feat!: serve the actions cache from the runner (#1229)

Make cache requests go through the runner and remove outbound traffic from the cache server. The runner forwards them to `cache.external_server` when one is set, and attaches itself to each job network so jobs can reach it.

Benefits:

1. Jobs need no http proxy rules for the cache server, which `@actions/http-client` mishandles for CIDR ranges in `no_proxy` (https://github.com/actions/toolkit/pull/2495).
2. Only runners reach a shared cache server, which needs no route to Gitea.
3. A runner container sharing the host Docker socket now caches, where jobs previously could not reach it.

Breaks:

1. Jobs that reach `external_server` but not their runner, such as a remote `DOCKER_HOST`. The built-in cache has always needed that reachability, so both modes now agree. Set `cache.host` and `cache.port` to an address those jobs reach.
2. Jobs on `container.network: bridge`, where Docker rejects the alias, so the runner cannot attach and their cache fails as it does today. Leave `container.network` empty, or name a user-defined network.

*Written by Claude.*

---------

Co-authored-by: bircni <[email protected]>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1229
Reviewed-by: bircni <[email protected]>
Co-authored-by: silverwind <[email protected]>

act/container/docker_network_test.go

@@ -18,11 +18,12 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-func TestIsolatedNetwork(t *testing.T) {
+func TestIsolatedCacheContainer(t *testing.T) {
 	ctx := context.Background()
 	cache, lan := netip.MustParseAddr("172.18.0.3"), netip.MustParseAddr("192.168.1.20")
 	client := &mockDockerClient{}
 	client.On("ContainerList", ctx, mobyclient.ContainerListOptions{}).Return(mobyclient.ContainerListResult{Items: []container.Summary{{}, {
+		ID: "0123456789abcdef",
 		NetworkSettings: &container.NetworkSettingsSummary{Networks: map[string]*network.EndpointSettings{
 			"compose": {NetworkID: "c0ffee", IPAddress: cache},
 			"lan":     {NetworkID: "beef", IPAddress: lan},
@@ -38,12 +39,12 @@ func TestIsolatedNetwork(t *testing.T) {
 			Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Network: inspected}}, nil)
 	}
 	check := func(addr netip.Addr, jobNetwork, want string) {
-		got, err := isolatedNetwork(ctx, client, addr, jobNetwork)
+		got, err := isolatedCacheContainer(ctx, client, addr, jobNetwork)
 		require.NoError(t, err)
 		assert.Equal(t, want, got, jobNetwork)
 	}
-	check(cache, "", "compose")
-	check(cache, "jobs", "compose")
+	check(cache, "", "0123456789ab")
+	check(cache, "jobs", "0123456789ab")
 	check(cache, "c0f", "")
 	check(cache, "host", "")
 	check(lan, "", "")
@@ -56,9 +57,44 @@ func TestIsAddressPoolExhausted(t *testing.T) {
 	assert.False(t, isAddressPoolExhausted(cerrdefs.ErrInvalidArgument.WithMessage("invalid subnet 10.0.0.0/8: it overlaps with an existing network")))
 }
 
-// Of this runner's networks, only the ones nothing is attached to and old enough to predate
-// any job now starting are the runner's to reclaim. An unexpected NetworkRemove fails the
-// test on its own, since testify has no expectation to match it against.
+func (m *mockDockerClient) NetworkConnect(ctx context.Context, id string, opts mobyclient.NetworkConnectOptions) (mobyclient.NetworkConnectResult, error) {
+	return mobyclient.NetworkConnectResult{}, m.Called(ctx, id, opts).Error(0)
+}
+
+func (m *mockDockerClient) NetworkDisconnect(ctx context.Context, id string, opts mobyclient.NetworkDisconnectOptions) (mobyclient.NetworkDisconnectResult, error) {
+	return mobyclient.NetworkDisconnectResult{}, m.Called(ctx, id, opts).Error(0)
+}
+
+func TestConnectAndDisconnectNetwork(t *testing.T) {
+	ctx := context.Background()
+	connect := func(answer error) error {
+		client := &mockDockerClient{}
+		client.On("NetworkConnect", ctx, "jobs", mobyclient.NetworkConnectOptions{
+			Container:      "0123456789ab",
+			EndpointConfig: &network.EndpointSettings{Aliases: []string{"0123456789ab"}},
+		}).Return(answer).Once()
+		return connectNetwork(ctx, client, "jobs", "0123456789ab")
+	}
+	require.NoError(t, connect(nil))
+	require.NoError(t, connect(errors.New("endpoint with name runner already exists in network jobs")))
+	require.NoError(t, connect(errors.New("container runner already connected to network jobs")))
+	require.ErrorIs(t, connect(context.Canceled), context.Canceled)
+
+	disconnect := func(answer error) error {
+		client := &mockDockerClient{}
+		client.On("NetworkDisconnect", ctx, "jobs", mobyclient.NetworkDisconnectOptions{
+			Container: "0123456789ab", Force: true,
+		}).Return(answer).Once()
+		return disconnectNetwork(ctx, client, "jobs", "0123456789ab")
+	}
+	require.NoError(t, disconnect(cerrdefs.ErrNotFound))
+	require.NoError(t, disconnect(errors.New("container runner is not connected to the network jobs")))
+	require.ErrorIs(t, disconnect(context.Canceled), context.Canceled)
+}
+
+// Of this runner's networks, only the ones nothing but its own cache container is attached to
+// and old enough to predate any job now starting are the runner's to reclaim. An unexpected
+// NetworkRemove fails the test on its own, since testify has no expectation to match it against.
 func TestRemoveOrphanNetworks(t *testing.T) {
 	ctx := context.Background()
 	cutoff := time.Date(2026, time.April, 29, 20, 0, 0, 0, time.UTC)
@@ -69,16 +105,29 @@ func TestRemoveOrphanNetworks(t *testing.T) {
 		{ID: "orphan"},
 		{ID: "busy"},
 		{ID: "starting"},
+		{ID: "cached"},
+		{ID: "shared"},
 	}}, nil)
 	client.On("NetworkInspect", ctx, "orphan", mobyclient.NetworkInspectOptions{}).
 		Return(mobyclient.NetworkInspectResult{}, nil)
 	client.On("NetworkInspect", ctx, "busy", mobyclient.NetworkInspectOptions{}).
 		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Containers: map[string]network.EndpointResource{"c": {}}}}, nil)
 	client.On("NetworkInspect", ctx, "starting", mobyclient.NetworkInspectOptions{}).
-		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Network: network.Network{Created: cutoff.Add(time.Second)}}}, nil)
+		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{
+			Network:    network.Network{Created: cutoff.Add(time.Second)},
+			Containers: map[string]network.EndpointResource{"0123456789abcdef": {}},
+		}}, nil)
+	client.On("NetworkInspect", ctx, "cached", mobyclient.NetworkInspectOptions{}).
+		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Containers: map[string]network.EndpointResource{"0123456789abcdef": {}}}}, nil)
+	client.On("NetworkInspect", ctx, "shared", mobyclient.NetworkInspectOptions{}).
+		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Containers: map[string]network.EndpointResource{"0123456789abcdef": {}, "job": {}}}}, nil)
 	client.On("NetworkRemove", ctx, "orphan", mobyclient.NetworkRemoveOptions{}).
 		Return(mobyclient.NetworkRemoveResult{}, nil)
+	client.On("NetworkDisconnect", ctx, "cached", mobyclient.NetworkDisconnectOptions{Container: "0123456789abcdef", Force: true}).
+		Return(nil).Once()
+	client.On("NetworkRemove", ctx, "cached", mobyclient.NetworkRemoveOptions{}).
+		Return(mobyclient.NetworkRemoveResult{}, nil).Once()
 
-	require.NoError(t, removeOrphanNetworks(ctx, client, "runner-1", cutoff))
+	require.NoError(t, removeOrphanNetworks(ctx, client, "runner-1", "0123456789ab", cutoff))
 	client.AssertExpectations(t)
 }