feat(worker): provisioned worker should not take when memory requirements (#4084)

Signed-off-by: Benjamin Coenen <[email protected]>

Author: bnjjj

docs/content/docs/concepts/requirement/_index.md

@@ -13,7 +13,7 @@ Type of requirements:
 - Hostname
 - [Network access]({{< relref "/docs/concepts/requirement/requirement_network.md" >}})
 - [Service]({{< relref "/docs/concepts/requirement/requirement_service.md" >}})
-- Memory
+- [Memory]({{< relref "/docs/concepts/requirement/requirement_memory.md" >}})
 - [OS & Architecture]({{< relref "/docs/concepts/requirement/requirement_os_arch.md" >}})
 
 A [Job]({{< relref "/docs/concepts/job.md" >}}) will be executed by a **worker**.

docs/content/docs/concepts/requirement/requirement_memory.md

@@ -0,0 +1,8 @@
+---
+title: "Memory"
+weight: 6
+---
+
+The Memory requirement allows you to require a worker to have a specific number of mb of RAM.
+
+For example if you need 2Gb of RAM for your worker you can put `2048` in your memory requirement.

engine/hatchery/kubernetes/kubernetes.go

@@ -15,9 +15,9 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/client-go/kubernetes"
+	_ "k8s.io/client-go/plugin/pkg/client/auth"
 	"k8s.io/client-go/rest"
 	"k8s.io/client-go/tools/clientcmd"
-  _ "k8s.io/client-go/plugin/pkg/client/auth"
 	clientcmdapi "k8s.io/client-go/tools/clientcmd/api"
 
 	"github.com/ovh/cds/engine/api"
@@ -233,6 +233,10 @@ func (h *HatcheryKubernetes) SpawnWorker(ctx context.Context, spawnArgs hatchery
 		label = "register"
 	}
 
+	// Kubernetes pod name must not be > 63 chars
+	if len(name) > 63 {
+		name = name[:60]
+	}
 	log.Debug("hatchery> kubernetes> SpawnWorker> %s", name)
 
 	var logJob string
@@ -290,6 +294,7 @@ func (h *HatcheryKubernetes) SpawnWorker(ctx context.Context, spawnArgs hatchery
 	}
 	envsWm := map[string]string{}
 	envsWm["CDS_FORCE_EXIT"] = "1"
+	envsWm["CDS_MODEL_MEMORY"] = fmt.Sprintf("%d", memory)
 	envsWm["CDS_API"] = udataParam.API
 	envsWm["CDS_TOKEN"] = udataParam.Token
 	envsWm["CDS_NAME"] = udataParam.Name

engine/hatchery/marathon/marathon.go

@@ -285,6 +285,7 @@ func (h *HatcheryMarathon) SpawnWorker(ctx context.Context, spawnArgs hatchery.S
 
 	envsWm := map[string]string{}
 	envsWm["CDS_FORCE_EXIT"] = "0"
+	envsWm["CDS_MODEL_MEMORY"] = fmt.Sprintf("%d", memory)
 	envsWm["CDS_API"] = udataParam.API
 	envsWm["CDS_TOKEN"] = udataParam.Token
 	envsWm["CDS_NAME"] = udataParam.Name

engine/hatchery/swarm/swarm.go

@@ -380,6 +380,7 @@ func (h *HatcherySwarm) SpawnWorker(ctx context.Context, spawnArgs hatchery.Spaw
 
 	envsWm := map[string]string{}
 	envsWm["CDS_FORCE_EXIT"] = "1"
+	envsWm["CDS_MODEL_MEMORY"] = fmt.Sprintf("%d", memory)
 	envsWm["CDS_API"] = udataParam.API
 	envsWm["CDS_TOKEN"] = udataParam.Token
 	envsWm["CDS_NAME"] = udataParam.Name

engine/worker/cmd_run.go

@@ -229,7 +229,7 @@ func runCmd(w *currentWorker) func(cmd *cobra.Command, args []string) {
 				var requirementsOK, pluginsOK bool
 				var t string
 				if exceptJobID != j.ID && w.bookedWJobID == 0 { // If we already check the requirements before and it was OK
-					requirementsOK, _ = checkRequirements(w, &j.Job.Action, nil, j.ID)
+					requirementsOK, _ = checkRequirements(w, &j.Job.Action, j.ID)
 					if j.ID == w.bookedWJobID {
 						t = ", this was my booked job"
 					}
@@ -304,7 +304,7 @@ func (w *currentWorker) processBookedWJob(ctx context.Context, wjobs chan<- sdk.
 		return sdk.WrapError(err, "Unable to load workflow node job %d", w.bookedWJobID)
 	}
 
-	requirementsOK, errRequirements := checkRequirements(w, &wjob.Job.Action, nil, wjob.ID)
+	requirementsOK, errRequirements := checkRequirements(w, &wjob.Job.Action, wjob.ID)
 	if !requirementsOK {
 		var details string
 		for _, r := range errRequirements {

engine/worker/requirement.go

@@ -28,31 +28,12 @@ var requirementCheckFuncs = map[string]func(w *currentWorker, r sdk.Requirement)
 	sdk.OSArchRequirement:        checkOSArchRequirement,
 }
 
-func checkRequirements(w *currentWorker, a *sdk.Action, execGroups []sdk.Group, bookedJobID int64) (bool, []sdk.Requirement) {
+func checkRequirements(w *currentWorker, a *sdk.Action, bookedJobID int64) (bool, []sdk.Requirement) {
 	requirementsOK := true
 	errRequirements := []sdk.Requirement{}
 
 	log.Debug("checkRequirements> for JobID:%d model of worker: %s", bookedJobID, w.model.Name)
 
-	// DEPRECATED
-	// this code is useful for pipelineBuildJob
-	// with CDS Workflows, the queue contains only jobs executable by worker
-	// after removing pbBuildJob, check execGroups here can be removed
-	if execGroups != nil && len(execGroups) > 0 && w.model.GroupID > 0 {
-		checkGroup := false
-		for _, g := range execGroups {
-			if g.ID == w.model.GroupID {
-				checkGroup = true
-				break
-			}
-		}
-		if !checkGroup {
-			requirementsOK = false
-			log.Debug("checkRequirements> model %s attached to group %d can't run this job", w.model.Name, w.model.GroupID)
-			return requirementsOK, nil
-		}
-	}
-
 	log.Debug("requirements for %s >>> %+v\n", a.Name, a.Requirements)
 	for _, r := range a.Requirements {
 		ok, err := checkRequirement(w, r)
@@ -177,19 +158,33 @@ func checkServiceRequirement(w *currentWorker, r sdk.Requirement) (bool, error)
 }
 
 func checkMemoryRequirement(w *currentWorker, r sdk.Requirement) (bool, error) {
-	v, err := mem.VirtualMemory()
+	var totalMemory int64
+	neededMemory, err := strconv.ParseInt(r.Value, 10, 64)
 	if err != nil {
 		return false, err
 	}
-	totalMemory := v.Total
 
-	neededMemory, err := strconv.ParseInt(r.Value, 10, 64)
-	if err != nil {
-		return false, err
+	switch w.model.Type {
+	// Check env variables in a docker is safer than mem.VirtualMemory
+	case sdk.Docker:
+		var err error
+		// Useful for provisioned worker
+		memoryEnv := os.Getenv("CDS_MODEL_MEMORY")
+		totalMemory, err = strconv.ParseInt(memoryEnv, 10, 64)
+		if err != nil {
+			return false, err
+		}
+		totalMemory = totalMemory * 1024 * 1024
+	default:
+		v, err := mem.VirtualMemory()
+		if err != nil {
+			return false, err
+		}
+		totalMemory = int64(v.Total)
 	}
 	//Assuming memory is in megabytes
 	//If we have more than 90% of neededMemory, lets do it
-	return int64(totalMemory) >= (neededMemory*1024*1024)*90/100, nil
+	return totalMemory >= (neededMemory*1024*1024)*90/100, nil
 }
 
 func checkVolumeRequirement(w *currentWorker, r sdk.Requirement) (bool, error) {