solver: add per-step CPU and memory resource limits

client/client_test.go

@@ -154,6 +154,8 @@ var allTests = []func(t *testing.T, sb integration.Sandbox){
 	testShmSize,
 	testUlimit,
 	testCgroupParent,
+	testLinuxResources,
+	testLinuxResourcesMergeOnDedup,
 	testNetworkMode,
 	testFrontendMetadataReturn,
 	testFrontendUseSolveResults,
@@ -1242,6 +1244,131 @@ func testCgroupParent(t *testing.T, sb integration.Sandbox) {
 	require.Equal(t, "", strings.TrimSpace(string(dt)))
 }
 
+func testLinuxResources(t *testing.T, sb integration.Sandbox) {
+	integration.SkipOnPlatform(t, "windows")
+	if sb.Rootless() {
+		t.SkipNow()
+	}
+
+	if _, err := os.Lstat("/sys/fs/cgroup/cgroup.subtree_control"); os.IsNotExist(err) {
+		t.Skipf("test requires cgroup v2")
+	}
+
+	c, err := New(sb.Context(), sb.Address())
+	require.NoError(t, err)
+	defer c.Close()
+
+	img := llb.Image("alpine:latest")
+	st := llb.Scratch()
+
+	run := func(cmd string, ro ...llb.RunOption) {
+		st = img.Run(append(ro, llb.Shlex(cmd), llb.Dir("/wd"))...).AddMount("/wd", st)
+	}
+
+	// Test memory limit: set 64MiB and verify via cgroup
+	run(`sh -c "cat /sys/fs/cgroup/memory.max > mem_limited"`, llb.MemoryLimit(64*1024*1024))
+	run(`sh -c "cat /sys/fs/cgroup/memory.max > mem_default"`)
+
+	// Test CPU quota: set quota=50000 period=100000 (50% CPU) and verify
+	run(`sh -c "cat /sys/fs/cgroup/cpu.max > cpu_limited"`, llb.CPUQuota(50000), llb.CPUPeriod(100000))
+
+	def, err := st.Marshal(sb.Context())
+	require.NoError(t, err)
+
+	destDir := t.TempDir()
+
+	_, err = c.Solve(sb.Context(), def, SolveOpt{
+		Exports: []ExportEntry{
+			{
+				Type:      ExporterLocal,
+				OutputDir: destDir,
+			},
+		},
+	}, nil)
+	require.NoError(t, err)
+
+	dt, err := os.ReadFile(filepath.Join(destDir, "mem_limited"))
+	require.NoError(t, err)
+	require.Equal(t, "67108864", strings.TrimSpace(string(dt)))
+
+	dt2, err := os.ReadFile(filepath.Join(destDir, "mem_default"))
+	require.NoError(t, err)
+	require.Equal(t, "max", strings.TrimSpace(string(dt2)))
+
+	dt3, err := os.ReadFile(filepath.Join(destDir, "cpu_limited"))
+	require.NoError(t, err)
+	require.Equal(t, "50000 100000", strings.TrimSpace(string(dt3)))
+}
+
+// testLinuxResourcesMergeOnDedup verifies that when two concurrent builds share
+// the same RUN step but specify different resource limits, the most relaxed
+// (least restrictive) limit is applied.
+func testLinuxResourcesMergeOnDedup(t *testing.T, sb integration.Sandbox) {
+	integration.SkipOnPlatform(t, "windows")
+	if sb.Rootless() {
+		t.SkipNow()
+	}
+
+	if _, err := os.Lstat("/sys/fs/cgroup/cgroup.subtree_control"); os.IsNotExist(err) {
+		t.Skipf("test requires cgroup v2")
+	}
+
+	c, err := New(sb.Context(), sb.Address())
+	require.NoError(t, err)
+	defer c.Close()
+
+	// Both builds share the exact same RUN command so the solver deduplicates
+	// them into one vertex. They differ only in memory limits (OpMetadata).
+	// Dedup is guaranteed because loadUnlocked (which merges resources) loads
+	// the full vertex graph in microseconds, while image resolution that must
+	// complete before the RUN vertex can execute takes orders of magnitude longer.
+	// Both goroutines will have loaded and merged before the RUN step starts.
+	sharedCmd := `sh -c "cat /sys/fs/cgroup/memory.max > /wd/mem_limit"`
+
+	// Build 1: 64 MiB memory limit
+	st1 := llb.Image("alpine:latest").
+		Run(llb.Shlex(sharedCmd), llb.MemoryLimit(64*1024*1024), llb.Dir("/wd")).
+		AddMount("/wd", llb.Scratch())
+	def1, err := st1.Marshal(sb.Context())
+	require.NoError(t, err)
+
+	// Build 2: 128 MiB memory limit (more relaxed — should win)
+	st2 := llb.Image("alpine:latest").
+		Run(llb.Shlex(sharedCmd), llb.MemoryLimit(128*1024*1024), llb.Dir("/wd")).
+		AddMount("/wd", llb.Scratch())
+	def2, err := st2.Marshal(sb.Context())
+	require.NoError(t, err)
+
+	destDir1 := t.TempDir()
+	destDir2 := t.TempDir()
+
+	eg, egCtx := errgroup.WithContext(sb.Context())
+	eg.Go(func() error {
+		_, err := c.Solve(egCtx, def1, SolveOpt{
+			Exports: []ExportEntry{{Type: ExporterLocal, OutputDir: destDir1}},
+		}, nil)
+		return err
+	})
+	eg.Go(func() error {
+		_, err := c.Solve(egCtx, def2, SolveOpt{
+			Exports: []ExportEntry{{Type: ExporterLocal, OutputDir: destDir2}},
+		}, nil)
+		return err
+	})
+	err = eg.Wait()
+	require.NoError(t, err)
+
+	// Both builds share the deduplicated vertex, so both outputs come from
+	// the same container. The memory limit should be 128 MiB (most relaxed).
+	for _, dir := range []string{destDir1, destDir2} {
+		dt, err := os.ReadFile(filepath.Join(dir, "mem_limit"))
+		require.NoError(t, err)
+		memLimit := strings.TrimSpace(string(dt))
+		require.Equal(t, "134217728", memLimit,
+			"expected 128 MiB (most relaxed limit) but got %s in %s", memLimit, dir)
+	}
+}
+
 func testNetworkMode(t *testing.T, sb integration.Sandbox) {
 	integration.SkipOnPlatform(t, "windows")
 	c, err := New(sb.Context(), sb.Address())

client/llb/exec.go

@@ -255,6 +255,10 @@ func (e *ExecOp) Marshal(ctx context.Context, c *Constraints) (digest.Digest, []
 		meta.Ulimit = ul
 	}
 
+	if e.constraints.Metadata.LinuxResources != nil {
+		addCap(&e.constraints, pb.CapExecMetaLinuxResources)
+	}
+
 	network, err := getNetwork(e.base)(ctx, c)
 	if err != nil {
 		return "", nil, nil, nil, err

client/llb/exec_test.go

@@ -51,6 +51,97 @@ func TestValidGetMountIndex(t *testing.T) {
 	require.Equal(t, pb.OutputIndex(1), mountIndex, "unexpected mount index")
 }
 
+func TestLinuxResourcesMarshal(t *testing.T) {
+	t.Parallel()
+
+	st := Image("busybox:latest").
+		Run(
+			Shlex("true"),
+			MemoryLimit(64*1024*1024),
+			CPUShares(512),
+			CPUQuota(50000),
+			CPUPeriod(100000),
+			CPUsetCPUs("0-3"),
+			CPUsetMems("0"),
+		).Root()
+
+	def, err := st.Marshal(context.TODO())
+	require.NoError(t, err)
+
+	// Resources should be in OpMetadata (not in the Op bytes / cache key)
+	var found bool
+	for _, md := range def.Metadata {
+		if md.LinuxResources == nil {
+			continue
+		}
+		found = true
+		res := md.LinuxResources
+		require.Equal(t, int64(64*1024*1024), res.Memory)
+		require.Equal(t, uint64(512), res.CpuShares)
+		require.Equal(t, int64(50000), res.CpuQuota)
+		require.Equal(t, uint64(100000), res.CpuPeriod)
+		require.Equal(t, "0-3", res.CpusetCpus)
+		require.Equal(t, "0", res.CpusetMems)
+	}
+	require.True(t, found, "LinuxResources not found in OpMetadata")
+}
+
+func TestLinuxResourcesNotInCacheKey(t *testing.T) {
+	t.Parallel()
+
+	// Two ops with same command but different resource limits must produce the same digest
+	st1 := Image("busybox:latest").
+		Run(Shlex("echo hello"), MemoryLimit(64*1024*1024)).Root()
+
+	st2 := Image("busybox:latest").
+		Run(Shlex("echo hello"), MemoryLimit(128*1024*1024)).Root()
+
+	st3 := Image("busybox:latest").
+		Run(Shlex("echo hello")).Root()
+
+	def1, err := st1.Marshal(context.TODO())
+	require.NoError(t, err)
+
+	def2, err := st2.Marshal(context.TODO())
+	require.NoError(t, err)
+
+	def3, err := st3.Marshal(context.TODO())
+	require.NoError(t, err)
+
+	// All three should produce the same definition bytes (same cache key)
+	require.Equal(t, def1.Def, def2.Def, "different resource limits should produce same digest")
+	require.Equal(t, def1.Def, def3.Def, "resource limits vs no limits should produce same digest")
+}
+
+func TestLinuxResourcesMerge(t *testing.T) {
+	t.Parallel()
+
+	// Test that individual resource limit functions merge correctly
+	st := Image("busybox:latest").
+		Run(
+			Shlex("true"),
+			MemoryLimit(64*1024*1024),
+			CPUShares(512),
+		).Root()
+
+	def, err := st.Marshal(context.TODO())
+	require.NoError(t, err)
+
+	for _, md := range def.Metadata {
+		if md.LinuxResources == nil {
+			continue
+		}
+		res := md.LinuxResources
+		require.Equal(t, int64(64*1024*1024), res.Memory)
+		require.Equal(t, uint64(512), res.CpuShares)
+		// Unset fields should be zero
+		require.Equal(t, int64(0), res.CpuQuota)
+		require.Equal(t, uint64(0), res.CpuPeriod)
+		return
+	}
+	t.Fatal("LinuxResources not found in OpMetadata")
+}
+
 func TestExecOpMarshalConsistency(t *testing.T) {
 	var prevDef [][]byte
 	st := Image("busybox:latest").

client/llb/meta.go

@@ -324,6 +324,17 @@ func getCgroupParent(s State) func(context.Context, *Constraints) (string, error
 	}
 }
 
+// LinuxResources holds CPU and memory resource limits for containers.
+type LinuxResources struct {
+	Memory     int64
+	MemorySwap int64
+	CPUShares  uint64
+	CPUPeriod  uint64
+	CPUQuota   int64
+	CPUsetCPUs string
+	CPUsetMems string
+}
+
 // Network returns a [StateOption] which sets the network mode used for containers created by [State.Run].
 // This is the equivalent of [State.Network]
 // See [State.With] for where to use this.