stdin.Close(), it returns no error but doesn't unblock the copy. It will
+		// exit when the process exits, instead.
+		go cp(v1.StreamTypeStdin, p.remoteStdin, p.Stdin)
+	}
+
+	waitCount := 0
+	completedStreams := 0
+
+	if p.Stdout != nil {
+		waitCount++
+		go cp(v1.StreamTypeStdout, p.Stdout, p.remoteStdout)
+	}
+
+	if p.Stderr != nil && !p.Tty {
+		waitCount++
+		go cp(v1.StreamTypeStderr, p.Stderr, p.remoteStderr)
+	}
+
+Loop:
+	for {
+		select {
+		case <-doneChan:
+			completedStreams++
+			if completedStreams == waitCount {
+				break Loop
+			}
+		case err := <-errorChan:
+			return err
+		}
+	}
+
+	return nil
+}