Add release v1.9.8

Allow any non-local traffic onto the weave bridge

.gitignore

@@ -70,3 +70,6 @@ test/coverage.*
 *qemu-*-static
 terraform.tfstate
 terraform.tfstate.backup
+
+# Website
+site-build

.ruby-version

@@ -0,0 +1 @@
+2.3.4

CHANGELOG.md

@@ -1,3 +1,49 @@
+## Release 1.9.8
+
+Bug fixes and minor improvements
+
+* Fix weave-npc blocking NodePort and any other non-local access #3011,#3014
+* Fix bug where IPAM would duplicate a fixed IP address assigned via Docker plugin #3003,#3010
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/56?closed=1).
+
+
+## Release 1.9.7
+
+This is identical to 1.9.6 with one additional bug-fix:
+
+* weave-npc would block everything if `kubelet --hostname-override` was used #2995,#2996
+
+## Release 1.9.6
+
+Bug fixes and minor improvements
+
+* Ensure that Kubernetes pods can contact a service implemented within
+  the same pod, by turning on "hairpin mode". This is required because
+  of a quiet change between Kubernetes 1.5 and 1.6. #2993
+* Network Policy Controller (`weave-npc`) now checks local addresses
+  only, so it doesn't interfere with cross-cluster traffic. It should
+  be more efficient too #2622,#2973,#2979
+* Stop reporting back to Kubernetes any issues encountered when
+  deleting a pod's network interface. This is required because
+  of a quiet change between Kubernetes 1.5 and 1.6. #2921,#2928
+* Fixed an issue whereby `weave-npc` couldn't start because one
+  `ipset` was referring to another one and could not be destroyed #2915,#2949
+* Improved the code which checks whether the kernel supports `ipset` #2934,#2935
+* `weave-npc` now creates ipsets with only valid xml characters in the
+  name #2958,#2959
+
+Build and Testing
+
+* In build container use cross-compilers from debian package
+  repository, so they match other components #2940
+* Pin the version of the linting tool `shfmt` so the set of things it
+  checks is stable #2987
+* Fix lint error in script that runs smoke-tests #2962
+* Moved website publishing from Wordpress to Netlify #2986
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/55?closed=1).
+
 ## Release 1.9.5
 
 Bug fixes and minor improvements

Makefile

@@ -378,3 +378,14 @@ integration-tests: all testrunner
 #     integration-tests
 #
 	RUNNER_ARGS="-parallel" ./test/run-integration-tests.sh
+
+# This target is only intended for use in Netlify CI environment for generating preview pages on feature branches and pull requests.
+# We need to obtain website repository (checked out under `site-build`) and place `site` directory into the context (`site-build/_weave_net_docs`).
+# We then run make in `site-build` and Netlify will publish the output (`site-build/_site`).
+netlify-site-preview:
+	@mkdir -p site-build
+	@curl --user $(WEBSITE_GITHUB_USER) --silent 'https://codeload.github.com/weaveworks/website-next/tar.gz/$(WEBSITE_BRANCH)' \
+	  | tar --extract --gunzip --directory site-build --strip 1
+	@cp -r site site-build/_weave_net_docs
+	@$(MAKE) -C site-build netlify_ensure_install
+	@$(MAKE) -C site-build BUILD_ENV=netlify

bin/circle-deploy-issues

@@ -1,43 +0,0 @@
-#!/bin/bash
-
-set -e
-set -o pipefail
-
-source "$STATE"
-
-site_changes() {
-    curl -s https://api.github.com/repos/weaveworks/weave/pulls/$1/files |
-        jq -r '.[]|.status + " " + .filename' |
-        grep -oP '^(added|modified|removed) site/.*(?=\.md$)'
-}
-
-format_comment() {
-    if [ ! -d .git ] || ! BRANCH=$(git symbolic-ref --short HEAD) || [ -z "$BRANCH" ] ; then
-        echo "Could not determine branch" >&2
-        return 1
-    fi
-
-    echo "#### Documentation changes:"
-    while read ACTION PAGE; do
-        echo "$ACTION [${PAGE#site/}]($WP_DEV_URL/docs/net/${BRANCH#issues/}/${PAGE#site/}/)"
-    done | sort
-}
-
-post_comment() {
-    jq -s -R '. | tostring | { body: . }' |
-        curl -s -d @- \
-            -H "Authorization: Bearer $WP_GH_COMMENT_TOKEN" \
-            -H 'Content-Type: application/json' \
-            https://api.github.com/repos/weaveworks/weave/issues/$GH_PR/comments > /dev/null
-}
-
-cd $SRCDIR
-bin/install-wordepress v1.0.0
-bin/publish-site "$WP_DEV_URL" "$WP_DEV_USER" "$WP_DEV_PASSWORD"
-
-if [ -n "$WP_GH_COMMENT_TOKEN" -a -n "$GH_PR" ] ; then
-    echo ">>> Adding doc changelog to GitHub PR $GH_PR"
-    if CHANGES=$(site_changes $GH_PR) && [ -n "$CHANGES" ] ; then
-        echo "$CHANGES" | format_comment | post_comment
-    fi
-fi

bin/circle-deploy-master

@@ -11,7 +11,3 @@ if [ -n "$TEST_AND_PUBLISH" ] ; then
     # CircleCI doesn't let us install the qemu module to do ARM docker builds
     make publish ALL_ARCHITECTURES=amd64 UPDATE_LATEST=latest-only
 fi
-
-cd $SRCDIR
-bin/install-wordepress v1.0.0
-bin/publish-site "$WP_DEV_URL" "$WP_DEV_USER" "$WP_DEV_PASSWORD"

bin/circle-deploy-release

@@ -1,7 +0,0 @@
-#!/bin/bash
-
-set -e
-
-cd $SRCDIR
-bin/install-wordepress v1.0.0
-bin/publish-site "$WP_LIVE_URL" "$WP_LIVE_USER" "$WP_LIVE_PASSWORD"

bin/install-wordepress

@@ -1,12 +0,0 @@
-#!/bin/bash
-
-if [ $# -ne 1 ] ; then
-    echo USAGE: install-wordepress [COMMIT-ISH] >&2
-    exit 1
-fi
-
-go get github.com/weaveworks/wordepress
-cd /home/ubuntu/src/github.com/weaveworks/wordepress
-git checkout $1
-cd cmd/wordepress
-go get

bin/publish-site

@@ -1,46 +0,0 @@
-#!/bin/bash
-
-set -e
-set -o pipefail
-
-PRODUCT=net
-
-fatal() {
-    echo "$@" >&2
-    exit 1
-}
-
-if [ ! -d .git ] ; then
-    fatal "Current directory is not a git clone"
-fi
-
-if ! BRANCH=$(git symbolic-ref --short HEAD) || [ -z "$BRANCH" ] ; then
-    fatal "Could not determine branch"
-fi
-
-case "$BRANCH" in
-    issues/*)
-        VERSION="${BRANCH#issues/}"
-        TAGS="$VERSION"
-        ;;
-    *)
-        if echo "$BRANCH" | grep -qE '^[0-9]+\.[0-9]+' ; then
-            DESCRIBE=$(git describe --match 'v*')
-            if ! VERSION=$(echo "$DESCRIBE" | grep -oP '(?<=^v)[0-9]+\.[0-9]+\.[0-9]+') ; then
-                fatal "Could not infer latest $BRANCH version from $DESCRIBE"
-            fi
-            TAGS="latest"
-        else
-            VERSION="$BRANCH"
-            TAGS="$VERSION"
-        fi
-        ;;
-esac
-
-for TAG in $TAGS ; do
-    echo ">>> Publishing $PRODUCT $VERSION to $1/docs/$PRODUCT/$TAG"
-    wordepress \
-        --url "$1" --user "$2" --password "$3" \
-        --product "$PRODUCT" --version "$VERSION" --tag "$TAG" \
-        publish site
-done

build/Dockerfile

@@ -20,13 +20,12 @@ ENV GCC_CROSSCOMPILERS \
 
 # Temporarily add the Ubuntu repositories, because we're going to install gcc cross-compilers from there
 # Install the build-essential and crossbuild-essential-ARCH packages
-RUN echo "deb http://archive.ubuntu.com/ubuntu xenial main universe" > /etc/apt/sources.list.d/cgocrosscompiling.list \
-  && apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 40976EAF437D05B5 3B4FE6ACC0B21F32 \
+RUN echo "deb http://emdebian.org/tools/debian/ jessie main" > /etc/apt/sources.list.d/cgocrosscompiling.list \
+  && curl http://emdebian.org/tools/debian/emdebian-toolchain-archive.key | apt-key add - \
+  && for platform in ${DEB_CROSSPLATFORMS}; do dpkg --add-architecture $platform; done \
   && apt-get update \
   && apt-get install -y build-essential \
   && for platform in ${DEB_CROSSPLATFORMS}; do apt-get install -y crossbuild-essential-${platform}; done \
-  && rm /etc/apt/sources.list.d/cgocrosscompiling.list \
-  && apt-get update \
   && apt-get clean && rm -rf /var/lib/apt/lists/*
 
 # Install some required packages
@@ -41,11 +40,15 @@ RUN apt-get update \
 		flex \
 		bison
 
+RUN curl -fsSLo shfmt https://github.com/mvdan/sh/releases/download/v1.3.0/shfmt_v1.3.0_linux_amd64 && \
+	echo "b1925c2c405458811f0c227266402cf1868b4de529f114722c2e3a5af4ac7bb2  shfmt" | sha256sum -c && \
+	chmod +x shfmt && \
+	mv shfmt /usr/bin
+
 # Install common Go tools
 RUN go get \
 	github.com/golang/lint/golint \
 	github.com/fzipp/gocyclo \
-	gopkg.in/mvdan/sh.v1/cmd/shfmt \
 	github.com/fatih/hclfmt \
 	github.com/client9/misspell/cmd/misspell
 

circle.yml

@@ -59,18 +59,8 @@ teardown:
         parallel: true
 
 deployment:
-  release:
-    branch: /[0-9]+\.[0-9]+/
-    owner: weaveworks
-    commands:
-      - "$SRCDIR/bin/circle-deploy-release"
   master:
     branch: master
     owner: weaveworks
     commands:
       - "$SRCDIR/bin/circle-deploy-master"
-  issues:
-    branch: /.*/
-    owner: weaveworks
-    commands:
-      - "$SRCDIR/bin/circle-deploy-issues"

docs/weavenpc-design.md

@@ -77,8 +77,10 @@ iptables -A WEAVE-NPC -m state --state NEW -j WEAVE-NPC-INGRESS
 
 To direct traffic into the policy engine:
 
+```
 iptables -A FORWARD -o weave -m physdev ! --physdev-out vethwe-bridge -j WEAVE-NPC
 iptables -A FORWARD -o weave -m physdev ! --physdev-out vethwe-bridge -j DROP
+```
 
 Note this only affects traffic which egresses the bridge on a physical
 port which is not the Weave Net router - in other words, it is

ipam/allocator_test.go

@@ -9,6 +9,7 @@ import (
 
 	"github.com/stretchr/testify/require"
 
+	"github.com/weaveworks/weave/api"
 	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
 	"github.com/weaveworks/weave/testing/gossip"
@@ -30,6 +31,10 @@ func (alloc *Allocator) SimplyClaim(ident string, cidr address.CIDR) error {
 	return alloc.Claim(ident, cidr, true, true, returnFalse)
 }
 
+func (alloc *Allocator) AnonymousClaim(cidr address.CIDR) error {
+	return alloc.Claim(api.NoContainerID, cidr, false, true, returnFalse)
+}
+
 func TestAllocFree(t *testing.T) {
 	const (
 		container1 = "abcdef"
@@ -170,6 +175,8 @@ func TestAllocatorClaim(t *testing.T) {
 		universe   = "10.0.3.0/24"
 		testAddr1  = "10.0.3.5/24"
 		testAddr2  = "10.0.4.2/24"
+		testAddr4  = "10.0.2.3/23"
+		subnet4    = "10.0.2.0/23"
 		testPre    = "10.0.3.1/24"
 	)
 
@@ -203,6 +210,9 @@ func TestAllocatorClaim(t *testing.T) {
 	// claim for a different container should fail
 	err = alloc.SimplyClaim(container1, addr1)
 	require.Error(t, err)
+	// claim for an anonymous container, will assume it's the same one
+	err = alloc.AnonymousClaim(addr1)
+	require.NoError(t, err)
 	// claiming the address allocated on the other peer should fail
 	err = alloc.SimplyClaim(container1, address.MakeCIDR(subnet, addrx))
 	require.Error(t, err, "claiming address allocated on other peer should fail")
@@ -215,6 +225,15 @@ func TestAllocatorClaim(t *testing.T) {
 	addr2, _ := address.ParseCIDR(testAddr2)
 	err = alloc.SimplyClaim(container1, addr2)
 	require.NoError(t, err)
+	err = alloc.SimplyClaim(container3, addr2)
+	require.Error(t, err)
+	// Anonymous claim outside of our universe
+	addr4, _ := address.ParseCIDR(testAddr4)
+	subnet4cidr, _ := address.ParseCIDR(subnet4)
+	require.NoError(t, alloc.AnonymousClaim(addr4))
+	addr5, err := alloc.Allocate(api.NoContainerID, subnet4cidr, false, returnFalse)
+	require.NoError(t, err)
+	require.NotEqual(t, addr4.Start(), addr5)
 }
 
 func (alloc *Allocator) pause() func() {

ipam/claim.go

@@ -40,10 +40,29 @@ func (c *claim) Try(alloc *Allocator) bool {
 		return true
 	}
 
+	addOwned := func() {
+		if c.ident == api.NoContainerID {
+			alloc.addOwned(c.cidr.Addr.String(), c.cidr, c.isContainer)
+		} else {
+			alloc.addOwned(c.ident, c.cidr, c.isContainer)
+		}
+	}
+
 	if !alloc.ring.Contains(c.cidr.Addr) {
-		// Address not within our universe; assume user knows what they are doing
 		alloc.infof("Address %s claimed by %s - not in our range", c.cidr, c.ident)
-		alloc.addOwned(c.ident, c.cidr, c.isContainer)
+		previousOwner := alloc.findOwner(c.cidr.Addr)
+		switch {
+		case previousOwner == "":
+			addOwned()
+		case previousOwner == c.ident: // already owned by this ID
+		case c.ident == api.NoContainerID: // already owned by anonymous container
+			// do nothing (no automatic fall-through in Go)
+		case !alloc.dead[previousOwner].IsZero(): // already owned by dead container
+			alloc.removeOwned(previousOwner, c.cidr.Addr)
+			addOwned()
+		default:
+			c.sendResult(fmt.Errorf("address %s already in use by %s", c.cidr, previousOwner))
+		}
 		c.sendResult(nil)
 		return true
 	}
@@ -85,11 +104,7 @@ func (c *claim) Try(alloc *Allocator) bool {
 		// Unused address, we try to claim it:
 		if err := alloc.space.Claim(c.cidr.Addr); err == nil {
 			alloc.debugln("Claimed", c.cidr, "for", c.ident)
-			if c.ident == api.NoContainerID {
-				alloc.addOwned(c.cidr.Addr.String(), c.cidr, c.isContainer)
-			} else {
-				alloc.addOwned(c.ident, c.cidr, c.isContainer)
-			}
+			addOwned()
 			c.sendResult(nil)
 		} else {
 			c.sendResult(err)

net/veth.go

@@ -113,7 +113,7 @@ func interfaceExistsInNamespace(netNSPath string, ifName string) bool {
 
 // NB: This function can be used only by a process that terminates immediately
 //     after calling the function as it changes netns via WithNetNSLinkUnsafe.
-func AttachContainer(netNSPath, id, ifName, bridgeName string, mtu int, withMulticastRoute bool, cidrs []*net.IPNet, keepTXOn bool) error {
+func AttachContainer(netNSPath, id, ifName, bridgeName string, mtu int, withMulticastRoute bool, cidrs []*net.IPNet, keepTXOn bool, hairpinMode bool) error {
 	ns, err := netns.GetFromPath(netNSPath)
 	if err != nil {
 		return err
@@ -131,7 +131,7 @@ func AttachContainer(netNSPath, id, ifName, bridgeName string, mtu int, withMult
 			id = id[:maxIDLen] // trim passed ID if too long
 		}
 		name, peerName := vethPrefix+"pl"+id, vethPrefix+"pg"+id
-		_, err := CreateAndAttachVeth(name, peerName, bridgeName, mtu, keepTXOn, func(veth netlink.Link) error {
+		veth, err := CreateAndAttachVeth(name, peerName, bridgeName, mtu, keepTXOn, func(veth netlink.Link) error {
 			if err := netlink.LinkSetNsFd(veth, int(ns)); err != nil {
 				return fmt.Errorf("failed to move veth to container netns: %s", err)
 			}
@@ -143,6 +143,10 @@ func AttachContainer(netNSPath, id, ifName, bridgeName string, mtu int, withMult
 		if err != nil {
 			return err
 		}
+		if err = netlink.LinkSetHairpin(veth, hairpinMode); err != nil {
+			return fmt.Errorf("unable to set hairpin mode to %t for bridge side of veth %s: %s", hairpinMode, name, err)
+		}
+
 	}
 
 	if err := WithNetNSLinkUnsafe(ns, ifName, func(veth netlink.Link) error {

npc/constants.go

@@ -8,4 +8,6 @@ const (
 	IngressChain = "WEAVE-NPC-INGRESS"
 
 	IpsetNamePrefix = "weave-"
+
+	LocalIpset = IpsetNamePrefix + "local-pods"
 )

npc/controller.go

@@ -29,6 +29,8 @@ type NetworkPolicyController interface {
 type controller struct {
 	sync.Mutex
 
+	nodeName string // my node name
+
 	ipt *iptables.IPTables
 	ips ipset.Interface
 
@@ -36,11 +38,12 @@ type controller struct {
 	nsSelectors *selectorSet   // selector string -> nsSelector
 }
 
-func New(ipt *iptables.IPTables, ips ipset.Interface) NetworkPolicyController {
+func New(nodeName string, ipt *iptables.IPTables, ips ipset.Interface) NetworkPolicyController {
 	c := &controller{
-		ipt: ipt,
-		ips: ips,
-		nss: make(map[string]*ns)}
+		nodeName: nodeName,
+		ipt:      ipt,
+		ips:      ips,
+		nss:      make(map[string]*ns)}
 
 	c.nsSelectors = newSelectorSet(ips, c.onNewNsSelector)
 
@@ -82,10 +85,21 @@ func (npc *controller) withNS(name string, f func(ns *ns) error) error {
 	return nil
 }
 
+func (npc *controller) checkLocalPod(obj *coreapi.Pod, event string) bool {
+	if obj.Spec.NodeName != npc.nodeName {
+		common.Log.Debugf("%s ignored for pod %s/%s on node %s", event, obj.ObjectMeta.Namespace, obj.ObjectMeta.Name, obj.Spec.NodeName)
+		return false
+	}
+	return true
+}
+
 func (npc *controller) AddPod(obj *coreapi.Pod) error {
 	npc.Lock()
 	defer npc.Unlock()
 
+	if !npc.checkLocalPod(obj, "AddPod") {
+		return nil
+	}
 	common.Log.Debugf("EVENT AddPod %s", js(obj))
 	return npc.withNS(obj.ObjectMeta.Namespace, func(ns *ns) error {
 		return errors.Wrap(ns.addPod(obj), "add pod")
@@ -96,6 +110,10 @@ func (npc *controller) UpdatePod(oldObj, newObj *coreapi.Pod) error {
 	npc.Lock()
 	defer npc.Unlock()
 
+	// Pods don't move across nodes so we only check the new version
+	if !npc.checkLocalPod(newObj, "UpdatePod") {
+		return nil
+	}
 	common.Log.Debugf("EVENT UpdatePod %s %s", js(oldObj), js(newObj))
 	return npc.withNS(oldObj.ObjectMeta.Namespace, func(ns *ns) error {
 		return errors.Wrap(ns.updatePod(oldObj, newObj), "update pod")
@@ -106,6 +124,9 @@ func (npc *controller) DeletePod(obj *coreapi.Pod) error {
 	npc.Lock()
 	defer npc.Unlock()
 
+	if !npc.checkLocalPod(obj, "DeletePod") {
+		return nil
+	}
 	common.Log.Debugf("EVENT DeletePod %s", js(obj))
 	return npc.withNS(obj.ObjectMeta.Namespace, func(ns *ns) error {
 		return errors.Wrap(ns.deletePod(obj), "delete pod")

npc/ipset/ipset.go

@@ -1,6 +1,7 @@
 package ipset
 
 import (
+	"log"
 	"os/exec"
 	"strings"
 
@@ -31,10 +32,11 @@ type Interface interface {
 
 type ipset struct {
 	refCount
+	*log.Logger
 }
 
-func New() Interface {
-	return &ipset{refCount: newRefCount()}
+func New(logger *log.Logger) Interface {
+	return &ipset{refCount: newRefCount(), Logger: logger}
 }
 
 func (i *ipset) Create(ipsetName Name, ipsetType Type) error {
@@ -42,6 +44,7 @@ func (i *ipset) Create(ipsetName Name, ipsetType Type) error {
 }
 
 func (i *ipset) AddEntry(ipsetName Name, entry string) error {
+	i.Logger.Printf("adding entry %s to %s", entry, ipsetName)
 	if i.inc(ipsetName, entry) > 1 { // already in the set
 		return nil
 	}
@@ -49,6 +52,7 @@ func (i *ipset) AddEntry(ipsetName Name, entry string) error {
 }
 
 func (i *ipset) DelEntry(ipsetName Name, entry string) error {
+	i.Logger.Printf("deleting entry %s from %s", entry, ipsetName)
 	if i.dec(ipsetName, entry) > 0 { // still needed
 		return nil
 	}

npc/namespace.go

@@ -88,6 +88,7 @@ func (ns *ns) addPod(obj *coreapi.Pod) error {
 		return nil
 	}
 
+	ns.ips.AddEntry(LocalIpset, obj.Status.PodIP)
 	return ns.podSelectors.addToMatching(obj.ObjectMeta.Labels, obj.Status.PodIP)
 }
 
@@ -100,10 +101,12 @@ func (ns *ns) updatePod(oldObj, newObj *coreapi.Pod) error {
 	}
 
 	if hasIP(oldObj) && !hasIP(newObj) {
+		ns.ips.DelEntry(LocalIpset, oldObj.Status.PodIP)
 		return ns.podSelectors.delFromMatching(oldObj.ObjectMeta.Labels, oldObj.Status.PodIP)
 	}
 
 	if !hasIP(oldObj) && hasIP(newObj) {
+		ns.ips.AddEntry(LocalIpset, newObj.Status.PodIP)
 		return ns.podSelectors.addToMatching(newObj.ObjectMeta.Labels, newObj.Status.PodIP)
 	}
 
@@ -139,6 +142,7 @@ func (ns *ns) deletePod(obj *coreapi.Pod) error {
 		return nil
 	}
 
+	ns.ips.DelEntry(LocalIpset, obj.Status.PodIP)
 	return ns.podSelectors.delFromMatching(obj.ObjectMeta.Labels, obj.Status.PodIP)
 }
 

npc/selector.go

@@ -43,12 +43,10 @@ func (s *selector) matches(labelMap map[string]string) bool {
 }
 
 func (s *selector) addEntry(entry string) error {
-	common.Log.Infof("adding entry %s to %s", entry, s.spec.ipsetName)
 	return s.ips.AddEntry(s.spec.ipsetName, entry)
 }
 
 func (s *selector) delEntry(entry string) error {
-	common.Log.Infof("deleting entry %s from %s", entry, s.spec.ipsetName)
 	return s.ips.DelEntry(s.spec.ipsetName, entry)
 }
 

npc/shortname.go

@@ -7,12 +7,13 @@ import (
 
 const (
 	// This array:
-	// * Must include only printable UTF8 characters that are represented with a single byte
+	// * Must include only printable UTF8 characters that are represented with a single byte (http://www.utf8-chartable.de/)
 	// * Must be at least of length 85 (`len("weave-") + l(2^160)/l(85)` equals 31, the maximum ipset name length)
 	// * Must not include commas as those are treated specially by `ipset add` when adding a named set to a list:set
 	// * Should not include space for readability
+	// * Should not include invalid chars for XML `<>&"'`
 	// * Should not include single quote or backslash to be nice to shell users
-	ShortNameSymbols = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789(){}[]<>_$%^&*|/?.;:@#~"
+	ShortNameSymbols = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789(){}[]+=_$%^!*|/?.;:@#~"
 )
 
 // sha1 hash an arbitrary string and represent it using the full range of

plugin/net/cni.go

@@ -6,7 +6,6 @@ import (
 	"fmt"
 	"net"
 	"os"
-	"strings"
 	"syscall"
 
 	"github.com/containernetworking/cni/pkg/ipam"
@@ -35,7 +34,8 @@ func NewCNIPlugin(weave *weaveapi.Client) *CNIPlugin {
 
 func loadNetConf(bytes []byte) (*NetConf, error) {
 	n := &NetConf{
-		BrName: weavenet.WeaveBridgeName,
+		BrName:      weavenet.WeaveBridgeName,
+		HairpinMode: true,
 	}
 	if err := json.Unmarshal(bytes, n); err != nil {
 		return nil, fmt.Errorf("failed to load netconf: %v", err)
@@ -129,7 +129,7 @@ func (c *CNIPlugin) CmdAdd(args *skel.CmdArgs) error {
 		id = fmt.Sprintf("%x", data)
 	}
 
-	if err := weavenet.AttachContainer(args.Netns, id, args.IfName, conf.BrName, conf.MTU, false, []*net.IPNet{&ip.Address}, false); err != nil {
+	if err := weavenet.AttachContainer(args.Netns, id, args.IfName, conf.BrName, conf.MTU, false, []*net.IPNet{&ip.Address}, false, conf.HairpinMode); err != nil {
 		return err
 	}
 	if err := weavenet.WithNetNSLinkUnsafe(ns, args.IfName, func(link netlink.Link) error {
@@ -183,16 +183,21 @@ func assignBridgeIP(bridgeName string, ipnet net.IPNet) error {
 	return nil
 }
 
+// As of CNI 0.5 spec:
+//   "Plugins should generally complete a DEL action without error even if some resources are missing"
+// this method should therefore return nil in most, if not all, cases.
 func (c *CNIPlugin) CmdDel(args *skel.CmdArgs) error {
 	conf, err := loadNetConf(args.StdinData)
 	if err != nil {
-		return err
+		logOnStderr(err)
+		return nil
 	}
 
 	// As of CNI 0.3 spec, runtimes can send blank if they just want the address deallocated
 	if args.Netns != "" {
 		if _, err = weavenet.WithNetNS(args.Netns, "del-iface", args.IfName); err != nil {
-			return fmt.Errorf("error removing interface %q: %s", args.IfName, err)
+			// We log the error and carry on instead of returning nil, as there may still be resources to free in IPAM.
+			logOnStderr(fmt.Errorf("error removing interface %q: %s", args.IfName, err))
 		}
 	}
 
@@ -202,20 +207,21 @@ func (c *CNIPlugin) CmdDel(args *skel.CmdArgs) error {
 	} else {
 		err = ipam.ExecDel(conf.IPAM.Type, args.StdinData)
 	}
-	// Hack - don't know how we should detect this situation properly
-	if args.Netns == "" && strings.Contains(err.Error(), "no addresses") {
-		err = nil
-	}
-	if err != nil {
-		return fmt.Errorf("unable to release IP address: %s", err)
-	}
+	logOnStderr(fmt.Errorf("unable to release IP address: %s", err))
 	return nil
 }
 
+// Standard error is inherited from the caller (e.g. Kubelet, when running Weave Net's CNI plugin
+// under Kubernetes) hence provided errors should eventually appear in the caller's log files.
+func logOnStderr(err error) {
+	fmt.Fprintln(os.Stderr, "weave-cni:", err)
+}
+
 type NetConf struct {
 	types.NetConf
-	BrName string `json:"bridge"`
-	IsGW   bool   `json:"isGateway"`
-	IPMasq bool   `json:"ipMasq"`
-	MTU    int    `json:"mtu"`
+	BrName      string `json:"bridge"`
+	IsGW        bool   `json:"isGateway"`
+	IPMasq      bool   `json:"ipMasq"`
+	MTU         int    `json:"mtu"`
+	HairpinMode bool   `json:"hairpinMode"`
 }

prog/weave-kube/launch.sh

@@ -8,6 +8,11 @@ modprobe_safe() {
 
 # Check whether xt_set actually exists
 xt_set_exists() {
+    # Clean everything up in advance, in case there's leftovers
+    iptables -F WEAVE-KUBE-TEST 2>/dev/null || true
+    iptables -X WEAVE-KUBE-TEST 2>/dev/null || true
+    ipset destroy weave-kube-test 2>/dev/null || true
+
     ipset create weave-kube-test hash:ip
     iptables -t filter -N WEAVE-KUBE-TEST
     if ! iptables -A WEAVE-KUBE-TEST -m set --match-set weave-kube-test src -j DROP; then

prog/weave-kube/weave-daemonset-k8s-1.6.yaml

@@ -92,6 +92,12 @@ spec:
               cpu: 10m
           securityContext:
             privileged: true
+          env:
+          - name: HOSTNAME
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: spec.nodeName
       restartPolicy: Always
       tolerations:
       - key: node-role.kubernetes.io/master

prog/weave-kube/weave-daemonset.yaml

@@ -59,6 +59,12 @@ spec:
               cpu: 10m
           securityContext:
             privileged: true
+          env:
+          - name: HOSTNAME
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: spec.nodeName
       restartPolicy: Always
       volumes:
         - name: weavedb

prog/weave-npc/main.go

@@ -28,6 +28,7 @@ var (
 	metricsAddr string
 	logLevel    string
 	allowMcast  bool
+	nodeName    string
 )
 
 func handleError(err error) { common.CheckFatal(err) }
@@ -53,6 +54,41 @@ func resetIPTables(ipt *iptables.IPTables) error {
 		return err
 	}
 
+	return nil
+}
+
+func resetIPSets(ips ipset.Interface) error {
+	// Remove ipsets prefixed `weave-` only
+
+	sets, err := ips.List(npc.IpsetNamePrefix)
+	if err != nil {
+		common.Log.Errorf("Failed to retrieve list of ipsets")
+		return err
+	}
+
+	common.Log.Debugf("Got list of ipsets: %v", sets)
+
+	// Must remove references to ipsets by other ipsets before they're destroyed
+	for _, s := range sets {
+		common.Log.Debugf("Flushing ipset '%s'", string(s))
+		if err := ips.Flush(s); err != nil {
+			common.Log.Errorf("Failed to flush ipset '%s'", string(s))
+			return err
+		}
+	}
+
+	for _, s := range sets {
+		common.Log.Debugf("Destroying ipset '%s'", string(s))
+		if err := ips.Destroy(s); err != nil {
+			common.Log.Errorf("Failed to destroy ipset '%s'", string(s))
+			return err
+		}
+	}
+
+	return nil
+}
+
+func createBaseRules(ipt *iptables.IPTables, ips ipset.Interface) error {
 	// Configure main chain static rules
 	if err := ipt.Append(npc.TableFilter, npc.MainChain,
 		"-m", "state", "--state", "RELATED,ESTABLISHED", "-j", "ACCEPT"); err != nil {
@@ -76,26 +112,13 @@ func resetIPTables(ipt *iptables.IPTables) error {
 		return err
 	}
 
-	return nil
-}
-
-func resetIPSets(ips ipset.Interface) error {
-	// Remove ipsets prefixed `weave-` only
-
-	sets, err := ips.List(npc.IpsetNamePrefix)
-	if err != nil {
-		common.Log.Errorf("Failed to retrieve list of ipsets")
+	// If the destination address is not any of the local pods, let it through
+	if err := ips.Create(npc.LocalIpset, ipset.HashIP); err != nil {
 		return err
 	}
-
-	common.Log.Debugf("Got list of ipsets: %v", sets)
-
-	for _, s := range sets {
-		common.Log.Debugf("Destroying ipsets '%s'", string(s))
-		if err := ips.Destroy(s); err != nil {
-			common.Log.Errorf("Failed to destroy ipset '%s'", string(s))
-			return err
-		}
+	if err := ipt.Append(npc.TableFilter, npc.MainChain,
+		"-m", "set", "!", "--match-set", npc.LocalIpset, "dst", "-j", "ACCEPT"); err != nil {
+		return err
 	}
 
 	return nil
@@ -103,7 +126,14 @@ func resetIPSets(ips ipset.Interface) error {
 
 func root(cmd *cobra.Command, args []string) {
 	common.SetLogLevel(logLevel)
-	common.Log.Infof("Starting Weaveworks NPC %s", version)
+	if nodeName == "" {
+		// HOSTNAME is set by Kubernetes for pods in the host network namespace
+		nodeName = os.Getenv("HOSTNAME")
+	}
+	if nodeName == "" {
+		common.Log.Fatalf("Must set node name via --node-name or $HOSTNAME")
+	}
+	common.Log.Infof("Starting Weaveworks NPC %s; node name %q", version, nodeName)
 
 	if err := metrics.Start(metricsAddr); err != nil {
 		common.Log.Fatalf("Failed to start metrics: %v", err)
@@ -122,12 +152,13 @@ func root(cmd *cobra.Command, args []string) {
 	ipt, err := iptables.New()
 	handleError(err)
 
-	ips := ipset.New()
+	ips := ipset.New(common.LogLogger())
 
 	handleError(resetIPTables(ipt))
 	handleError(resetIPSets(ips))
+	handleError(createBaseRules(ipt, ips))
 
-	npc := npc.New(ipt, ips)
+	npc := npc.New(nodeName, ipt, ips)
 
 	nsController := makeController(client.Core().RESTClient(), "namespaces", &coreapi.Namespace{},
 		cache.ResourceEventHandlerFuncs{
@@ -207,6 +238,7 @@ func main() {
 	rootCmd.PersistentFlags().StringVar(&metricsAddr, "metrics-addr", ":6781", "metrics server bind address")
 	rootCmd.PersistentFlags().StringVar(&logLevel, "log-level", "debug", "logging level (debug, info, warning, error)")
 	rootCmd.PersistentFlags().BoolVar(&allowMcast, "allow-mcast", true, "allow all multicast traffic")
+	rootCmd.PersistentFlags().StringVar(&nodeName, "node-name", "", "only generate rules that apply to this node")
 
 	handleError(rootCmd.Execute())
 }

prog/weaveutil/attach.go

@@ -14,11 +14,12 @@ import (
 
 func attach(args []string) error {
 	if len(args) < 4 {
-		cmdUsage("attach-container", "[--no-multicast-route] [--keep-tx-on] <container-id> <bridge-name> <mtu> <cidr>...")
+		cmdUsage("attach-container", "[--no-multicast-route] [--keep-tx-on] [--hairpin-mode=true|false] <container-id> <bridge-name> <mtu> <cidr>...")
 	}
 
 	keepTXOn := false
 	withMulticastRoute := true
+	hairpinMode := true
 	for i := 0; i < len(args); {
 		switch args[i] {
 		case "--no-multicast-route":
@@ -27,6 +28,9 @@ func attach(args []string) error {
 		case "--keep-tx-on":
 			keepTXOn = true
 			args = append(args[:i], args[i+1:]...)
+		case "--hairpin-mode=false":
+			hairpinMode = false
+			args = append(args[:i], args[i+1:]...)
 		default:
 			i++
 		}
@@ -55,7 +59,7 @@ func attach(args []string) error {
 		return err
 	}
 
-	err = weavenet.AttachContainer(weavenet.NSPathByPid(pid), fmt.Sprint(pid), weavenet.VethName, args[1], mtu, withMulticastRoute, cidrs, keepTXOn)
+	err = weavenet.AttachContainer(weavenet.NSPathByPid(pid), fmt.Sprint(pid), weavenet.VethName, args[1], mtu, withMulticastRoute, cidrs, keepTXOn, hairpinMode)
 	// If we detected an error but the container has died, tell the user that instead.
 	if err != nil && !processExists(pid) {
 		err = fmt.Errorf("Container %s died", args[0])

site/cni-plugin.md

@@ -12,7 +12,7 @@ search_type: Documentation
 CNI, the [_Container Network Interface_](https://github.com/containernetworking/cni),
 is a proposed standard for configuring network interfaces for Linux
 application containers.  CNI is supported by
-[Kubernetes](http://kubernetes.io/), [Apache Mesos](mesos.apache.org)
+[Kubernetes](http://kubernetes.io/), [Apache Mesos](http://mesos.apache.org)
 and 
 [others](https://github.com/containernetworking/cni#who-is-using-cni).
 

site/installing-weave.md

@@ -51,7 +51,7 @@ To disable this check, run the following before launching Weave Net:
 CoreOS users see [here](/guides/networking-docker-containers-with-weave-on-coreos/) for an example of installing Weave using cloud-config.
 
 Amazon ECS users see [here](https://github.com/weaveworks/integrations/blob/master/aws/ecs/README.md)
-for the latest Weave AMIs and [here](http://weave.works/guides/service-discovery-with-weave-aws-ecs.html) to get started with Weave Net on ECS.
+for the latest Weave AMIs and [here](https://www.weave.works/docs/tutorials/old-guides/ecs/) to get started with Weave Net on ECS.
 
 If you're on Amazon EC2, the standard installation instructions at the
 top of this page, provide the simplest setup and the most flexibility.

site/kube-addon.md

@@ -9,6 +9,8 @@ The following topics are discussed:
 * [Installation](#install)
  * [Upgrading Kubernetes to version 1.6](#kube-1.6-upgrade)
  * [Upgrading the Daemon Sets](#daemon-sets)
+ * [CPU and Memory Requirements](#resources)
+ * [Pod Eviction](#eviction)
 * [Network Policy Controller](#npc)
  * [Troubleshooting Blocked Connections](#blocked-connections)
  * [Changing Configuration Options](#configuration-options)
@@ -35,7 +37,8 @@ After a few seconds, a Weave Net pod should be running on each
 Node and any further pods you create will be automatically attached to the Weave
 network.
 
-**Note:** This command requires Kubernetes 1.4 or later.
+**Note:** This command requires Kubernetes 1.4 or later, and we
+recommend your master node has at least two CPU cores.
 
 > CNI, the [_Container Network Interface_](https://github.com/containernetworking/cni),
 > is a proposed standard for configuring network interfaces for Linux
@@ -92,6 +95,64 @@ and so you will need to perform the procedure manually:
   lose track of IP address range ownership, possibly leading to
   duplicate IP addresses if you then start a new copy of Weave Net.
 
+## <a name="resources"></a>CPU and Memory Requirements
+
+Kubernetes manages
+[resources](https://kubernetes.io/docs/concepts/configuration/manage-compute-resources-container/)
+on each node, and only schedules pods to run on nodes that have enough
+free resources.
+
+The components of a typical Kubernetes installation (with the master
+node running etcd, scheduler, api-server, etc.) take up about 95% of a
+CPU, which leaves little room to run anything else. For all of Weave
+Net's features to work, it must run on every node, including the
+master.
+
+The best way to resolve this issue is to use machines with at least
+two CPU cores. However if you are installing Kubernetes and Weave Net
+for the first time, you may not be aware of this requirement. For this
+reason, Weave Net launches as a DaemonSet with a specification that
+reserves at least 1% CPU for each container. This enables Weave Net to
+start up seamlessly on a single-CPU node.
+
+Depending on the workload, Weave Net may need more than 1% of the
+CPU. The percentage set in the DaemonSet is the minimum and not a
+limit. This minimum setting allows Weave Net to take advantage of
+available CPU and "burst" above that limit if it needs to.
+
+## <a name="eviction"></a>Pod Eviction
+
+If a node runs out of CPU, memory or disk, Kubernetes may [decide to
+evict](https://kubernetes.io/docs/concepts/cluster-administration/out-of-resource/)
+one or more pods. It may choose to evict the Weave Net pod, which will
+disrupt pod network operations.
+
+You can reduce the chance of eviction by changing the DaemonSet to
+have a much bigger request, and a limit of the same value.
+
+This causes Kubernetes to apply a ["guaranteed" rather than a
+"burstable" policy](https://github.com/kubernetes/community/blob/master/contributors/design-proposals/resource-qos.md).
+However a similar request for disk space can not
+be made, and so please be aware of this issue and monitor your
+resources to ensure that they stay below 100%.
+
+You can see when pods have been evicted via the `kubectl get events` command
+
+```
+LASTSEEN   COUNT     NAME          KIND    TYPE      REASON     SOURCE            MESSAGE
+1m         1         mypod-09vkd   Pod     Warning   Evicted    kubelet, node-1   The node was low on resource: memory.
+```
+
+or `kubectl get pods`
+
+```
+NAME                READY     STATUS    RESTARTS   AGE       IP          NODE
+mypod-09vkd         0/1       Evicted   0          1h        <none>      node-1
+```
+
+If you see this in your cluster, consider some of the above steps to
+reduce disruption.
+
 ## <a name="npc"></a>Network Policy Controller
 
 The addon also supports the [Kubernetes policy
@@ -180,3 +241,7 @@ The list of variables you can set is:
   a larger size for better performance if your network supports jumbo
   frames - see [here](/site/using-weave/fastdp.md#mtu) for more
   details.
+* `HAIRPIN_MODE` - Weave Net defaults to enabling hairpin on the bridge side of
+  the `veth` pair for containers attached. If you need to disable hairpin, e.g. your
+  kernel is one of those that can panic if hairpin is enabled, then you can disable it
+  by setting `HAIRPIN_MODE=false`.

site/operational-guide/tasks.md

@@ -44,7 +44,7 @@ or possible:
 In some cases you may already be aware of the problem, as you were
 unable to execute `weave reset` successfully or because you know
 through other channels that the host has died. In these cases you can
-proceed straight to [Manually Reclaiming Lost Address Space](#reclaim-address-space).
+proceed straight to [Manually Reclaiming Lost Address Space](#manually-reclaim-address-space).
 
 However in some scenarios it may not be obvious that space has been
 lost, in which case you can check for it periodically with the

site/plugin.md

@@ -6,7 +6,6 @@ search_type: Documentation
 
  * [Launching Weave Net and Running Containers Using the Plugin](#launching)
  * [Restarting the Plugin](#restarting)
- * [Configuring the Plugin](#configuring)
 
 Docker versions 1.9 and later have a plugin mechanism for adding
 different network providers. Weave Net installs itself as a network

site/plugin/plugin-how-it-works.md

@@ -23,7 +23,7 @@ If you do create additional networks using the `weavemesh` driver, containers at
 
 There's no specific documentation from Docker on using a cluster
 store, but the first part of
-[Getting Started with Docker Multi-host Networking](https://github.com/docker/docker/blob/master/docs/userguide/networking/get-started-overlay.md) is a good place to start.
+[Getting Started with Docker Multi-host Networking](https://docs.docker.com/engine/userguide/networking/get-started-overlay/) is a good place to start.
 
 >**Note:** In the case of multiple networks using the `weave` driver, all containers are on the same virtual network but Docker allocates their addresses on different subnets so they cannot talk to each other directly.
 

site/weavedns/load-balance-fault-weavedns.md

@@ -49,7 +49,7 @@ WeaveDNS removes the addresses of any container that dies. This offers
 a simple way to implement redundancy. E.g. if in our example we stop
 one of the `pingme` containers and re-run the ping tests, eventually
 (within ~30s at most, since that is the weaveDNS
-[cache expiry time](#ttl)) we will only be hitting the address of the
+[cache expiry time](/site/weavedns/managing-entries-weavedns.md#ttl)) we will only be hitting the address of the
 container that is still alive.
 
 **See Also**

site/weavedns/managing-entries-weavedns.md

@@ -4,17 +4,13 @@ menu_order: 40
 search_type: Documentation
 ---
 
-
 The following topics are discussed: 
 
 * [Adding and removing extra DNS entries](#add-remove)
 * [Resolving WeaveDNS entries from the Host](#resolve-weavedns-entries-from-host)
 * [Hot-swapping Service Containers](#hot-swapping)
-* [Retaining DNS Entries When Containers Stop](#retain-stopped)
 * [Configuring a Custom TTL](#ttl)
 
-
-
 ### <a name="add-remove"></a>Adding and Removing Extra DNS Entries
 
 If you want to give the container a name in DNS *other* than its
@@ -82,6 +78,6 @@ weaveDNS instance will receive.
 
 **See Also**
 
- * [How Weave Finds Containers](/site/how-works-weavedns.md)
+ * [How Weave Finds Containers](/site/weavedns/how-works-weavedns.md)
  * [Load Balancing and Fault Resilience with WeaveDNS](/site/weavedns/load-balance-fault-weavedns.md)
  * [Managing Domains](/site/weavedns/managing-domains-weavedns.md)

test/830_cni_plugin_test.sh

@@ -8,7 +8,7 @@ cni_connect() {
     pid=$(container_pid $1 $2)
     id=$(docker_on $1 inspect -f '{{.Id}}' $2)
     run_on $1 sudo CNI_COMMAND=ADD CNI_CONTAINERID=$id CNI_IFNAME=eth0 \
-    CNI_NETNS=/proc/$pid/ns/net CNI_PATH=/opt/cni/bin /opt/cni/bin/weave-net 
+    CNI_NETNS=/proc/$pid/ns/net CNI_PATH=/opt/cni/bin /opt/cni/bin/weave-net
 }
 
 run_on $HOST1 sudo mkdir -p /opt/cni/bin
@@ -19,6 +19,11 @@ weave_on $HOST1 launch
 C0=$(docker_on $HOST1 run --net=none --name=c0 -dt $SMALL_IMAGE /bin/sh)
 C1=$(docker_on $HOST1 run --net=none --privileged --name=c1 -dt $SMALL_IMAGE /bin/sh)
 C2=$(docker_on $HOST1 run --net=none --name=c2 -dt $SMALL_IMAGE /bin/sh)
+CH0=$(docker_on $HOST1 run --net=none --name=ch0 -dt $SMALL_IMAGE /bin/sh)
+CH1=$(docker_on $HOST1 run --net=none --name=ch1 -dt $SMALL_IMAGE /bin/sh)
+CH2=$(docker_on $HOST1 run --net=none --name=ch2 -dt $SMALL_IMAGE /bin/sh)
+
+
 # Enable unsolicited ARPs so that ping after the address reuse does not time out
 exec_on $HOST1 c1 sysctl -w net.ipv4.conf.all.arp_accept=1
 
@@ -52,6 +57,29 @@ cni_connect $HOST1 c2 <<EOF
 }
 EOF
 
+# examples to test hairpin
+cni_connect $HOST1 ch0 <<EOF
+{
+    "name": "weave",
+    "type": "weave-net",
+    "hairpinMode": true
+}
+EOF
+cni_connect $HOST1 ch1 <<EOF
+{
+    "name": "weave",
+    "type": "weave-net",
+    "hairpinMode": false
+}
+EOF
+cni_connect $HOST1 ch2 <<EOF
+{
+    "name": "weave",
+    "type": "weave-net"
+}
+EOF
+
+
 C0IP=$(container_ip $HOST1 c0)
 C1IP=$(container_ip $HOST1 c1)
 C2IP=$(container_ip $HOST1 c2)
@@ -85,6 +113,29 @@ assert_raises "[ $C2IP != $C3IP ]"
 assert_raises "[ $BRIP != $C3IP ]"
 assert_raises "exec_on $HOST1 c1 $PING $C3IP"
 
+#####
+#
+# Hairpin mode tests
+#
+# here we check that the bridge side of the veth is set to hairpin mode (true or false) correctly
+# check for each of
+# - ch0 explicitly true
+# - ch1 explicitly false
+# - ch2 unset, which should default to true
+#
+# we take advantage of the veth name being constructed from the container ID)
+#
+# should this reliance ever fail and we need to get it from the system, do the following for each:
+# CH0_PEER_ID=$(exec_on $HOST1 ch0 cat /sys/class/net/eth0/iflink)
+# CH0_PEER=$($SSH $HOST1 ip link show | awk "/^${CH0_PEER_ID}:/"' {print $2}' | sed 's/@.*$//g' )
+# CH0_HAIRPIN=$($SSH $HOST1 cat /sys/devices/virtual/net/weave/brif/$CH0_PEER/hairpin_mode)
+# assert_raises "[ $CH0_HAIRPIN == 1 ]"
+#
+assert "$SSH $HOST1 cat /sys/devices/virtual/net/weave/brif/vethwepl${CH0:0:7}/hairpin_mode" "1"
+assert "$SSH $HOST1 cat /sys/devices/virtual/net/weave/brif/vethwepl${CH1:0:7}/hairpin_mode" "0"
+assert "$SSH $HOST1 cat /sys/devices/virtual/net/weave/brif/vethwepl${CH2:0:7}/hairpin_mode" "1"
+
+
 
 # Ensure existing containers can reclaim their IP addresses after CNI has been used -- see #2548
 stop_weave_on $HOST1

test/run-integration-tests.sh

@@ -284,7 +284,7 @@ function configure() {
         local inventory_file=$(mktemp /tmp/ansible_inventory_XXXXX)
         echo "[all]" >"$inventory_file"
         # shellcheck disable=SC2001
-        echo "$2" | sed "s/$/:$3/" >>"$inventory_file"
+        echo "$2" | sed "s/\$/:$3/" >>"$inventory_file"
 
         # Configure the provisioned machines using Ansible, allowing up to 3 retries upon failure (e.g. APT connectivity issues, etc.):
         for i in $(seq 3); do

test/run_all.sh

@@ -23,4 +23,4 @@ if [ -n "$CIRCLECI" -o -n "$PARALLEL" ]; then
     RUNNER_ARGS="$RUNNER_ARGS -parallel"
 fi
 
-HOSTS="$HOSTS" "${DIR}/../tools/runner/runner" $RUNNER_ARGS $TESTS
+HOSTS="$HOSTS" "${DIR}/../tools/runner/runner" -timeout=240 $RUNNER_ARGS $TESTS

weave

@@ -1070,17 +1070,20 @@ create_cni_config() {
     cat >"$1" <<EOF
 {
     "name": "weave",
-    "type": "weave-net"
+    "type": "weave-net",
+    "hairpinMode": $2
 }
 EOF
 }
 
 setup_cni() {
+    # if env var HAIRPIN_MODE is not set, default it to true
+    HAIRPIN_MODE=${HAIRPIN_MODE:-true}
     if install_cni_plugin $CNI_PLUGIN_DIR ; then
         upgrade_cni_plugin $CNI_PLUGIN_DIR
     fi
     if [ -d $HOST_ROOT/etc/cni/net.d -a ! -f $HOST_ROOT/etc/cni/net.d/10-weave.conf ] ; then
-        create_cni_config $HOST_ROOT/etc/cni/net.d/10-weave.conf
+        create_cni_config $HOST_ROOT/etc/cni/net.d/10-weave.conf $HAIRPIN_MODE
     fi
 }