fix: stop the docker proxy probe from leaking its container (#1357)

When the probe's deadline expired during `ContainerCreate`, dockerd still created the container but the probe never got its ID, leaving it in `Created` with its image's anonymous volumes. A slow daemon leaked one per job.

The probe now runs detached with its own timeout, so the create is never cancelled and the container is always removed. Callers still fall back after their own deadline, and the runner waits for pending probes before exiting.

Fixes: https://gitea.com/gitea/runner/issues/1353
Reviewed-on: https://gitea.com/gitea/runner/pulls/1357
Reviewed-by: Zettat123 <[email protected]>
Co-authored-by: silverwind <[email protected]>
This commit is contained in:
silverwind authored and silverwind committed 2026-10-29 15:53:32 +00:00
1 parent e3607b576e
commit c6567296a0
3 files changed
+195 -93

No files matched your search

+122 -56
View File
@@ -94,6 +94,8 @@ func TestDockerProxy(t *testing.T) {
return
}
_, _ = w.Write([]byte(`[{"Id":"probe-image"}]`))
case strings.HasSuffix(r.URL.Path, "/containers/json"):
_, _ = w.Write([]byte(`[]`))
case strings.HasSuffix(r.URL.Path, "/json"):
http.NotFound(w, r)
case strings.HasSuffix(r.URL.Path, "/archive"):
@@ -235,7 +237,8 @@ func TestDockerProxy(t *testing.T) {
t.Skip("automatic proxy discovery requires Linux")
}
t.Setenv("DOCKER_HOST", "unix://"+daemonSocket)
t.Setenv("TMPDIR", shortTempDir(t))
tempDir := shortTempDir(t)
t.Setenv("TMPDIR", tempDir)
require.Nil(t, NewDockerProxy(t.Context(), "failed"))
for _, job := range []string{"first", "second"} {
proxy := NewDockerProxy(t.Context(), job)
@@ -243,12 +246,10 @@ func TestDockerProxy(t *testing.T) {
require.NoError(t, proxy.Close(context.Background()))
}
assert.EqualValues(t, 2, probes.Load())
dockerProxyDirectories.Range(func(key, value any) bool {
directory, _ := value.(dockerProxyDirectory)
directory.socket, _ = os.Stat(t.TempDir())
dockerProxyDirectories.Store(key, directory)
return true
})
workdir, err := os.Getwd()
require.NoError(t, err)
cached, _ := dockerProxyProbes.Load([3]string{"unix://" + daemonSocket, workdir, tempDir})
cached.(*dockerProxyProbe).directory.socket, _ = os.Stat(t.TempDir())
proxy := NewDockerProxy(t.Context(), "replaced daemon")
require.NotNil(t, proxy)
require.NoError(t, proxy.Close(context.Background()))
@@ -476,10 +477,23 @@ func BenchmarkDockerProxy(b *testing.B) {
type probeClient struct {
mobyclient.APIClient
create func(mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error)
remove func(context.Context, string, mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error)
stat func(string) error
mounts []container.MountPoint
create func(context.Context, mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error)
remove func(context.Context, string, mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error)
stat func(string) error
mounts []container.MountPoint
containers []container.Summary
}
func (c *probeClient) Close() error {
return nil
}
func (c *probeClient) ContainerList(context.Context, mobyclient.ContainerListOptions) (mobyclient.ContainerListResult, error) {
return mobyclient.ContainerListResult{Items: c.containers}, nil
}
func (c *probeClient) ContainerKill(context.Context, string, mobyclient.ContainerKillOptions) (mobyclient.ContainerKillResult, error) {
return mobyclient.ContainerKillResult{}, cerrdefs.ErrConflict
}
func (c *probeClient) ContainerInspect(context.Context, string, mobyclient.ContainerInspectOptions) (mobyclient.ContainerInspectResult, error) {
@@ -490,8 +504,8 @@ func (c *probeClient) ImageList(context.Context, mobyclient.ImageListOptions) (m
return mobyclient.ImageListResult{Items: []image.Summary{{ID: "removed-image"}, {ID: "probe-image"}}}, nil
}
func (c *probeClient) ContainerCreate(_ context.Context, opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
return c.create(opts)
func (c *probeClient) ContainerCreate(ctx context.Context, opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
return c.create(ctx, opts)
}
func (c *probeClient) ContainerRemove(ctx context.Context, id string, opts mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error) {
@@ -518,7 +532,7 @@ func TestDockerProxyDirectoryProbes(t *testing.T) {
var calls, removals atomic.Int32
visible := false
cli := &probeClient{
create: func(_opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
create: func(_ctx context.Context, _opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
calls.Add(1)
<-release
return mobyclient.ContainerCreateResult{ID: "probe"}, createErr
@@ -538,23 +552,18 @@ func TestDockerProxyDirectoryProbes(t *testing.T) {
results := make(chan result, 8)
for range cap(results) {
go func() {
_, seen, err := probeDockerProxyDirectory(t.Context(), cli, key, socket)
results <- result{seen, err}
probe, err := probeDockerProxyDirectory(t.Context(), cli, key, socket)
results <- result{probe.seen, err}
}()
}
synctest.Wait()
assert.EqualValues(t, 1, calls.Load())
ctx, cancel := context.WithCancel(t.Context())
cancel()
_, seen, err := probeDockerProxyDirectory(ctx, cli, key, socket)
require.ErrorIs(t, err, context.Canceled)
assert.False(t, seen)
replacement, err := os.Stat(t.TempDir())
require.NoError(t, err)
replaced := make(chan result, 1)
go func() {
_, seen, err := probeDockerProxyDirectory(t.Context(), cli, key, replacement)
replaced <- result{seen, err}
probe, err := probeDockerProxyDirectory(t.Context(), cli, key, replacement)
replaced <- result{probe.seen, err}
}()
synctest.Wait()
assert.EqualValues(t, 2, calls.Load())
@@ -574,11 +583,11 @@ func TestDockerProxyDirectoryProbes(t *testing.T) {
assert.Zero(t, removals.Load())
}
createErr, visible = nil, true
directory, seen, err := probeDockerProxyDirectory(t.Context(), cli, key, socket)
probe, err := probeDockerProxyDirectory(t.Context(), cli, key, socket)
require.NoError(t, err)
assert.True(t, seen)
assert.Equal(t, dir, directory.dir)
assert.Equal(t, dir, directory.daemonDir)
assert.True(t, probe.seen)
assert.Equal(t, dir, probe.directory.dir)
assert.Equal(t, dir, probe.directory.daemonDir)
assert.EqualValues(t, 3, calls.Load())
entries, err := os.ReadDir(dir)
require.NoError(t, err)
@@ -586,7 +595,7 @@ func TestDockerProxyDirectoryProbes(t *testing.T) {
})
})
}
t.Run("cancelled probe does not cancel another caller", func(t *testing.T) {
t.Run("cancelled caller leaves the probe to finish for the others", func(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
dir := t.TempDir()
socket, err := os.Stat(t.TempDir())
@@ -595,49 +604,105 @@ func TestDockerProxyDirectoryProbes(t *testing.T) {
ctx, cancel := context.WithCancel(t.Context())
defer cancel()
release := make(chan struct{})
calls := 0
var calls, removals atomic.Int32
cli := &probeClient{
create: func(_opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
calls++
if calls == 1 {
<-release
create: func(ctx context.Context, _opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
calls.Add(1)
<-release
return mobyclient.ContainerCreateResult{ID: "probe"}, ctx.Err()
},
stat: func(_path string) error { return nil },
remove: func(_ctx context.Context, _id string, _opts mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error) {
removals.Add(1)
return mobyclient.ContainerRemoveResult{}, nil
},
}
leader, follower, late := make(chan result, 1), make(chan result, 1), make(chan result, 1)
go func() {
probe, err := probeDockerProxyDirectory(ctx, cli, key, socket)
leader <- result{probe.seen, err}
}()
synctest.Wait()
go func() {
probe, err := probeDockerProxyDirectory(t.Context(), cli, key, socket)
follower <- result{probe.seen, err}
}()
synctest.Wait()
cancel()
leaderResult := <-leader
require.ErrorIs(t, leaderResult.err, context.Canceled)
assert.False(t, leaderResult.seen)
go func() {
probe, err := probeDockerProxyDirectory(t.Context(), cli, key, socket)
late <- result{probe.seen, err}
}()
synctest.Wait()
close(release)
for _, results := range []chan result{follower, late} {
result := <-results
require.NoError(t, result.err)
assert.True(t, result.seen)
}
assert.EqualValues(t, 1, calls.Load())
assert.EqualValues(t, 1, removals.Load())
_, cached := dockerProxyProbes.Load(key)
assert.True(t, cached)
})
})
t.Run("caller joining a probe that ran out of time retries", func(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
dir := t.TempDir()
socket, err := os.Stat(t.TempDir())
require.NoError(t, err)
var calls atomic.Int32
cli := &probeClient{
create: func(ctx context.Context, _opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
if calls.Add(1) == 1 {
<-ctx.Done()
return mobyclient.ContainerCreateResult{}, ctx.Err()
} else {
return mobyclient.ContainerCreateResult{ID: "probe"}, nil
}
return mobyclient.ContainerCreateResult{ID: "probe"}, nil
},
stat: func(_path string) error { return nil },
remove: func(_ctx context.Context, _id string, _opts mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error) {
return mobyclient.ContainerRemoveResult{}, nil
},
}
leader, follower := make(chan result, 1), make(chan result, 1)
go func() {
_, seen, err := probeDockerProxyDirectory(ctx, cli, key, socket)
leader <- result{seen, err}
}()
synctest.Wait()
go func() {
_, seen, err := probeDockerProxyDirectory(t.Context(), cli, key, socket)
follower <- result{seen, err}
}()
synctest.Wait()
assert.Equal(t, 1, calls)
cancel()
close(release)
leaderResult, followerResult := <-leader, <-follower
require.ErrorIs(t, leaderResult.err, context.Canceled)
assert.False(t, leaderResult.seen)
require.NoError(t, followerResult.err)
assert.True(t, followerResult.seen)
assert.Equal(t, 2, calls)
probe, err := probeDockerProxyDirectory(t.Context(), cli, [3]string{"daemon", dir, dir}, socket)
require.NoError(t, err)
assert.True(t, probe.seen)
assert.EqualValues(t, 2, calls.Load())
})
})
}
func TestRemoveStaleProbeContainers(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
probeMount := container.MountPoint{Type: mount.TypeBind, Destination: "/gitea-runner-probe"}
writableMount := container.MountPoint{Type: mount.TypeBind, Destination: "/gitea-runner-probe", RW: true}
var removed []string
cli := &probeClient{
containers: []container.Summary{
{ID: "abandoned", State: container.StateCreated, Mounts: []container.MountPoint{probeMount}},
{ID: "foreign", State: container.StateCreated, Mounts: []container.MountPoint{writableMount}},
{ID: "started", State: container.StateExited, Mounts: []container.MountPoint{probeMount}},
},
remove: func(_ctx context.Context, id string, _opts mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error) {
removed = append(removed, id)
return mobyclient.ContainerRemoveResult{}, nil
},
}
start := time.Now()
removeStaleProbeContainers(cli)
assert.Equal(t, []string{"abandoned"}, removed)
assert.GreaterOrEqual(t, time.Since(start), dockerProxyProbeWorkerTimeout+dockerProxyProbeTimeout)
})
}
func TestDaemonSeesDir(t *testing.T) {
dir := t.TempDir()
markers := make(map[string]bool)
for _, testCase := range []struct {
name string
daemonDir string
@@ -656,7 +721,7 @@ func TestDaemonSeesDir(t *testing.T) {
removed := false
var attemptedImages []string
cli := &probeClient{
create: func(opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
create: func(_ctx context.Context, opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
assert.Equal(t, []mount.Mount{{Type: mount.TypeBind, Source: testCase.daemonDir, Target: "/gitea-runner-probe", ReadOnly: true}}, opts.HostConfig.Mounts)
attemptedImages = append(attemptedImages, opts.Config.Image)
if opts.Config.Image == "removed-image" {
@@ -692,6 +757,7 @@ func TestDaemonSeesDir(t *testing.T) {
assert.Equal(t, []string{"removed-image", "probe-image"}, attemptedImages)
assert.Equal(t, testCase.seen, seen)
assert.Equal(t, testCase.removed, removed)
entries, err := os.ReadDir(dir)
require.NoError(t, err)
assert.Empty(t, entries)