diff --git a/net/bridge.go b/net/bridge.go
index f882ebb1..268e70c1 100644
--- a/net/bridge.go
+++ b/net/bridge.go
@@ -262,7 +262,7 @@ func EnsureBridge(procPath string, config *BridgeConfig, log *logrus.Logger, ips
 		break
 	}
 
-	if err := configureIPTables(config, ips); err != nil {
+	if err := ConfigureIPTables(config, ips); err != nil {
 		return bridgeType, errors.Wrap(err, "configuring iptables")
 	}
 
@@ -297,9 +297,7 @@ func EnsureBridge(procPath string, config *BridgeConfig, log *logrus.Logger, ips
 		return bridgeType, errors.Wrapf(err, "configuring ARP cache on bridge %q", config.WeaveBridgeName)
 	}
 
-	// NB: No concurrent call to Expose is possible, as EnsureBridge is called
-	// before any service has been started.
-	if err := reexpose(config, log); err != nil {
+	if err := Reexpose(config, log); err != nil {
 		return bridgeType, err
 	}
 
@@ -431,10 +429,40 @@ func (f fastdpImpl) attach(veth *netlink.Veth) error {
 	return odp.AddDatapathInterfaceIfNotExist(f.datapathName, veth.Attrs().Name)
 }
 
-func configureIPTables(config *BridgeConfig, ips ipset.Interface) error {
+// ResetIPTables resets IPTables in case they're in a strange state from a previous run.
+func ResetIPTables(config *BridgeConfig, ips ipset.Interface) error {
 	ipt, err := iptables.New()
 	if err != nil {
-		return errors.Wrap(err, "creating iptables object")
+		return errors.Wrap(err, "creating iptables object while resetting")
+	}
+
+	if !config.NPC {
+		// Create/Flush a chain for allowing ingress traffic when the bridge is exposed
+		if err := ipt.ClearChain("filter", "WEAVE-EXPOSE"); err != nil {
+			return errors.Wrap(err, "failed to clear/create filter/WEAVE-EXPOSE chain")
+		}
+	}
+
+	if err := ipt.ClearChain("nat", "WEAVE"); err != nil {
+		return errors.Wrap(err, "failed to clear/create nat/WEAVE chain")
+	}
+
+	if config.NoMasqLocal {
+		ips := ipset.New(common.LogLogger(), 0)
+		_ = ips.Destroy(NoMasqLocalIpset)
+		if err := ips.Create(NoMasqLocalIpset, ipset.HashNet); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// ConfigureIPTables idempotently configures all the iptables!
+func ConfigureIPTables(config *BridgeConfig, ips ipset.Interface) error {
+	ipt, err := iptables.New()
+	if err != nil {
+		return errors.Wrap(err, "creating iptables object while configuring")
 	}
 
 	// The order among weave filter/FORWARD rules is important!
@@ -515,8 +543,8 @@ func configureIPTables(config *BridgeConfig, ips ipset.Interface) error {
 
 	if !config.NPC {
 		// Create/Flush a chain for allowing ingress traffic when the bridge is exposed
-		if err := ipt.ClearChain("filter", "WEAVE-EXPOSE"); err != nil {
-			return errors.Wrap(err, "failed to clear/create filter/WEAVE-EXPOSE chain")
+		if err := ensureChains(ipt, "filter", "WEAVE-EXPOSE"); err != nil {
+			return errors.Wrap(err, "failed to ensure existence of filter/WEAVE-EXPOSE chain")
 		}
 
 		fwdRules = append(fwdRules, []string{"-o", config.WeaveBridgeName, "-j", "WEAVE-EXPOSE"})
@@ -532,26 +560,22 @@ func configureIPTables(config *BridgeConfig, ips ipset.Interface) error {
 	}
 
 	// Create a chain for masquerading
-	if err := ipt.ClearChain("nat", "WEAVE"); err != nil {
-		return errors.Wrap(err, "failed to clear/create nat/WEAVE chain")
+	if err := ensureChains(ipt, "nat", "WEAVE"); err != nil {
+		return errors.Wrap(err, "failed to ensure existence of nat/WEAVE chain")
 	}
 	if err := ipt.AppendUnique("nat", "POSTROUTING", "-j", "WEAVE"); err != nil {
 		return err
 	}
 
 	// For the cases where the weave bridge is the default gateway for
-	// containers (e.g. Kubernetes): create the ipset to store CIDRs allocated
-	// by IPAM for local containers. In the case of Kubernetes, external traffic
-	// sent to these CIDRs avoids SNAT'ing so that NodePort with
-	// `"externalTrafficPolicy":"Local"` would receive packets with correct
-	// src IP addr.
+	// containers (e.g. Kubernetes): In `ResetIPTables` (which we assume
+	// to have been called at this point) we create an ipset to store CIDRs
+	// allocated by IPAM for local containers.
+	// In the case of Kubernetes, external traffic sent to these CIDRs
+	// avoids SNAT'ing so that NodePort with `"externalTrafficPolicy":"Local"`
+	// would receive packets with correct src IP addr.
 	if config.NoMasqLocal {
-		ips := ipset.New(common.LogLogger(), 0)
-		_ = ips.Destroy(NoMasqLocalIpset)
-		if err := ips.Create(NoMasqLocalIpset, ipset.HashNet); err != nil {
-			return err
-		}
-		if err := ipt.Insert("nat", "WEAVE", 1,
+		if err := ipt.AppendUnique("nat", "WEAVE",
 			"-m", "set", "--match-set", string(NoMasqLocalIpset), "dst",
 			"-m", "comment", "--comment", "Prevent SNAT to locally running containers",
 			"-j", "RETURN"); err != nil {
@@ -609,7 +633,7 @@ func linkSetUpByName(linkName string) error {
 	return netlink.LinkSetUp(link)
 }
 
-func reexpose(config *BridgeConfig, log *logrus.Logger) error {
+func Reexpose(config *BridgeConfig, log *logrus.Logger) error {
 	// Get existing IP addrs of the weave bridge.
 	// If the bridge hasn't been exposed, then this functions does nothing.
 	//
diff --git a/net/expose.go b/net/expose.go
index 76ba41a1..d35f781e 100644
--- a/net/expose.go
+++ b/net/expose.go
@@ -103,13 +103,8 @@ func addNatRule(ipt *iptables.IPTables, rulespec ...string) error {
 	// Loop until we get an exit code other than "temporarily unavailable"
 	for {
 		if err := ipt.AppendUnique("nat", "WEAVE", rulespec...); err != nil {
-			if ierr, ok := err.(*iptables.Error); ok {
-				if status, ok := ierr.ExitError.Sys().(syscall.WaitStatus); ok {
-					// (magic exit code 4 found in iptables source code; undocumented)
-					if status.ExitStatus() == 4 {
-						continue
-					}
-				}
+			if isResourceError(err) {
+				continue
 			}
 			return err
 		}
diff --git a/net/ipset/ipset.go b/net/ipset/ipset.go
index b538a6c9..4c0b7481 100644
--- a/net/ipset/ipset.go
+++ b/net/ipset/ipset.go
@@ -26,7 +26,8 @@ type Interface interface {
 	Create(ipsetName Name, ipsetType Type) error
 	AddEntry(user types.UID, ipsetName Name, entry string, comment string) error
 	DelEntry(user types.UID, ipsetName Name, entry string) error
-	Exist(user types.UID, ipsetName Name, entry string) bool
+	EntryExists(user types.UID, ipsetName Name, entry string) bool
+	Exists(ipsetName Name) (bool, error)
 	Flush(ipsetName Name) error
 	Destroy(ipsetName Name) error
 
@@ -121,10 +122,24 @@ func (i *ipset) DelEntry(user types.UID, ipsetName Name, entry string) error {
 	return doExec("del", string(ipsetName), entry)
 }
 
-func (i *ipset) Exist(user types.UID, ipsetName Name, entry string) bool {
+func (i *ipset) EntryExists(user types.UID, ipsetName Name, entry string) bool {
 	return i.existUser(user, ipsetName, entry)
 }
 
+// Dummy way to check whether a given ipset exists.
+func (i *ipset) Exists(name Name) (bool, error) {
+	sets, err := i.List(string(name))
+	if err != nil {
+		return false, err
+	}
+	for _, s := range sets {
+		if s == name {
+			return true, nil
+		}
+	}
+	return false, nil
+}
+
 func (i *ipset) Flush(ipsetName Name) error {
 	i.removeSetFromUsers(ipsetName)
 	return doExec("flush", string(ipsetName))
diff --git a/net/iptables.go b/net/iptables.go
index a11ca533..cda4655a 100644
--- a/net/iptables.go
+++ b/net/iptables.go
@@ -2,9 +2,13 @@ package net
 
 import (
 	"strings"
+	"syscall"
+	"time"
 
 	"github.com/coreos/go-iptables/iptables"
 	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
+	utilwait "k8s.io/apimachinery/pkg/util/wait"
 )
 
 // AddChainWithRules creates a chain and appends given rules to it.
@@ -104,3 +108,119 @@ func ensureRulesAtTop(table, chain string, rulespecs [][]string, ipt *iptables.I
 
 	return nil
 }
+
+func chainExists(ipt *iptables.IPTables, table string, chain string) (bool, error) {
+	existingChains, err := ipt.ListChains(table)
+	if err != nil {
+		return false, errors.Wrapf(err, "ipt.ListChains(%s)", table)
+	}
+	chainMap := make(map[string]struct{})
+	for _, c := range existingChains {
+		chainMap[c] = struct{}{}
+	}
+
+	_, found := chainMap[chain]
+	return found, nil
+}
+
+const (
+	// Max time we wait for an iptables flush to complete after we notice it has started
+	iptablesFlushTimeout = 5 * time.Second
+	// How often we poll while waiting for an iptables flush to complete
+	iptablesFlushPollTime = 100 * time.Millisecond
+)
+
+// MonitorForIptablesFlush periodically checks for a canary chain in iptables. If this canary chain goes missing it calls the reloadFunc.
+// This is a more efficient way of detecting whether firewalld or another process has been removing rules that we rely on.
+// The reloadFunc can then check whether other chains that should exist are still there, fix things and restore the canary.
+func MonitorForIptablesFlush(log *logrus.Logger, canary string, tables []string, reloadFunc func(), interval time.Duration, stopCh <-chan struct{}) {
+	ipt, err := iptables.New()
+	if err != nil {
+		log.Errorf("creating iptables object while initializing iptable Monitoring %s", err)
+		return
+	}
+
+	for {
+		_ = PollImmediateUntil(interval, func() (bool, error) {
+			for _, table := range tables {
+				if err := ensureChains(ipt, table, canary); err != nil {
+					log.Warningf("Could not set up iptables canary %s/%s: %v", table, canary, err)
+					return false, nil
+				}
+			}
+			return true, nil
+		}, stopCh)
+
+		// Poll until stopCh is closed or iptables is flushed
+		err = utilwait.PollUntil(interval, func() (bool, error) {
+			if exists, err := chainExists(ipt, tables[0], canary); exists {
+				return false, nil
+			} else if isResourceError(err) {
+				log.Warningf("Could not check for iptables canary %s/%s: %v", tables[0], canary, err)
+				return false, nil
+			}
+			log.Infof("iptables canary %s/%s deleted", tables[0], canary)
+
+			// Wait for the other canaries to be deleted too before returning
+			// so we don't start reloading too soon.
+			err := utilwait.PollImmediate(iptablesFlushPollTime, iptablesFlushTimeout, func() (bool, error) {
+				for i := 1; i < len(tables); i++ {
+					if exists, err := chainExists(ipt, tables[i], canary); exists || isResourceError(err) {
+						return false, nil
+					}
+				}
+				return true, nil
+			})
+			if err != nil {
+				log.Warning("Inconsistent iptables state detected.")
+			}
+			return true, nil
+		}, stopCh)
+
+		if err != nil {
+			// stopCh was closed
+			for _, table := range tables {
+				_ = ipt.DeleteChain(table, canary)
+			}
+			return
+		}
+
+		log.Infof("Reloading after iptables flush")
+		reloadFunc()
+	}
+}
+
+const iptablesStatusResourceProblem = 4
+
+// isResourceError returns true if the error indicates that iptables ran into a "resource
+// problem" and was unable to attempt the request. In particular, this will be true if it
+// times out trying to get the iptables lock.
+func isResourceError(err error) bool {
+	if ierr, ok := err.(*iptables.Error); ok {
+		if status, ok := ierr.ExitError.Sys().(syscall.WaitStatus); ok {
+			return status.ExitStatus() == iptablesStatusResourceProblem
+		}
+	}
+
+	return false
+}
+
+// PollImmediateUntil tries a condition func until it returns true, an error or stopCh is closed.
+//
+// PollImmediateUntil runs the 'condition' before waiting for the interval.
+// 'condition' will always be invoked at least once.
+func PollImmediateUntil(interval time.Duration, condition utilwait.ConditionFunc, stopCh <-chan struct{}) error {
+	done, err := condition()
+	if err != nil {
+		return err
+	}
+	if done {
+		return nil
+	}
+	select {
+	case <-stopCh:
+		return utilwait.ErrWaitTimeout
+	default:
+		return utilwait.PollUntil(interval, condition, stopCh)
+	}
+}
diff --git a/npc/controller_test.go b/npc/controller_test.go
index 860b5e5c..618e453c 100644
--- a/npc/controller_test.go
+++ b/npc/controller_test.go
@@ -72,12 +72,17 @@ func (i *mockIPSet) DelEntry(user types.UID, ipsetName ipset.Name, entry string)
 	return nil
 }
 
-func (i *mockIPSet) Exist(user types.UID, ipsetName ipset.Name, entry string) bool {
+func (i *mockIPSet) EntryExists(user types.UID, ipsetName ipset.Name, entry string) bool {
 	_, ok := i.sets[string(ipsetName)].subSets[entry][user]
 	return ok
 }
 
-func (i *mockIPSet) entryExists(ipsetName ipset.Name, entry string) bool {
+func (i *mockIPSet) Exists(ipsetName ipset.Name) (bool, error) {
+	_, ok := i.sets[string(ipsetName)]
+	return ok, nil
+}
+
+func (i *mockIPSet) entriesExist(ipsetName ipset.Name, entry string) bool {
 	return len(i.sets[string(ipsetName)].subSets[entry]) > 0
 }
 
@@ -262,8 +267,8 @@ func TestDefaultAllow(t *testing.T) {
 	controller.AddPod(podFoo)
 
 	// Should add the foo pod to default-allow
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, fooPodIP))
-	require.True(t, m.entryExists(egressDefaultAllowIPSetName, fooPodIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, fooPodIP))
+	require.True(t, m.entriesExist(egressDefaultAllowIPSetName, fooPodIP))
 
 	podBar := &coreapi.Pod{
 		ObjectMeta: metav1.ObjectMeta{
@@ -278,8 +283,8 @@ func TestDefaultAllow(t *testing.T) {
 	controller.UpdatePod(podBarNoIP, podBar)
 
 	// Should add the bar pod to default-allow
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, barPodIP))
-	require.True(t, m.entryExists(egressDefaultAllowIPSetName, barPodIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, barPodIP))
+	require.True(t, m.entriesExist(egressDefaultAllowIPSetName, barPodIP))
 
 	// Allow access from the bar pod to the foo pod
 	netpol := &networkingv1.NetworkPolicy{
@@ -310,48 +315,48 @@ func TestDefaultAllow(t *testing.T) {
 	controller.AddNetworkPolicy(netpol)
 
 	// Should remove the foo pod from default-allow as the netpol selects it
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, fooPodIP))
-	require.False(t, m.entryExists(egressDefaultAllowIPSetName, fooPodIP))
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, barPodIP))
-	require.True(t, m.entryExists(egressDefaultAllowIPSetName, barPodIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, fooPodIP))
+	require.False(t, m.entriesExist(egressDefaultAllowIPSetName, fooPodIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, barPodIP))
+	require.True(t, m.entriesExist(egressDefaultAllowIPSetName, barPodIP))
 
 	podBarWithNewIP := *podBar
 	podBarWithNewIP.Status.PodIP = barPodNewIP
 	controller.UpdatePod(podBar, &podBarWithNewIP)
 
 	// Should update IP addr of the bar pod in default-allow
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, barPodIP))
-	require.False(t, m.entryExists(egressDefaultAllowIPSetName, barPodIP))
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, barPodNewIP))
-	require.True(t, m.entryExists(egressDefaultAllowIPSetName, barPodNewIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, barPodIP))
+	require.False(t, m.entriesExist(egressDefaultAllowIPSetName, barPodIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, barPodNewIP))
+	require.True(t, m.entriesExist(egressDefaultAllowIPSetName, barPodNewIP))
 
 	controller.UpdatePod(&podBarWithNewIP, podBarNoIP)
 	// Should remove the bar pod from default-allow as it does not have any IP addr
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, barPodNewIP))
-	require.False(t, m.entryExists(egressDefaultAllowIPSetName, barPodNewIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, barPodNewIP))
+	require.False(t, m.entriesExist(egressDefaultAllowIPSetName, barPodNewIP))
 
 	podFooWithNewLabel := *podFoo
 	podFooWithNewLabel.ObjectMeta.Labels = map[string]string{"run": "new-foo"}
 	controller.UpdatePod(podFoo, &podFooWithNewLabel)
 
 	// Should bring back the foo pod to default-allow as it does not match dst of any netpol
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, fooPodIP))
-	require.True(t, m.entryExists(egressDefaultAllowIPSetName, fooPodIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, fooPodIP))
+	require.True(t, m.entriesExist(egressDefaultAllowIPSetName, fooPodIP))
 
 	controller.UpdatePod(&podFooWithNewLabel, podFoo)
 	// Should remove from default-allow as it matches the netpol after the update
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, fooPodIP))
-	require.False(t, m.entryExists(egressDefaultAllowIPSetName, fooPodIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, fooPodIP))
+	require.False(t, m.entriesExist(egressDefaultAllowIPSetName, fooPodIP))
 
 	controller.DeleteNetworkPolicy(netpol)
 	// Should bring back the foo pod to default-allow as no netpol selects it
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, fooPodIP))
-	require.True(t, m.entryExists(egressDefaultAllowIPSetName, fooPodIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, fooPodIP))
+	require.True(t, m.entriesExist(egressDefaultAllowIPSetName, fooPodIP))
 
 	controller.DeletePod(podFoo)
 	// Should remove foo pod from default-allow
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, fooPodIP))
-	require.False(t, m.entryExists(egressDefaultAllowIPSetName, fooPodIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, fooPodIP))
+	require.False(t, m.entriesExist(egressDefaultAllowIPSetName, fooPodIP))
 
 	controller.DeleteNamespace(defaultNamespace)
 	// Should remove default ipset
@@ -388,7 +393,7 @@ func TestOutOfOrderPodEvents(t *testing.T) {
 	controller.AddPod(podFoo)
 
 	// Should be in default-allow as no netpol selects podFoo
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
 
 	netpol := &networkingv1.NetworkPolicy{
 		ObjectMeta: metav1.ObjectMeta{
@@ -410,7 +415,7 @@ func TestOutOfOrderPodEvents(t *testing.T) {
 	controller.AddNetworkPolicy(netpol)
 
 	// Shouldn't be in default-allow as netpol above selects podFoo
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
 
 	podBar := &coreapi.Pod{
 		ObjectMeta: metav1.ObjectMeta{
@@ -422,10 +427,10 @@ func TestOutOfOrderPodEvents(t *testing.T) {
 	controller.AddPod(podBar)
 
 	// Should be in default-allow as no netpol selects podBar
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
-	require.True(t, m.Exist(podBar.ObjectMeta.UID, ingressDefaultAllowIPSetName, podIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
+	require.True(t, m.EntryExists(podBar.ObjectMeta.UID, ingressDefaultAllowIPSetName, podIP))
 	// Should be in run=bar ipset
-	require.True(t, m.entryExists(runBarIPSetName, podIP))
+	require.True(t, m.entriesExist(runBarIPSetName, podIP))
 
 	controller.DeletePod(podFoo)
 	// Multiple duplicate events should not affect npc state
@@ -434,13 +439,13 @@ func TestOutOfOrderPodEvents(t *testing.T) {
 
 	// Should be in default-allow as no netpol selects podBar and podFoo removal
 	// should not affect podBar in default-allow
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
 
 	controller.DeletePod(podBar)
 
 	// Should remove from default-allow and run=bar ipsets
 	require.Equal(t, 0, len(m.sets[ingressDefaultAllowIPSetName].subSets))
-	require.False(t, m.entryExists(runBarIPSetName, podIP))
+	require.False(t, m.entriesExist(runBarIPSetName, podIP))
 }
 
 // Test case for https://github.com/weaveworks/weave/issues/3222
@@ -471,7 +476,7 @@ func TestNewTargetSelector(t *testing.T) {
 		Status: coreapi.PodStatus{PodIP: podIP}}
 	controller.AddPod(podFoo)
 
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
 
 	netpolBar := &networkingv1.NetworkPolicy{
 		ObjectMeta: metav1.ObjectMeta{
@@ -492,7 +497,7 @@ func TestNewTargetSelector(t *testing.T) {
 	controller.AddNetworkPolicy(netpolBar)
 
 	// netpolBar target selector selects podFoo
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
 
 	netpolFoo := &networkingv1.NetworkPolicy{
 		ObjectMeta: metav1.ObjectMeta{
@@ -514,10 +519,10 @@ func TestNewTargetSelector(t *testing.T) {
 
 	controller.DeleteNetworkPolicy(netpolBar)
 	// netpolFoo target-selects podFoo
-	require.False(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
+	require.False(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
 	controller.DeleteNetworkPolicy(netpolFoo)
 	// No netpol target-selects podFoo
-	require.True(t, m.entryExists(ingressDefaultAllowIPSetName, podIP))
+	require.True(t, m.entriesExist(ingressDefaultAllowIPSetName, podIP))
 }
 
 func TestEgressPolicyWithIPBlock(t *testing.T) {
@@ -573,8 +578,8 @@ func TestEgressPolicyWithIPBlock(t *testing.T) {
 	require.NoError(t, err)
 
 	require.Equal(t, 2, len(m.sets[exceptIPSetName].subSets))
-	require.True(t, m.entryExists(exceptIPSetName, "192.168.48.1/32"))
-	require.True(t, m.entryExists(exceptIPSetName, "192.168.48.2/32"))
+	require.True(t, m.entriesExist(exceptIPSetName, "192.168.48.1/32"))
+	require.True(t, m.entriesExist(exceptIPSetName, "192.168.48.2/32"))
 
 	// Each egress rule is represented as two iptables rules (-J MARK and -J RETURN).
 	require.Equal(t, 2, len(ipt.rules[EgressCustomChain]))
@@ -683,7 +688,7 @@ func TestIngressPolicyWithIPBlockAndPortSpecified(t *testing.T) {
 	defer controller.DeleteNetworkPolicy(netpolicty)
 
 	require.Equal(t, 1, len(m.sets[runBarIPSetName].subSets))
-	require.True(t, m.entryExists(runBarIPSetName, barPodIP))
+	require.True(t, m.entriesExist(runBarIPSetName, barPodIP))
 
 	require.Equal(t, 1, len(ipt.rules[IngressChain]))
 	for rule := range ipt.rules[IngressChain] {
diff --git a/npc/namespace.go b/npc/namespace.go
index 3c061111..6423b4cd 100644
--- a/npc/namespace.go
+++ b/npc/namespace.go
@@ -441,7 +441,7 @@ func (ns *ns) updateDefaultAllowIPSetEntry(oldObj, newObj *coreapi.Pod, ipsetNam
 	// Instead of iterating over all selectors we check whether old pod IP
 	// has been inserted into default-allow ipset to decide whether the IP
 	// in the ipset has to be updated.
-	if ns.ips.Exist(oldObj.ObjectMeta.UID, ipsetName, oldObj.Status.PodIP) {
+	if ns.ips.EntryExists(oldObj.ObjectMeta.UID, ipsetName, oldObj.Status.PodIP) {
 
 		if err := ns.ips.DelEntry(oldObj.ObjectMeta.UID, ipsetName, oldObj.Status.PodIP); err != nil {
 			return err
diff --git a/prog/weave-npc/main.go b/prog/weave-npc/main.go
index 4da4a1de..22e4e556 100644
--- a/prog/weave-npc/main.go
+++ b/prog/weave-npc/main.go
@@ -196,7 +196,7 @@ func createBaseRules(ipt *iptables.IPTables, ips ipset.Interface) error {
 	}
 
 	// delete `weave-local-pods` ipset which is no longer used by weave-npc
-	weaveLocalPodExist, err := ipsetExist(ips, npc.LocalIpset)
+	weaveLocalPodExist, err := ips.Exists(npc.LocalIpset)
 	if err != nil {
 		common.Log.Errorf("Failed to look if ipset '%s' exists", npc.LocalIpset)
 	} else if weaveLocalPodExist {
@@ -209,22 +209,6 @@ func createBaseRules(ipt *iptables.IPTables, ips ipset.Interface) error {
 	return nil
 }
 
-// Dummy way to check whether a given ipset exists.
-// TODO(brb) Use "ipset -exist create <..>" for our purpose instead (for some reasons
-// creating an ipset with -exist fails).
-func ipsetExist(ips ipset.Interface, name ipset.Name) (bool, error) {
-	sets, err := ips.List(string(name))
-	if err != nil {
-		return false, err
-	}
-	for _, s := range sets {
-		if s == name {
-			return true, nil
-		}
-	}
-	return false, nil
-}
-
 func root(cmd *cobra.Command, args []string) {
 	var npController cache.Controller
 
diff --git a/prog/weaver/main.go b/prog/weaver/main.go
index 3edaed16..811fc5d9 100644
--- a/prog/weaver/main.go
+++ b/prog/weaver/main.go
@@ -313,8 +313,11 @@ func main() {
 		}
 	}
 	ips := ipset.New(common.LogLogger(), 0)
+	err = weavenet.ResetIPTables(&bridgeConfig, ips)
+	checkFatal(err)
 	bridgeType, err := weavenet.EnsureBridge(procPath, &bridgeConfig, Log, ips)
 	checkFatal(err)
+
 	Log.Println("Bridge type is", bridgeType)
 
 	config.Password = determinePassword(password)
@@ -512,6 +515,18 @@ func main() {
 		go exposeForAWSVPC(allocator, defaultSubnet, bridgeConfig.WeaveBridgeName, waitReady.Add())
 	}
 
+	applyIPTables := func() {
+		Log.Info("Re-configuring iptables")
+		err := weavenet.ConfigureIPTables(&bridgeConfig, ips)
+		if err != nil {
+			Log.Errorf("Error configuring iptables: %s", err)
+		}
+		weavenet.Reexpose(&bridgeConfig, Log)
+	}
+	stopChan := make(chan struct{})
+	go weavenet.MonitorForIptablesFlush(Log, "WEAVE-CANARY", []string{"mangle", "nat", "filter"}, applyIPTables, 10*time.Second, stopChan)
+	defer close(stopChan)
+
 	signals.SignalHandlerLoop(common.Log, router)
 }
 
diff --git a/test/135_expose_iptables_refresh.sh b/test/135_expose_iptables_refresh.sh
new file mode 100755
index 00000000..a844e7cc
--- /dev/null
+++ b/test/135_expose_iptables_refresh.sh
@@ -0,0 +1,47 @@
+#!/bin/bash
+
+. "$(dirname "$0")/config.sh"
+
+run_on1() {
+    assert_raises "run_on $HOST1 $@"
+}
+
+weave_on1() {
+    assert_raises "weave_on $HOST1 $@"
+}
+
+stop_weave_on1() {
+    assert_raises "stop_weave_on $HOST1 $@"
+}
+
+get_weave_iptable_rules() {
+    get_command_output_on $HOST1 "sudo iptables-save | grep -i weave"
+}
+
+wait_for_iptable_refresh() {
+    # Canary is checked every 10s
+    sleep 12
+}
+
+start_suite "exposing weave network to host"
+
+# Launch
+weave_on1 "launch"
+IPT_BEFORE=$(get_weave_iptable_rules)
+run_on1 "sudo iptables -t nat -X WEAVE-CANARY"
+wait_for_iptable_refresh
+IPT_AFTER=$(get_weave_iptable_rules)
+assert_raises "diff <(echo $IPT_BEFORE) <(echo $IPT_AFTER)" 0
+stop_weave_on1
+
+# After exposing
+weave_on1 "launch"
+weave_on1 "expose"
+IPT_BEFORE=$(get_weave_iptable_rules)
+run_on1 "sudo iptables -t nat -X WEAVE-CANARY"
+wait_for_iptable_refresh
+IPT_AFTER=$(get_weave_iptable_rules)
+assert_raises "diff <(echo $IPT_BEFORE) <(echo $IPT_AFTER)" 0
+stop_weave_on1
+
+end_suite