Cpu resources (#642)

* fn: cpu quota implementation
2022-10-28 21:29:17 +03:00 · 2018-01-12 11:38:28 -08:00
parent fa59400a97
commit 39b2cb2d9b
29 changed files with 856 additions and 91 deletions
--- a/api/agent/agent.go
+++ b/api/agent/agent.go
@@ -338,7 +338,7 @@ func (a *agent) hotLauncher(ctx context.Context, callObj *call) {
 		}).Info("Hot function launcher starting hot container")

 		select {
-		case tok, isOpen := <-a.resources.GetResourceToken(resourceCtx, callObj.Memory, isAsync):
+		case tok, isOpen := <-a.resources.GetResourceToken(resourceCtx, callObj.Memory, uint64(callObj.CPUs), isAsync):
 			cancel()
 			if isOpen {
 				a.wg.Add(1)
@@ -404,7 +404,7 @@ func (a *agent) launchCold(ctx context.Context, call *call) (Slot, error) {
 	ch := make(chan Slot)

 	select {
-	case tok, isOpen := <-a.resources.GetResourceToken(ctx, call.Memory, isAsync):
+	case tok, isOpen := <-a.resources.GetResourceToken(ctx, call.Memory, uint64(call.CPUs), isAsync):
 		if !isOpen {
 			return nil, models.ErrCallTimeoutServerBusy
 		}
@@ -544,6 +544,7 @@ func (a *agent) prepCold(ctx context.Context, call *call, tok ResourceToken, ch
 		image:   call.Image,
 		env:     map[string]string(call.Config),
 		memory:  call.Memory,
+		cpus:    uint64(call.CPUs),
 		timeout: time.Duration(call.Timeout) * time.Second, // this is unnecessary, but in case removal fails...
 		stdin:   call.req.Body,
 		stdout:  call.w,
@@ -604,12 +605,13 @@ func (a *agent) runHot(ctxArg context.Context, call *call, tok ResourceToken) {
 		image:  call.Image,
 		env:    map[string]string(call.Config),
 		memory: call.Memory,
+		cpus:   uint64(call.CPUs),
 		stdin:  stdinRead,
 		stdout: stdoutWrite,
 		stderr: &ghostWriter{inner: stderr},
 	}

-	logger := logrus.WithFields(logrus.Fields{"id": container.id, "app": call.AppName, "route": call.Path, "image": call.Image, "memory": call.Memory, "format": call.Format, "idle_timeout": call.IdleTimeout})
+	logger := logrus.WithFields(logrus.Fields{"id": container.id, "app": call.AppName, "route": call.Path, "image": call.Image, "memory": call.Memory, "cpus": call.CPUs, "format": call.Format, "idle_timeout": call.IdleTimeout})
 	ctx = common.WithLogger(ctx, logger)

 	cookie, err := a.driver.Prepare(ctx, container)
@@ -698,6 +700,7 @@ type container struct {
 	image   string
 	env     map[string]string
 	memory  uint64
+	cpus    uint64
 	timeout time.Duration // cold only (superfluous, but in case)

 	stdin  io.Reader
@@ -735,6 +738,7 @@ func (c *container) Image() string                  { return c.image }
 func (c *container) Timeout() time.Duration         { return c.timeout }
 func (c *container) EnvVars() map[string]string     { return c.env }
 func (c *container) Memory() uint64                 { return c.memory * 1024 * 1024 } // convert MB
+func (c *container) CPUs() uint64                   { return c.cpus }

 // WriteStat publishes each metric in the specified Stats structure as a histogram metric
 func (c *container) WriteStat(ctx context.Context, stat drivers.Stat) {
--- a/api/agent/agent_test.go
+++ b/api/agent/agent_test.go
@@ -185,6 +185,7 @@ func TestCallConfigurationModel(t *testing.T) {
 	const timeout = 1
 	const idleTimeout = 20
 	const memory = 256
+	CPUs := models.MilliCPUs(1000)
 	method := "GET"
 	url := "http://127.0.0.1:8080/r/" + appName + path
 	payload := "payload"
@@ -195,6 +196,7 @@ func TestCallConfigurationModel(t *testing.T) {
 		"FN_APP_NAME": appName,
 		"FN_PATH":     path,
 		"FN_MEMORY":   strconv.Itoa(memory),
+		"FN_CPUS":     CPUs.String(),
 		"FN_TYPE":     typ,
 		"APP_VAR":     "FOO",
 		"ROUTE_VAR":   "BAR",
@@ -210,6 +212,7 @@ func TestCallConfigurationModel(t *testing.T) {
 		Timeout:     timeout,
 		IdleTimeout: idleTimeout,
 		Memory:      memory,
+		CPUs:        CPUs,
 		Payload:     payload,
 		URL:         url,
 		Method:      method,
@@ -243,6 +246,7 @@ func TestAsyncCallHeaders(t *testing.T) {
 	const timeout = 1
 	const idleTimeout = 20
 	const memory = 256
+	CPUs := models.MilliCPUs(200)
 	method := "GET"
 	url := "http://127.0.0.1:8080/r/" + appName + path
 	payload := "payload"
@@ -255,6 +259,7 @@ func TestAsyncCallHeaders(t *testing.T) {
 		"FN_APP_NAME": appName,
 		"FN_PATH":     path,
 		"FN_MEMORY":   strconv.Itoa(memory),
+		"FN_CPUS":     CPUs.String(),
 		"FN_TYPE":     typ,
 		"APP_VAR":     "FOO",
 		"ROUTE_VAR":   "BAR",
@@ -277,6 +282,7 @@ func TestAsyncCallHeaders(t *testing.T) {
 		Timeout:     timeout,
 		IdleTimeout: idleTimeout,
 		Memory:      memory,
+		CPUs:        CPUs,
 		Payload:     payload,
 		URL:         url,
 		Method:      method,
@@ -343,6 +349,7 @@ func TestSubmitError(t *testing.T) {
 	const timeout = 10
 	const idleTimeout = 20
 	const memory = 256
+	CPUs := models.MilliCPUs(200)
 	method := "GET"
 	url := "http://127.0.0.1:8080/r/" + appName + path
 	payload := "payload"
@@ -353,6 +360,7 @@ func TestSubmitError(t *testing.T) {
 		"FN_APP_NAME": appName,
 		"FN_PATH":     path,
 		"FN_MEMORY":   strconv.Itoa(memory),
+		"FN_CPUS":     CPUs.String(),
 		"FN_TYPE":     typ,
 		"APP_VAR":     "FOO",
 		"ROUTE_VAR":   "BAR",
@@ -369,6 +377,7 @@ func TestSubmitError(t *testing.T) {
 		Timeout:     timeout,
 		IdleTimeout: idleTimeout,
 		Memory:      memory,
+		CPUs:        CPUs,
 		Payload:     payload,
 		URL:         url,
 		Method:      method,
--- a/api/agent/call.go
+++ b/api/agent/call.go
@@ -103,6 +103,7 @@ func FromRequest(appName, path string, req *http.Request) CallOpt {
 			Timeout:     route.Timeout,
 			IdleTimeout: route.IdleTimeout,
 			Memory:      route.Memory,
+			CPUs:        route.CPUs,
 			Config:      buildConfig(app, route),
 			Headers:     req.Header,
 			CreatedAt:   strfmt.DateTime(time.Now()),
@@ -130,6 +131,11 @@ func buildConfig(app *models.App, route *models.Route) models.Config {
 	// TODO: might be a good idea to pass in: "FN_BASE_PATH" = fmt.Sprintf("/r/%s", appName) || "/" if using DNS entries per app
 	conf["FN_MEMORY"] = fmt.Sprintf("%d", route.Memory)
 	conf["FN_TYPE"] = route.Type
+
+	CPUs := route.CPUs.String()
+	if CPUs != "" {
+		conf["FN_CPUS"] = CPUs
+	}
 	return conf
 }

--- a/api/agent/drivers/docker/docker.go
+++ b/api/agent/drivers/docker/docker.go
@@ -115,6 +115,15 @@ func (drv *DockerDriver) Prepare(ctx context.Context, task drivers.ContainerTask
 		Context:    ctx,
 	}

+	// Translate milli cpus into CPUQuota & CPUPeriod (see Linux cGroups CFS cgroup v1 documentation)
+	// eg: task.CPUQuota() of 8000 means CPUQuota of 8 * 100000 usecs in 100000 usec period,
+	// which is approx 8 CPUS in CFS world.
+	// Also see docker run options --cpu-quota and --cpu-period
+	if task.CPUs() != 0 {
+		container.HostConfig.CPUQuota = int64(task.CPUs() * 100)
+		container.HostConfig.CPUPeriod = 100000
+	}
+
 	volumes := task.Volumes()
 	for _, mapping := range volumes {
 		hostDir := mapping[0]
@@ -140,7 +149,7 @@ func (drv *DockerDriver) Prepare(ctx context.Context, task drivers.ContainerTask
 		// since we retry under the hood, if the container gets created and retry fails, we can just ignore error
 		if err != docker.ErrContainerAlreadyExists {
 			log.WithFields(logrus.Fields{"call_id": task.Id(), "command": container.Config.Cmd, "memory": container.Config.Memory,
-				"cpu_shares": container.Config.CPUShares, "hostname": container.Config.Hostname, "name": container.Name,
+				"cpu_shares": container.Config.CPUShares, "cpu_quota": task.CPUs(), "hostname": container.Config.Hostname, "name": container.Name,
 				"image": container.Config.Image, "volumes": container.Config.Volumes, "binds": container.HostConfig.Binds, "container": container.Name,
 			}).WithError(err).Error("Could not create container")

--- a/api/agent/drivers/docker/docker_test.go
+++ b/api/agent/drivers/docker/docker_test.go
@@ -31,6 +31,7 @@ func (f *taskDockerTest) Logger() (stdout, stderr io.Writer)      { return f.out
 func (f *taskDockerTest) WriteStat(context.Context, drivers.Stat) { /* TODO */ }
 func (f *taskDockerTest) Volumes() [][2]string                    { return [][2]string{} }
 func (f *taskDockerTest) Memory() uint64                          { return 256 * 1024 * 1024 }
+func (f *taskDockerTest) CPUs() uint64                            { return 0 }
 func (f *taskDockerTest) WorkDir() string                         { return "" }
 func (f *taskDockerTest) Close()                                  {}
 func (f *taskDockerTest) Input() io.Reader                        { return f.input }
--- a/api/agent/drivers/driver.go
+++ b/api/agent/drivers/driver.go
@@ -104,6 +104,9 @@ type ContainerTask interface {
 	// 0 is unlimited.
 	Memory() uint64

+	// CPUs in milli CPU units
+	CPUs() uint64
+
 	// WorkDir returns the working directory to use for the task. Empty string
 	// leaves it unset.
 	WorkDir() string
--- a/api/agent/resource.go
+++ b/api/agent/resource.go
@@ -26,7 +26,7 @@ const (
 type ResourceTracker interface {
 	WaitAsyncResource() chan struct{}
 	// returns a closed channel if the resource can never me met.
-	GetResourceToken(ctx context.Context, memory uint64, isAsync bool) <-chan ResourceToken
+	GetResourceToken(ctx context.Context, memory uint64, cpuQuota uint64, isAsync bool) <-chan ResourceToken
 }

 type resourceTracker struct {
@@ -42,6 +42,17 @@ type resourceTracker struct {
 	ramAsyncUsed uint64
 	// memory in use for async area in which agent stops dequeuing async jobs
 	ramAsyncHWMark uint64
+
+	// cpuTotal is the total usable cpu for sync functions
+	cpuSyncTotal uint64
+	// cpuSyncUsed is cpu reserved for running sync containers including hot/idle
+	cpuSyncUsed uint64
+	// cpuAsyncTotal is the total usable cpu for async + sync functions
+	cpuAsyncTotal uint64
+	// cpuAsyncUsed is cpu reserved for running async + sync containers including hot/idle
+	cpuAsyncUsed uint64
+	// cpu in use for async area in which agent stops dequeuing async jobs
+	cpuAsyncHWMark uint64
 }

 func NewResourceTracker() ResourceTracker {
@@ -51,6 +62,7 @@ func NewResourceTracker() ResourceTracker {
 	}

 	obj.initializeMemory()
+	obj.initializeCPU()
 	return obj
 }

@@ -60,46 +72,52 @@ type ResourceToken interface {
 }

 type resourceToken struct {
+	once      sync.Once
 	decrement func()
 }

 func (t *resourceToken) Close() error {
-	t.decrement()
+	t.once.Do(func() {
+		t.decrement()
+	})
 	return nil
 }

-func (a *resourceTracker) isResourceAvailableLocked(memory uint64, isAsync bool) bool {
+func (a *resourceTracker) isResourceAvailableLocked(memory uint64, cpuQuota uint64, isAsync bool) bool {

-	asyncAvail := a.ramAsyncTotal - a.ramAsyncUsed
-	syncAvail := a.ramSyncTotal - a.ramSyncUsed
+	asyncAvailMem := a.ramAsyncTotal - a.ramAsyncUsed
+	syncAvailMem := a.ramSyncTotal - a.ramSyncUsed
+
+	asyncAvailCPU := a.cpuAsyncTotal - a.cpuAsyncUsed
+	syncAvailCPU := a.cpuSyncTotal - a.cpuSyncUsed

 	// For sync functions, we can steal from async pool. For async, we restrict it to sync pool
 	if isAsync {
-		return asyncAvail >= memory
+		return asyncAvailMem >= memory && asyncAvailCPU >= cpuQuota
 	} else {
-		return asyncAvail+syncAvail >= memory
+		return asyncAvailMem+syncAvailMem >= memory && asyncAvailCPU+syncAvailCPU >= cpuQuota
 	}
 }

 // is this request possible to meet? If no, fail quick
-func (a *resourceTracker) isResourcePossible(memory uint64, isAsync bool) bool {
+func (a *resourceTracker) isResourcePossible(memory uint64, cpuQuota uint64, isAsync bool) bool {
 	if isAsync {
-		return memory <= a.ramAsyncTotal
+		return memory <= a.ramAsyncTotal && cpuQuota <= a.cpuAsyncTotal
 	} else {
-		return memory <= a.ramSyncTotal+a.ramAsyncTotal
+		return memory <= a.ramSyncTotal+a.ramAsyncTotal && cpuQuota <= a.cpuSyncTotal+a.cpuAsyncTotal
 	}
 }

 // the received token should be passed directly to launch (unconditionally), launch
 // will close this token (i.e. the receiver should not call Close)
-func (a *resourceTracker) GetResourceToken(ctx context.Context, memory uint64, isAsync bool) <-chan ResourceToken {
+func (a *resourceTracker) GetResourceToken(ctx context.Context, memory uint64, cpuQuota uint64, isAsync bool) <-chan ResourceToken {

 	memory = memory * Mem1MB

 	c := a.cond
 	ch := make(chan ResourceToken)

-	if !a.isResourcePossible(memory, isAsync) {
+	if !a.isResourcePossible(memory, cpuQuota, isAsync) {
 		close(ch)
 		return ch
 	}
@@ -107,39 +125,44 @@ func (a *resourceTracker) GetResourceToken(ctx context.Context, memory uint64, i
 	go func() {
 		c.L.Lock()

-		for !a.isResourceAvailableLocked(memory, isAsync) {
-			select {
-			case <-ctx.Done():
-				c.L.Unlock()
-				return
-			default:
-			}
-
+		for !a.isResourceAvailableLocked(memory, cpuQuota, isAsync) && ctx.Err() == nil {
 			c.Wait()
 		}

+		if ctx.Err() != nil {
+			c.L.Unlock()
+			return
+		}
+
 		var asyncMem, syncMem uint64
+		var asyncCPU, syncCPU uint64

 		if isAsync {
 			// async uses async pool only
 			asyncMem = memory
-		} else if a.ramSyncTotal-a.ramSyncUsed >= memory {
-			// if sync fits in sync pool
-			syncMem = memory
+			asyncCPU = cpuQuota
 		} else {
 			// if sync fits async + sync pool
-			syncMem = a.ramSyncTotal - a.ramSyncUsed
+			syncMem = minUint64(a.ramSyncTotal-a.ramSyncUsed, memory)
+			syncCPU = minUint64(a.cpuSyncTotal-a.cpuSyncUsed, cpuQuota)
+
 			asyncMem = memory - syncMem
+			asyncCPU = cpuQuota - syncCPU
 		}

 		a.ramAsyncUsed += asyncMem
 		a.ramSyncUsed += syncMem
+		a.cpuAsyncUsed += asyncCPU
+		a.cpuSyncUsed += syncCPU
 		c.L.Unlock()

 		t := &resourceToken{decrement: func() {
+
 			c.L.Lock()
 			a.ramAsyncUsed -= asyncMem
 			a.ramSyncUsed -= syncMem
+			a.cpuAsyncUsed -= asyncCPU
+			a.cpuSyncUsed -= syncCPU
 			c.L.Unlock()

 			// WARNING: yes, we wake up everyone even async waiters when only sync pool has space, but
@@ -159,7 +182,7 @@ func (a *resourceTracker) GetResourceToken(ctx context.Context, memory uint64, i
 	return ch
 }

-// WaitAsyncResource will send a signal on the returned channel when RAM in-use
+// WaitAsyncResource will send a signal on the returned channel when RAM and CPU in-use
 // in the async area is less than high water mark
 func (a *resourceTracker) WaitAsyncResource() chan struct{} {
 	ch := make(chan struct{})
@@ -167,7 +190,7 @@ func (a *resourceTracker) WaitAsyncResource() chan struct{} {
 	c := a.cond
 	go func() {
 		c.L.Lock()
-		for a.ramAsyncUsed >= a.ramAsyncHWMark {
+		for a.ramAsyncUsed >= a.ramAsyncHWMark || a.cpuAsyncUsed >= a.cpuAsyncHWMark {
 			c.Wait()
 		}
 		c.L.Unlock()
@@ -198,6 +221,71 @@ func clampUint64(val, min, max uint64) uint64 {
 	return val
 }

+func (a *resourceTracker) initializeCPU() {
+
+	var maxSyncCPU, maxAsyncCPU, cpuAsyncHWMark uint64
+	var totalCPU, availCPU uint64
+
+	if runtime.GOOS == "linux" {
+
+		// Why do we prefer /proc/cpuinfo for Linux and not just use runtime.NumCPU?
+		// This is because NumCPU is sched_getaffinity based and we prefer to check
+		// cgroup which will more likely be same cgroup for container runtime
+		numCPU, err := checkProcCPU()
+		if err != nil {
+			logrus.WithError(err).Error("Error checking for CPU, falling back to runtime CPU count.")
+			numCPU = uint64(runtime.NumCPU())
+		}
+
+		totalCPU = 1000 * numCPU
+		availCPU = totalCPU
+
+		// Clamp further if cgroups CFS quota/period limits are in place
+		cgroupCPU := checkCgroupCPU()
+		if cgroupCPU > 0 {
+			availCPU = minUint64(availCPU, cgroupCPU)
+		}
+
+		// TODO: check cgroup cpuset to clamp this further. We might be restricted into
+		// a subset of CPUs. (eg. /sys/fs/cgroup/cpuset/cpuset.effective_cpus)
+
+		// TODO: skip CPU headroom for ourselves for now
+	} else {
+		totalCPU = uint64(runtime.NumCPU() * 1000)
+		availCPU = totalCPU
+	}
+
+	logrus.WithFields(logrus.Fields{
+		"totalCPU": totalCPU,
+		"availCPU": availCPU,
+	}).Info("available cpu")
+
+	// %20 of cpu for sync only reserve
+	maxSyncCPU = uint64(availCPU * 2 / 10)
+	maxAsyncCPU = availCPU - maxSyncCPU
+	cpuAsyncHWMark = maxAsyncCPU * 8 / 10
+
+	logrus.WithFields(logrus.Fields{
+		"cpuSync":        maxSyncCPU,
+		"cpuAsync":       maxAsyncCPU,
+		"cpuAsyncHWMark": cpuAsyncHWMark,
+	}).Info("sync and async cpu reservations")
+
+	if maxSyncCPU == 0 || maxAsyncCPU == 0 {
+		logrus.Fatal("Cannot get the proper CPU information to size server")
+	}
+
+	if maxSyncCPU+maxAsyncCPU < 1000 {
+		logrus.Warn("Severaly Limited CPU: cpuSync + cpuAsync < 1000m (1 CPU)")
+	} else if maxAsyncCPU < 1000 {
+		logrus.Warn("Severaly Limited CPU: cpuAsync < 1000m (1 CPU)")
+	}
+
+	a.cpuAsyncHWMark = cpuAsyncHWMark
+	a.cpuSyncTotal = maxSyncCPU
+	a.cpuAsyncTotal = maxAsyncCPU
+}
+
 func (a *resourceTracker) initializeMemory() {

 	var maxSyncMemory, maxAsyncMemory, ramAsyncHWMark uint64
@@ -205,7 +293,7 @@ func (a *resourceTracker) initializeMemory() {
 	if runtime.GOOS == "linux" {

 		// system wide available memory
-		totalMemory, err := checkProc()
+		totalMemory, err := checkProcMem()
 		if err != nil {
 			logrus.WithError(err).Fatal("Cannot get the proper memory information to size server.")
 		}
@@ -213,7 +301,7 @@ func (a *resourceTracker) initializeMemory() {
 		availMemory := totalMemory

 		// cgroup limit restriction on memory usage
-		cGroupLimit, err := checkCgroup()
+		cGroupLimit, err := checkCgroupMem()
 		if err != nil {
 			logrus.WithError(err).Error("Error checking for cgroup memory limits, falling back to host memory available..")
 		} else {
@@ -251,7 +339,7 @@ func (a *resourceTracker) initializeMemory() {
 		"ramSync":        maxSyncMemory,
 		"ramAsync":       maxAsyncMemory,
 		"ramAsyncHWMark": ramAsyncHWMark,
-	}).Info("sync and async reservations")
+	}).Info("sync and async ram reservations")

 	if maxSyncMemory == 0 || maxAsyncMemory == 0 {
 		logrus.Fatal("Cannot get the proper memory pool information to size server")
@@ -286,24 +374,56 @@ func getMemoryHeadRoom(usableMemory uint64) (uint64, error) {
 	return headRoom, nil
 }

-func checkCgroup() (uint64, error) {
-	f, err := os.Open("/sys/fs/cgroup/memory/memory.limit_in_bytes")
+func readString(fileName string) (string, error) {
+	b, err := ioutil.ReadFile(fileName)
+	if err != nil {
+		return "", err
+	}
+	value := string(b)
+	return strings.TrimSpace(value), nil
+}
+
+func checkCgroupMem() (uint64, error) {
+	value, err := readString("/sys/fs/cgroup/memory/memory.limit_in_bytes")
 	if err != nil {
 		return 0, err
 	}
-	defer f.Close()
-	b, err := ioutil.ReadAll(f)
+	return strconv.ParseUint(value, 10, 64)
+}
+
+func checkCgroupCPU() uint64 {
+
+	periodStr, err := readString("/sys/fs/cgroup/cpu/cpu.cfs_period_us")
 	if err != nil {
-		return 0, err
+		return 0
 	}
-	limBytes := string(b)
-	limBytes = strings.TrimSpace(limBytes)
-	return strconv.ParseUint(limBytes, 10, 64)
+	quotaStr, err := readString("/sys/fs/cgroup/cpu/cpu.cfs_quota_us")
+	if err != nil {
+		return 0
+	}
+
+	period, err := strconv.ParseUint(periodStr, 10, 64)
+	if err != nil {
+		logrus.Warn("Cannot parse CFS period", err)
+		return 0
+	}
+
+	quota, err := strconv.ParseInt(quotaStr, 10, 64)
+	if err != nil {
+		logrus.Warn("Cannot parse CFS quota", err)
+		return 0
+	}
+
+	if quota <= 0 || period <= 0 {
+		return 0
+	}
+
+	return uint64(quota) * 1000 / period
 }

 var errCantReadMemInfo = errors.New("Didn't find MemAvailable in /proc/meminfo, kernel is probably < 3.14")

-func checkProc() (uint64, error) {
+func checkProcMem() (uint64, error) {
 	f, err := os.Open("/proc/meminfo")
 	if err != nil {
 		return 0, err
@@ -341,3 +461,30 @@ func checkProc() (uint64, error) {

 	return 0, errCantReadMemInfo
 }
+
+func checkProcCPU() (uint64, error) {
+	f, err := os.Open("/proc/cpuinfo")
+	if err != nil {
+		return 0, err
+	}
+	defer f.Close()
+
+	total := uint64(0)
+
+	scanner := bufio.NewScanner(f)
+	for scanner.Scan() {
+		b := scanner.Text()
+
+		// processor       : 0
+		toks := strings.Fields(b)
+		if len(toks) == 3 && toks[0] == "processor" && toks[1] == ":" {
+			total += 1
+		}
+	}
+
+	if total == 0 {
+		return 0, errors.New("Could not parse cpuinfo")
+	}
+
+	return total, nil
+}
--- a/api/agent/resource_test.go
+++ b/api/agent/resource_test.go
@@ -1,6 +1,8 @@
 package agent

 import (
+	"context"
+	"errors"
 	"testing"
 	"time"
 )
@@ -8,11 +10,17 @@ import (
 func setTrackerTestVals(tr *resourceTracker, vals *trackerVals) {
 	tr.cond.L.Lock()

-	tr.ramSyncTotal = vals.st
-	tr.ramSyncUsed = vals.su
-	tr.ramAsyncTotal = vals.at
-	tr.ramAsyncUsed = vals.au
-	tr.ramAsyncHWMark = vals.am
+	tr.ramSyncTotal = vals.mst
+	tr.ramSyncUsed = vals.msu
+	tr.ramAsyncTotal = vals.mat
+	tr.ramAsyncUsed = vals.mau
+	tr.ramAsyncHWMark = vals.mam
+
+	tr.cpuSyncTotal = vals.cst
+	tr.cpuSyncUsed = vals.csu
+	tr.cpuAsyncTotal = vals.cat
+	tr.cpuAsyncUsed = vals.cau
+	tr.cpuAsyncHWMark = vals.cam

 	tr.cond.L.Unlock()
 	tr.cond.Broadcast()
@@ -22,24 +30,72 @@ func getTrackerTestVals(tr *resourceTracker, vals *trackerVals) {

 	tr.cond.L.Lock()

-	vals.st = tr.ramSyncTotal
-	vals.su = tr.ramSyncUsed
-	vals.at = tr.ramAsyncTotal
-	vals.au = tr.ramAsyncUsed
-	vals.am = tr.ramAsyncHWMark
+	vals.mst = tr.ramSyncTotal
+	vals.msu = tr.ramSyncUsed
+	vals.mat = tr.ramAsyncTotal
+	vals.mau = tr.ramAsyncUsed
+	vals.mam = tr.ramAsyncHWMark
+
+	vals.cst = tr.cpuSyncTotal
+	vals.csu = tr.cpuSyncUsed
+	vals.cat = tr.cpuAsyncTotal
+	vals.cau = tr.cpuAsyncUsed
+	vals.cam = tr.cpuAsyncHWMark

 	tr.cond.L.Unlock()
 }

+// helper to debug print (fields correspond to resourceTracker CPU/MEM fields)
 type trackerVals struct {
-	st uint64
-	su uint64
-	at uint64
-	au uint64
-	am uint64
+	mst uint64
+	msu uint64
+	mat uint64
+	mau uint64
+	mam uint64
+	cst uint64
+	csu uint64
+	cat uint64
+	cau uint64
+	cam uint64
 }

-func TestResourceAsyncMem(t *testing.T) {
+func (vals *trackerVals) setDefaults() {
+	// set set these to known vals (4GB total: 1GB sync, 3 async)
+	vals.mst = 1 * Mem1GB
+	vals.msu = 0
+	vals.mat = 3 * Mem1GB
+	vals.mau = 0
+	vals.mam = 1 * Mem1GB
+
+	// let's assume 10 CPUs (2 CPU sync, 8 CPU async)
+	vals.cst = 2000
+	vals.csu = 0
+	vals.cat = 8000
+	vals.cau = 0
+	vals.cam = 6000
+}
+
+func fetchToken(ch <-chan ResourceToken) (ResourceToken, error) {
+	select {
+	case tok := <-ch:
+		return tok, nil
+	case <-time.After(time.Duration(500) * time.Millisecond):
+		return nil, errors.New("expected token")
+	}
+}
+
+func isClosed(ch <-chan ResourceToken) bool {
+	select {
+	case _, ok := <-ch:
+		if !ok {
+			return true
+		}
+	default:
+	}
+	return false
+}
+
+func TestResourceAsyncWait(t *testing.T) {

 	var vals trackerVals

@@ -48,36 +104,242 @@ func TestResourceAsyncMem(t *testing.T) {
 	tr := trI.(*resourceTracker)

 	getTrackerTestVals(tr, &vals)
-	if vals.st <= 0 || vals.su != 0 || vals.at <= 0 || vals.au != 0 || vals.am <= 0 {
-		t.Fatalf("faulty init %#v", vals)
+	if vals.mst <= 0 || vals.msu != 0 || vals.mat <= 0 || vals.mau != 0 || vals.mam <= 0 {
+		t.Fatalf("faulty init MEM %#v", vals)
+	}
+	if vals.cst <= 0 || vals.csu != 0 || vals.cat <= 0 || vals.cau != 0 || vals.cam <= 0 {
+		t.Fatalf("faulty init CPU %#v", vals)
 	}

-	// set set these to known vals
-	vals.st = 1 * 1024 * 1024
-	vals.su = 0
-	vals.at = 2 * 1024 * 1024
-	vals.au = 0
-	vals.am = 1 * 1024 * 1024
+	vals.setDefaults()

 	// should block & wait
-	vals.au = vals.am
+	vals.mau = vals.mam
 	setTrackerTestVals(tr, &vals)
 	ch := tr.WaitAsyncResource()

 	select {
 	case <-ch:
-		t.Fatal("high water mark over, should not trigger")
+		t.Fatal("high water mark MEM over, should not trigger")
 	case <-time.After(time.Duration(500) * time.Millisecond):
 	}

 	// should not block & wait
-	vals.au = 0
+	vals.mau = 0
 	setTrackerTestVals(tr, &vals)

 	select {
 	case <-ch:
 	case <-time.After(time.Duration(500) * time.Millisecond):
-		t.Fatal("high water mark not over, should trigger")
+		t.Fatal("high water mark MEM not over, should trigger")
+	}
+
+	// get a new channel to prevent previous test interference
+	ch = tr.WaitAsyncResource()
+
+	// should block & wait
+	vals.cau = vals.cam
+	setTrackerTestVals(tr, &vals)
+
+	select {
+	case <-ch:
+		t.Fatal("high water mark CPU over, should not trigger")
+	case <-time.After(time.Duration(500) * time.Millisecond):
+	}
+
+	// should not block & wait
+	vals.cau = 0
+	setTrackerTestVals(tr, &vals)
+
+	select {
+	case <-ch:
+	case <-time.After(time.Duration(500) * time.Millisecond):
+		t.Fatal("high water mark CPU not over, should trigger")
+	}
+
+}
+
+func TestResourceGetSimple(t *testing.T) {
+
+	var vals trackerVals
+	trI := NewResourceTracker()
+	tr := trI.(*resourceTracker)
+
+	vals.setDefaults()
+
+	// let's make it like CPU and MEM are 100% full
+	vals.mau = vals.mat
+	vals.cau = vals.cat
+
+	setTrackerTestVals(tr, &vals)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	// ask for 4GB and 10 CPU
+	ch := trI.GetResourceToken(ctx, 4*1024, 1000, false)
+
+	_, err := fetchToken(ch)
+	if err == nil {
+		t.Fatalf("full system should not hand out token")
+	}
+
+	// reset back
+	vals.setDefaults()
+	setTrackerTestVals(tr, &vals)
+
+	tok, err := fetchToken(ch)
+	if err != nil {
+		t.Fatalf("empty system should hand out token")
+	}
+
+	// ask for another 4GB and 10 CPU
+	ch = trI.GetResourceToken(ctx, 4*1024, 1000, false)
+
+	_, err = fetchToken(ch)
+	if err == nil {
+		t.Fatalf("full system should not hand out token")
+	}
+
+	// close means, giant token resources released
+	tok.Close()
+
+	tok, err = fetchToken(ch)
+	if err != nil {
+		t.Fatalf("empty system should hand out token")
+	}
+
+	tok.Close()
+
+	// POOLS should all be empty now
+	getTrackerTestVals(tr, &vals)
+	if vals.msu != 0 || vals.mau != 0 {
+		t.Fatalf("faulty state MEM %#v", vals)
+	}
+	if vals.csu != 0 || vals.cau != 0 {
+		t.Fatalf("faulty state CPU %#v", vals)
+	}
+}
+
+func TestResourceGetCombo(t *testing.T) {
+
+	var vals trackerVals
+	trI := NewResourceTracker()
+	tr := trI.(*resourceTracker)
+
+	vals.setDefaults()
+	setTrackerTestVals(tr, &vals)
+
+	ctx, cancel := context.WithCancel(context.Background())
+
+	// impossible request
+	ch := trI.GetResourceToken(ctx, 20*1024, 20000, false)
+	if !isClosed(ch) {
+		t.Fatalf("impossible request should return closed channel")
+	}
+
+	cancel()
+	ctx, cancel = context.WithCancel(context.Background())
+
+	// let's use up 2 GB of 3GB async pool
+	ch = trI.GetResourceToken(ctx, 2*1024, 10, true)
+	tok1, err := fetchToken(ch)
+	if err != nil {
+		t.Fatalf("empty async system should hand out token1")
+	}
+
+	cancel()
+	ctx, cancel = context.WithCancel(context.Background())
+
+	// remaining 1 GB async
+	ch = trI.GetResourceToken(ctx, 1*1024, 11, true)
+	tok2, err := fetchToken(ch)
+	if err != nil {
+		t.Fatalf("empty async system should hand out token2")
+	}
+
+	cancel()
+	ctx, cancel = context.WithCancel(context.Background())
+
+	// NOW ASYNC POOL IS FULL
+	// SYNC POOL HAS 1GB
+
+	// we no longer can get async token
+	ch = trI.GetResourceToken(ctx, 1*1024, 12, true)
+	_, err = fetchToken(ch)
+	if err == nil {
+		t.Fatalf("full async system should not hand out a token")
+	}
+
+	cancel()
+	ctx, cancel = context.WithCancel(context.Background())
+
+	// but we should get 1GB sync token
+	ch = trI.GetResourceToken(ctx, 1*1024, 13, false)
+	tok3, err := fetchToken(ch)
+	if err != nil {
+		t.Fatalf("empty sync system should hand out token3")
+	}
+	cancel()
+	ctx, cancel = context.WithCancel(context.Background())
+
+	// NOW ASYNC AND SYNC POOLS ARE FULL
+
+	// this should fail
+	ch = trI.GetResourceToken(ctx, 1*1024, 14, false)
+	_, err = fetchToken(ch)
+	if err == nil {
+		t.Fatalf("full system should not hand out a token")
+	}
+	cancel()
+	ctx, cancel = context.WithCancel(context.Background())
+
+	// now let's free up some async pool, release tok2 (1GB)
+	tok2.Close()
+
+	// NOW ASYNC POOL HAS 1GB FREE
+	// SYNC POOL IS FULL
+
+	// async pool should provide this
+	ch = trI.GetResourceToken(ctx, 1*1024, 15, false)
+	tok4, err := fetchToken(ch)
+	if err != nil {
+		t.Fatalf("async system should hand out token4")
+	}
+	cancel()
+	ctx, cancel = context.WithCancel(context.Background())
+
+	// NOW ASYNC AND SYNC POOLS ARE FULL
+
+	tok4.Close()
+	tok3.Close()
+
+	// NOW ASYNC POOL HAS 1GB FREE
+	// SYNC POOL HAS 1GB FREE
+
+	// now, we ask for 2GB sync token, it should be provided from both async+sync pools
+	ch = trI.GetResourceToken(ctx, 2*1024, 16, false)
+	tok5, err := fetchToken(ch)
+	if err != nil {
+		t.Fatalf("async+sync system should hand out token5")
+	}
+	cancel()
+
+	// NOW ASYNC AND SYNC POOLS ARE FULL
+
+	tok1.Close()
+	tok5.Close()
+
+	// attempt to close tok2 twice.. This should be OK.
+	tok2.Close()
+
+	// POOLS should all be empty now
+	getTrackerTestVals(tr, &vals)
+	if vals.msu != 0 || vals.mau != 0 {
+		t.Fatalf("faulty state MEM %#v", vals)
+	}
+	if vals.csu != 0 || vals.cau != 0 {
+		t.Fatalf("faulty state CPU %#v", vals)
 	}

 }
--- a/api/agent/slots.go
+++ b/api/agent/slots.go
@@ -334,6 +334,7 @@ func getSlotQueueKey(call *call) string {
 	fmt.Fprint(hash, call.Timeout, "\x00")
 	fmt.Fprint(hash, call.IdleTimeout, "\x00")
 	fmt.Fprint(hash, call.Memory, "\x00")
+	fmt.Fprint(hash, call.CPUs, "\x00")
 	fmt.Fprint(hash, call.Format, "\x00")

 	// we have to sort these before printing, yay. TODO do better
--- a/api/datastore/internal/datastoretest/test.go
+++ b/api/datastore/internal/datastoretest/test.go
@@ -529,6 +529,7 @@ func Test(t *testing.T, dsf func(t *testing.T) models.Datastore) {
 				Format:      "http",
 				IdleTimeout: testRoute.IdleTimeout,
 				Memory:      testRoute.Memory,
+				CPUs:        testRoute.CPUs,
 				// updated
 				Timeout: 2,
 				Config: map[string]string{
@@ -572,6 +573,7 @@ func Test(t *testing.T, dsf func(t *testing.T) models.Datastore) {
 				Format:      "http",
 				Timeout:     2,
 				Memory:      testRoute.Memory,
+				CPUs:        testRoute.CPUs,
 				IdleTimeout: testRoute.IdleTimeout,
 				// updated
 				Config: map[string]string{
--- a/api/datastore/sql/migrations/7_add_route_cpus.down.sql
+++ b/api/datastore/sql/migrations/7_add_route_cpus.down.sql
@@ -0,0 +1 @@
+ALTER TABLE routes DROP COLUMN cpus;
--- a/api/datastore/sql/migrations/7_add_route_cpus.up.sql
+++ b/api/datastore/sql/migrations/7_add_route_cpus.up.sql
@@ -0,0 +1 @@
+ALTER TABLE routes ADD cpus int;
--- a/api/datastore/sql/migrations/README.md
+++ b/api/datastore/sql/migrations/README.md
@@ -22,7 +22,7 @@ README:
 $ go generate
 ```

-NOTE: You may need to `go get github.com/jteeuwen/go-bindata` before running `go
+NOTE: You may need to `go get -u github.com/jteeuwen/go-bindata/...` before running `go
 generate` in order for it to work.

 After running `go generate`, the `migrations.go` file should be updated. Check
--- a/api/datastore/sql/migrations/migrations.go
+++ b/api/datastore/sql/migrations/migrations.go
@@ -12,6 +12,8 @@
 // 5_add_app_created_at.up.sql
 // 6_add_app_updated_at.down.sql
 // 6_add_app_updated_at.up.sql
+// 7_add_route_cpus.down.sql
+// 7_add_route_cpus.up.sql
 // DO NOT EDIT!

 package migrations
@@ -94,7 +96,7 @@ func _1_add_route_created_atDownSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "1_add_route_created_at.down.sql", size: 43, mode: os.FileMode(420), modTime: time.Unix(1511259011, 0)}
+	info := bindataFileInfo{name: "1_add_route_created_at.down.sql", size: 43, mode: os.FileMode(420), modTime: time.Unix(1510963763, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -114,7 +116,7 @@ func _1_add_route_created_atUpSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "1_add_route_created_at.up.sql", size: 40, mode: os.FileMode(420), modTime: time.Unix(1511919777, 0)}
+	info := bindataFileInfo{name: "1_add_route_created_at.up.sql", size: 40, mode: os.FileMode(420), modTime: time.Unix(1510963763, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -134,7 +136,7 @@ func _2_add_call_statsDownSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "2_add_call_stats.down.sql", size: 37, mode: os.FileMode(420), modTime: time.Unix(1511259011, 0)}
+	info := bindataFileInfo{name: "2_add_call_stats.down.sql", size: 37, mode: os.FileMode(420), modTime: time.Unix(1511917353, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -154,7 +156,7 @@ func _2_add_call_statsUpSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "2_add_call_stats.up.sql", size: 34, mode: os.FileMode(420), modTime: time.Unix(1511259011, 0)}
+	info := bindataFileInfo{name: "2_add_call_stats.up.sql", size: 34, mode: os.FileMode(420), modTime: time.Unix(1511917353, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -174,7 +176,7 @@ func _3_add_call_errorDownSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "3_add_call_error.down.sql", size: 37, mode: os.FileMode(420), modTime: time.Unix(1511301534, 0)}
+	info := bindataFileInfo{name: "3_add_call_error.down.sql", size: 37, mode: os.FileMode(420), modTime: time.Unix(1511989827, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -194,7 +196,7 @@ func _3_add_call_errorUpSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "3_add_call_error.up.sql", size: 34, mode: os.FileMode(420), modTime: time.Unix(1511301534, 0)}
+	info := bindataFileInfo{name: "3_add_call_error.up.sql", size: 34, mode: os.FileMode(420), modTime: time.Unix(1511989827, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -214,7 +216,7 @@ func _4_add_route_updated_atDownSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "4_add_route_updated_at.down.sql", size: 43, mode: os.FileMode(420), modTime: time.Unix(1513728957, 0)}
+	info := bindataFileInfo{name: "4_add_route_updated_at.down.sql", size: 43, mode: os.FileMode(420), modTime: time.Unix(1514060619, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -234,7 +236,7 @@ func _4_add_route_updated_atUpSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "4_add_route_updated_at.up.sql", size: 48, mode: os.FileMode(420), modTime: time.Unix(1513730369, 0)}
+	info := bindataFileInfo{name: "4_add_route_updated_at.up.sql", size: 48, mode: os.FileMode(420), modTime: time.Unix(1514060619, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -254,7 +256,7 @@ func _5_add_app_created_atDownSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "5_add_app_created_at.down.sql", size: 41, mode: os.FileMode(420), modTime: time.Unix(1513730497, 0)}
+	info := bindataFileInfo{name: "5_add_app_created_at.down.sql", size: 41, mode: os.FileMode(420), modTime: time.Unix(1514060619, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -274,7 +276,7 @@ func _5_add_app_created_atUpSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "5_add_app_created_at.up.sql", size: 46, mode: os.FileMode(420), modTime: time.Unix(1513730527, 0)}
+	info := bindataFileInfo{name: "5_add_app_created_at.up.sql", size: 46, mode: os.FileMode(420), modTime: time.Unix(1514060619, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -294,7 +296,7 @@ func _6_add_app_updated_atDownSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "6_add_app_updated_at.down.sql", size: 41, mode: os.FileMode(420), modTime: time.Unix(1513733616, 0)}
+	info := bindataFileInfo{name: "6_add_app_updated_at.down.sql", size: 41, mode: os.FileMode(420), modTime: time.Unix(1514060619, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -314,7 +316,47 @@ func _6_add_app_updated_atUpSql() (*asset, error) {
 		return nil, err
 	}

-	info := bindataFileInfo{name: "6_add_app_updated_at.up.sql", size: 46, mode: os.FileMode(420), modTime: time.Unix(1513733621, 0)}
+	info := bindataFileInfo{name: "6_add_app_updated_at.up.sql", size: 46, mode: os.FileMode(420), modTime: time.Unix(1514060619, 0)}
+	a := &asset{bytes: bytes, info: info}
+	return a, nil
+}
+
+var __7_add_route_cpusDownSql = []byte("\x1f\x8b\x08\x00\x00\x00\x00\x00\x00\xff\x72\xf4\x09\x71\x0d\x52\x08\x71\x74\xf2\x71\x55\x28\xca\x2f\x2d\x49\x2d\x56\x70\x09\xf2\x0f\x50\x70\xf6\xf7\x09\xf5\xf5\x53\x48\x2e\x28\x2d\xb6\xe6\x02\x04\x00\x00\xff\xff\xec\x60\x24\xd0\x25\x00\x00\x00")
+
+func _7_add_route_cpusDownSqlBytes() ([]byte, error) {
+	return bindataRead(
+		__7_add_route_cpusDownSql,
+		"7_add_route_cpus.down.sql",
+	)
+}
+
+func _7_add_route_cpusDownSql() (*asset, error) {
+	bytes, err := _7_add_route_cpusDownSqlBytes()
+	if err != nil {
+		return nil, err
+	}
+
+	info := bindataFileInfo{name: "7_add_route_cpus.down.sql", size: 37, mode: os.FileMode(420), modTime: time.Unix(1515624756, 0)}
+	a := &asset{bytes: bytes, info: info}
+	return a, nil
+}
+
+var __7_add_route_cpusUpSql = []byte("\x1f\x8b\x08\x00\x00\x00\x00\x00\x00\xff\x72\xf4\x09\x71\x0d\x52\x08\x71\x74\xf2\x71\x55\x28\xca\x2f\x2d\x49\x2d\x56\x70\x74\x71\x51\x48\x2e\x28\x2d\x56\xc8\xcc\x2b\xb1\xe6\x02\x04\x00\x00\xff\xff\xf1\x18\xf8\xa9\x21\x00\x00\x00")
+
+func _7_add_route_cpusUpSqlBytes() ([]byte, error) {
+	return bindataRead(
+		__7_add_route_cpusUpSql,
+		"7_add_route_cpus.up.sql",
+	)
+}
+
+func _7_add_route_cpusUpSql() (*asset, error) {
+	bytes, err := _7_add_route_cpusUpSqlBytes()
+	if err != nil {
+		return nil, err
+	}
+
+	info := bindataFileInfo{name: "7_add_route_cpus.up.sql", size: 33, mode: os.FileMode(420), modTime: time.Unix(1515628068, 0)}
 	a := &asset{bytes: bytes, info: info}
 	return a, nil
 }
@@ -383,6 +425,8 @@ var _bindata = map[string]func() (*asset, error){
 	"5_add_app_created_at.up.sql":     _5_add_app_created_atUpSql,
 	"6_add_app_updated_at.down.sql":   _6_add_app_updated_atDownSql,
 	"6_add_app_updated_at.up.sql":     _6_add_app_updated_atUpSql,
+	"7_add_route_cpus.down.sql":       _7_add_route_cpusDownSql,
+	"7_add_route_cpus.up.sql":         _7_add_route_cpusUpSql,
 }

 // AssetDir returns the file names below a certain
@@ -438,6 +482,8 @@ var _bintree = &bintree{nil, map[string]*bintree{
 	"5_add_app_created_at.up.sql":     &bintree{_5_add_app_created_atUpSql, map[string]*bintree{}},
 	"6_add_app_updated_at.down.sql":   &bintree{_6_add_app_updated_atDownSql, map[string]*bintree{}},
 	"6_add_app_updated_at.up.sql":     &bintree{_6_add_app_updated_atUpSql, map[string]*bintree{}},
+	"7_add_route_cpus.down.sql":       &bintree{_7_add_route_cpusDownSql, map[string]*bintree{}},
+	"7_add_route_cpus.up.sql":         &bintree{_7_add_route_cpusUpSql, map[string]*bintree{}},
 }}

 // RestoreAsset restores an asset under the given directory
--- a/api/datastore/sql/sql.go
+++ b/api/datastore/sql/sql.go
@@ -49,6 +49,7 @@ var tables = [...]string{`CREATE TABLE IF NOT EXISTS routes (
 	image varchar(256) NOT NULL,
 	format varchar(16) NOT NULL,
 	memory int NOT NULL,
+	cpus int,
 	timeout int NOT NULL,
 	idle_timeout int NOT NULL,
 	type varchar(16) NOT NULL,
@@ -87,7 +88,7 @@ var tables = [...]string{`CREATE TABLE IF NOT EXISTS routes (
 }

 const (
-	routeSelector = `SELECT app_name, path, image, format, memory, type, timeout, idle_timeout, headers, config, created_at, updated_at FROM routes`
+	routeSelector = `SELECT app_name, path, image, format, memory, cpus, type, timeout, idle_timeout, headers, config, created_at, updated_at FROM routes`
 	callSelector  = `SELECT id, created_at, started_at, completed_at, status, app_name, path, stats, error FROM calls`
 )

@@ -453,6 +454,7 @@ func (ds *sqlStore) InsertRoute(ctx context.Context, route *models.Route) (*mode
 			image,
 			format,
 			memory,
+			cpus,
 			type,
 			timeout,
 			idle_timeout,
@@ -467,6 +469,7 @@ func (ds *sqlStore) InsertRoute(ctx context.Context, route *models.Route) (*mode
 			:image,
 			:format,
 			:memory,
+			:cpus,
 			:type,
 			:timeout,
 			:idle_timeout,
@@ -507,6 +510,7 @@ func (ds *sqlStore) UpdateRoute(ctx context.Context, newroute *models.Route) (*m
 			image = :image,
 			format = :format,
 			memory = :memory,
+			cpus = :cpus,
 			type = :type,
 			timeout = :timeout,
 			idle_timeout = :idle_timeout,
--- a/api/models/call.go
+++ b/api/models/call.go
@@ -118,6 +118,11 @@ type Call struct {
 	// Memory is the amount of RAM this call is allocated.
 	Memory uint64 `json:"memory,omitempty" db:"-"`

+	// CPU as in MilliCPUs where each CPU core is split into 1000 units, specified either
+	// *) milliCPUs as "100m" which is 1/10 of a CPU or
+	// *) as floating point number "0.1" which is 1/10 of a CPU
+	CPUs MilliCPUs `json:"cpus,omitempty" db:"-"`
+
 	// Config is the set of configuration variables for the call
 	Config Config `json:"config,omitempty" db:"-"`

--- a/api/models/config.go
+++ b/api/models/config.go
@@ -6,6 +6,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"net/http"
+	"strconv"
 )

 type Config map[string]string
@@ -125,3 +126,67 @@ func (h *Headers) Scan(value interface{}) error {
 	// otherwise, return an error
 	return fmt.Errorf("headers invalid db format: %T %T value, err: %v", value, bv, err)
 }
+
+// MilliCPU units
+type MilliCPUs uint64
+
+const (
+	MinMilliCPUs = 0       // 0 is unlimited
+	MaxMilliCPUs = 1024000 // 1024 CPUs
+)
+
+// implements fmt.Stringer
+func (c MilliCPUs) String() string {
+	if c == 0 {
+		return ""
+	}
+	return fmt.Sprintf("%dm", c)
+}
+
+// implements json.Unmarshaler
+func (c *MilliCPUs) UnmarshalJSON(data []byte) error {
+
+	outer := bytes.TrimSpace(data)
+	if !bytes.HasSuffix(outer, []byte("\"")) || !bytes.HasPrefix(outer, []byte("\"")) {
+		return ErrInvalidJSON
+	}
+
+	outer = bytes.TrimPrefix(outer, []byte("\""))
+	outer = bytes.TrimSuffix(outer, []byte("\""))
+	outer = bytes.TrimSpace(outer)
+	if len(outer) == 0 {
+		*c = 0
+		return nil
+	}
+
+	if bytes.HasSuffix(outer, []byte("m")) {
+
+		// Support milli cores as "100m"
+		outer = bytes.TrimSuffix(outer, []byte("m"))
+		mCPU, err := strconv.ParseUint(string(outer), 10, 64)
+		if err != nil || mCPU > MaxMilliCPUs || mCPU < MinMilliCPUs {
+			return ErrInvalidCPUs
+		}
+		*c = MilliCPUs(mCPU)
+	} else {
+		// Support for floating point "0.1" style CPU units
+		fCPU, err := strconv.ParseFloat(string(outer), 64)
+		if err != nil || fCPU < MinMilliCPUs/1000 || fCPU > MaxMilliCPUs/1000 {
+			return ErrInvalidCPUs
+		}
+		*c = MilliCPUs(fCPU * 1000)
+	}
+
+	return nil
+}
+
+// implements json.Marshaler
+func (c *MilliCPUs) MarshalJSON() ([]byte, error) {
+
+	if *c < MinMilliCPUs || *c > MaxMilliCPUs {
+		return nil, ErrInvalidCPUs
+	}
+
+	// always use milli cpus "1000m" format
+	return []byte(fmt.Sprintf("\"%s\"", c.String())), nil
+}
--- a/api/models/config_test.go
+++ b/api/models/config_test.go
@@ -0,0 +1,112 @@
+package models
+
+import (
+	"encoding/json"
+	"fmt"
+	"testing"
+)
+
+func checkStr(input string, expected MilliCPUs) error {
+	var res MilliCPUs
+	err := json.Unmarshal([]byte(input), &res)
+	if err != nil {
+		return err
+	}
+	if expected != res {
+		return fmt.Errorf("mismatch %s != %s", res, expected)
+	}
+	return nil
+}
+
+func checkErr(input string) (MilliCPUs, error) {
+	var res MilliCPUs
+	err := json.Unmarshal([]byte(input), &res)
+	if err != nil {
+		return res, err
+	}
+	return res, nil
+}
+
+func TestMilliCPUsUnmarshal(t *testing.T) {
+
+	err := checkStr("\"1.00\"", 1000)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"1\"", 1000)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"0\"", 0)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"00000\"", 0)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"+00000\"", 0)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"-00000\"", 0)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"0.01\"", 10)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	tmp, err := checkErr("\"1000000000000000000000000\"")
+	if err == nil {
+		t.Fatal("failed, should get error, got: ", tmp)
+	}
+
+	// 0.2341 is too high of a precision for CPUs
+	err = checkStr("\"0.2341\"", 234)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"1m\"", 1)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"1000m\"", 1000)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	tmp, err = checkErr("\"-100\"")
+	if err == nil {
+		t.Fatal("failed, should get error, got: ", tmp)
+	}
+
+	tmp, err = checkErr("\"100000000000m\"")
+	if err == nil {
+		t.Fatal("failed, should get error, got: ", tmp)
+	}
+
+	err = checkStr("\".2\"", 200)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	err = checkStr("\"100.2000\"", 100200)
+	if err != nil {
+		t.Fatal("failed: ", err)
+	}
+
+	tmp, err = checkErr("\"-0.20\"")
+	if err == nil {
+		t.Fatal("failed, should get error got: ", tmp)
+	}
+}
--- a/api/models/error.go
+++ b/api/models/error.go
@@ -184,6 +184,11 @@ var (
 		code:  http.StatusNotFound,
 		error: errors.New("Path not found"),
 	}
+	ErrInvalidCPUs = err{
+		code: http.StatusBadRequest,
+		error: fmt.Errorf("Cpus is invalid. Value should be either between [%.3f and %.3f] or [%dm and %dm] milliCPU units",
+			float64(MinMilliCPUs)/1000.0, float64(MaxMilliCPUs)/1000.0, MinMilliCPUs, MaxMilliCPUs),
+	}
 )

 // APIError any error that implements this interface will return an API response
@@ -202,6 +207,11 @@ func (e err) Code() int { return e.code }

 func NewAPIError(code int, e error) APIError { return err{code, e} }

+func IsAPIError(e error) bool {
+	_, ok := e.(APIError)
+	return ok
+}
+
 // Error uniform error output
 type Error struct {
 	Error *ErrorBody `json:"error,omitempty"`
--- a/api/models/route.go
+++ b/api/models/route.go
@@ -29,6 +29,7 @@ type Route struct {
 	Path        string          `json:"path" db:"path"`
 	Image       string          `json:"image" db:"image"`
 	Memory      uint64          `json:"memory" db:"memory"`
+	CPUs        MilliCPUs       `json:"cpus" db:"cpus"`
 	Headers     Headers         `json:"headers,omitempty" db:"headers"`
 	Type        string          `json:"type" db:"type"`
 	Format      string          `json:"format" db:"format"`
@@ -161,6 +162,7 @@ func (r1 *Route) Equals(r2 *Route) bool {
 	eq = eq && r1.Path == r2.Path
 	eq = eq && r1.Image == r2.Image
 	eq = eq && r1.Memory == r2.Memory
+	eq = eq && r1.CPUs == r2.CPUs
 	eq = eq && r1.Headers.Equals(r2.Headers)
 	eq = eq && r1.Type == r2.Type
 	eq = eq && r1.Format == r2.Format
@@ -186,6 +188,9 @@ func (r *Route) Update(new *Route) {
 	if new.Memory != 0 {
 		r.Memory = new.Memory
 	}
+	if new.CPUs != 0 {
+		r.CPUs = new.CPUs
+	}
 	if new.Type != "" {
 		r.Type = new.Type
 	}
--- a/api/models/route_test.go
+++ b/api/models/route_test.go
@@ -0,0 +1,42 @@
+package models
+
+import (
+	"testing"
+)
+
+func TestRouteSimple(t *testing.T) {
+
+	route1 := &Route{
+		AppName:     "test",
+		Path:        "/some",
+		Image:       "foo",
+		Memory:      128,
+		CPUs:        100,
+		Type:        "sync",
+		Format:      "http",
+		Timeout:     10,
+		IdleTimeout: 10,
+	}
+
+	err := route1.Validate()
+	if err != nil {
+		t.Fatal("should not have failed, got: ", err)
+	}
+
+	route2 := &Route{
+		AppName:     "test",
+		Path:        "/some",
+		Image:       "foo",
+		Memory:      128,
+		CPUs:        100,
+		Type:        "sync",
+		Format:      "nonsense",
+		Timeout:     10,
+		IdleTimeout: 10,
+	}
+
+	err = route2.Validate()
+	if err == nil {
+		t.Fatalf("should have failed route: %#v", route2)
+	}
+}
--- a/api/server/apps_create.go
+++ b/api/server/apps_create.go
@@ -14,7 +14,11 @@ func (s *Server) handleAppCreate(c *gin.Context) {

 	err := c.BindJSON(&wapp)
 	if err != nil {
-		handleErrorResponse(c, models.ErrInvalidJSON)
+		if models.IsAPIError(err) {
+			handleErrorResponse(c, err)
+		} else {
+			handleErrorResponse(c, models.ErrInvalidJSON)
+		}
 		return
 	}

--- a/api/server/apps_update.go
+++ b/api/server/apps_update.go
@@ -15,7 +15,11 @@ func (s *Server) handleAppUpdate(c *gin.Context) {

 	err := c.BindJSON(&wapp)
 	if err != nil {
-		handleErrorResponse(c, models.ErrInvalidJSON)
+		if models.IsAPIError(err) {
+			handleErrorResponse(c, err)
+		} else {
+			handleErrorResponse(c, models.ErrInvalidJSON)
+		}
 		return
 	}

--- a/api/server/hybrid.go
+++ b/api/server/hybrid.go
@@ -17,7 +17,11 @@ func (s *Server) handleRunnerEnqueue(c *gin.Context) {
 	var call models.Call
 	err := c.BindJSON(&call)
 	if err != nil {
-		handleErrorResponse(c, models.ErrInvalidJSON)
+		if models.IsAPIError(err) {
+			handleErrorResponse(c, err)
+		} else {
+			handleErrorResponse(c, models.ErrInvalidJSON)
+		}
 		return
 	}

@@ -92,7 +96,11 @@ func (s *Server) handleRunnerStart(c *gin.Context) {
 	var call models.Call
 	err := c.BindJSON(&call)
 	if err != nil {
-		handleErrorResponse(c, models.ErrInvalidJSON)
+		if models.IsAPIError(err) {
+			handleErrorResponse(c, err)
+		} else {
+			handleErrorResponse(c, models.ErrInvalidJSON)
+		}
 		return
 	}

@@ -143,7 +151,11 @@ func (s *Server) handleRunnerFinish(c *gin.Context) {
 	}
 	err := c.BindJSON(&body)
 	if err != nil {
-		handleErrorResponse(c, models.ErrInvalidJSON)
+		if models.IsAPIError(err) {
+			handleErrorResponse(c, err)
+		} else {
+			handleErrorResponse(c, models.ErrInvalidJSON)
+		}
 		return
 	}

--- a/api/server/middleware_test.go
+++ b/api/server/middleware_test.go
@@ -73,7 +73,7 @@ func TestRootMiddleware(t *testing.T) {
 			{Name: "myapp2", Config: models.Config{}},
 		},
 		[]*models.Route{
-			{Path: "/", AppName: "myapp", Image: "fnproject/hello", Type: "sync", Memory: 128, Timeout: 30, IdleTimeout: 30, Headers: map[string][]string{"X-Function": {"Test"}}},
+			{Path: "/", AppName: "myapp", Image: "fnproject/hello", Type: "sync", Memory: 128, CPUs: 100, Timeout: 30, IdleTimeout: 30, Headers: map[string][]string{"X-Function": {"Test"}}},
 			{Path: "/myroute", AppName: "myapp", Image: "fnproject/hello", Type: "sync", Memory: 128, Timeout: 30, IdleTimeout: 30, Headers: map[string][]string{"X-Function": {"Test"}}},
 			{Path: "/app2func", AppName: "myapp2", Image: "fnproject/hello", Type: "sync", Memory: 128, Timeout: 30, IdleTimeout: 30, Headers: map[string][]string{"X-Function": {"Test"}},
 				Config: map[string]string{"NAME": "johnny"},
--- a/api/server/routes_create_update.go
+++ b/api/server/routes_create_update.go
@@ -141,6 +141,9 @@ func (s *Server) ensureApp(ctx context.Context, wroute *models.RouteWrapper, met
 func bindRoute(c *gin.Context, method string, wroute *models.RouteWrapper) error {
 	err := c.BindJSON(wroute)
 	if err != nil {
+		if models.IsAPIError(err) {
+			return err
+		}
 		return models.ErrInvalidJSON
 	}

--- a/api/server/routes_test.go
+++ b/api/server/routes_test.go
@@ -107,6 +107,7 @@ func TestRouteCreate(t *testing.T) {
 		{datastore.NewMock(), logs.NewMock(), http.MethodPost, "/v1/apps/a/routes", `{ "route": { "image": "fnproject/hello", "type": "sync" } }`, http.StatusBadRequest, models.ErrRoutesMissingPath},
 		{datastore.NewMock(), logs.NewMock(), http.MethodPost, "/v1/apps/a/routes", `{ "route": { "image": "fnproject/hello", "path": "myroute", "type": "sync" } }`, http.StatusBadRequest, models.ErrRoutesInvalidPath},
 		{datastore.NewMock(), logs.NewMock(), http.MethodPost, "/v1/apps/$/routes", `{ "route": { "image": "fnproject/hello", "path": "/myroute", "type": "sync" } }`, http.StatusBadRequest, models.ErrAppsInvalidName},
+		{datastore.NewMock(), logs.NewMock(), http.MethodPost, "/v1/apps/a/routes", `{ "route": { "image": "fnproject/hello", "path": "/myroute", "type": "sync", "cpus": "-100" } }`, http.StatusBadRequest, models.ErrInvalidCPUs},
 		{datastore.NewMockInit(nil,
 			[]*models.Route{
 				{
@@ -118,6 +119,8 @@ func TestRouteCreate(t *testing.T) {

 		// success
 		{datastore.NewMock(), logs.NewMock(), http.MethodPost, "/v1/apps/a/routes", `{ "route": { "image": "fnproject/hello", "path": "/myroute", "type": "sync" } }`, http.StatusOK, nil},
+		{datastore.NewMock(), logs.NewMock(), http.MethodPost, "/v1/apps/a/routes", `{ "route": { "image": "fnproject/hello", "path": "/myroute", "type": "sync", "cpus": "100m" } }`, http.StatusOK, nil},
+		{datastore.NewMock(), logs.NewMock(), http.MethodPost, "/v1/apps/a/routes", `{ "route": { "image": "fnproject/hello", "path": "/myroute", "type": "sync", "cpus": "0.2" } }`, http.StatusOK, nil},
 	} {
 		test.run(t, i, buf)
 	}
@@ -332,6 +335,7 @@ func TestRouteUpdate(t *testing.T) {
 		{ds, logs.NewMock(), http.MethodPatch, "/v1/apps/a/routes/myroute/do", `{ "route": { "type": "async", "timeout": 121, "idle_timeout": 240 } }`, http.StatusOK, nil}, // should work if async
 		{ds, logs.NewMock(), http.MethodPatch, "/v1/apps/a/routes/myroute/do", `{ "route": { "idle_timeout": 3601 } }`, http.StatusBadRequest, models.ErrRoutesInvalidIdleTimeout},
 		{ds, logs.NewMock(), http.MethodPatch, "/v1/apps/a/routes/myroute/do", `{ "route": { "memory": 100000000000000 } }`, http.StatusBadRequest, models.ErrRoutesInvalidMemory},
+		{ds, logs.NewMock(), http.MethodPatch, "/v1/apps/a/routes/myroute/do", `{ "route": { "cpus": "foo" } }`, http.StatusBadRequest, models.ErrInvalidCPUs},
 		// TODO this should be correct, waiting for patch to come in
 		//{ds, logs.NewMock(), http.MethodPatch, "/v1/apps/b/routes/myroute/dont", `{ "route": {} }`, http.StatusNotFound, models.ErrAppsNotFound},
 		{ds, logs.NewMock(), http.MethodPatch, "/v1/apps/a/routes/myroute/dont", `{ "route": {} }`, http.StatusNotFound, models.ErrRoutesNotFound},
--- a/api/server/runner_async_test.go
+++ b/api/server/runner_async_test.go
@@ -42,7 +42,7 @@ func TestRouteRunnerAsyncExecution(t *testing.T) {
 		[]*models.Route{
 			{Type: "async", Path: "/hot-http", AppName: "myapp", Image: "fnproject/fn-test-utils", Format: "http", Config: map[string]string{"test": "true"}, Memory: 128, Timeout: 4, IdleTimeout: 30},
 			{Type: "async", Path: "/hot-json", AppName: "myapp", Image: "fnproject/fn-test-utils", Format: "json", Config: map[string]string{"test": "true"}, Memory: 128, Timeout: 4, IdleTimeout: 30},
-			{Type: "async", Path: "/myroute", AppName: "myapp", Image: "fnproject/hello", Config: map[string]string{"test": "true"}, Memory: 128, Timeout: 30, IdleTimeout: 30},
+			{Type: "async", Path: "/myroute", AppName: "myapp", Image: "fnproject/hello", Config: map[string]string{"test": "true"}, Memory: 128, CPUs: 200, Timeout: 30, IdleTimeout: 30},
 			{Type: "async", Path: "/myerror", AppName: "myapp", Image: "fnproject/error", Config: map[string]string{"test": "true"}, Memory: 128, Timeout: 30, IdleTimeout: 30},
 			{Type: "async", Path: "/myroute/:param", AppName: "myapp", Image: "fnproject/hello", Config: map[string]string{"test": "true"}, Memory: 128, Timeout: 30, IdleTimeout: 30},
 		}, nil,