Fix several bugs

Author: adityapb

Makefile

@@ -213,8 +213,5 @@ volcano-scheduler-deploy: volcano-scheduler-crd
 
 .PHONY: charm
 charm: ${BIN_DIR}
-rm -rf $(PROJECT_DIR)/dep-libs/charm
-mkdir -p $(PROJECT_DIR)/dep-libs/charm
-git clone $(CHARM_REPO) $(PROJECT_DIR)/dep-libs/charm
-cd $(PROJECT_DIR)/dep-libs/charm && git checkout shrinkexpand-fix && ./build charm++ netlrts-linux-x86_64 --enable-shrinkexpand -j8 --force
+cd $(PROJECT_DIR)/dep-libs/charm && git checkout kubernetes && ./build charm++ netlrts-linux-x86_64 --enable-shrinkexpand -j8 --force
 cd pkg/controller && $(PROJECT_DIR)/dep-libs/charm/bin/charmc -language c++ -seq -o ${PROJECT_DIR}/${BIN_DIR}/rescale_client rescale_client.C -lccs-client

cmd/mpi-operator/app/options/options.go

@@ -59,7 +59,7 @@ func (s *ServerOption) AddFlags(fs *flag.FlagSet) {
 `The namespace to monitor mpijobs. If unset, it monitors all namespaces cluster-wide. 
  If set, it only monitors mpijobs in the given namespace.`)
 
-fs.IntVar(&s.Threadiness, "threadiness", 2,
+fs.IntVar(&s.Threadiness, "threadiness", 1,
 `How many threads to process the main logic`)
 
 fs.BoolVar(&s.PrintVersion, "version", false, "Show version and quit")

deploy/v2beta1/mpi-operator.yaml

@@ -8241,7 +8241,7 @@ spec:
  - args:
  - -alsologtostderr
  - --lock-namespace=mpi-operator
- image: adityapb/mpi-operator:autoscale
+ image: adityapb/mpi-operator:rescale-gap
  imagePullPolicy: Always
  name: mpi-operator
  serviceAccountName: mpi-operator

examples/v2beta1/charm/charm-0.yaml

@@ -8,7 +8,7 @@ spec:
  priority: 1
  runPolicy:
  cleanPodPolicy: Running
- ttlSecondsAfterFinished: 60
+ ttlSecondsAfterFinished: 600
  sshAuthMountPath: /home/mpiuser/.ssh
  mpiReplicaSpecs:
  Launcher:
@@ -25,9 +25,9 @@ spec:
  - /app/charmrun
  args:
  - /app/jacobi2d
- - "4096"
+ - "8192"
  - "512"
- - "3000"
+ - "600"
  - +balancer
  - MetisLB
  - +LBDebug
@@ -36,7 +36,6 @@ spec:
  resources:
  limits:
  cpu: 1
- memory: 1Gi
  Worker:
  minReplicas: 1
  maxReplicas: 10
@@ -46,6 +45,9 @@ spec:
  - image: adityapb/charm-netlrts:dev-new
  imagePullPolicy: Always
  name: mpi-worker
+ volumeMounts:
+ - mountPath: /dev/shm
+ name: shm-volume
  securityContext:
  runAsUser: 1000
  command:
@@ -57,3 +59,8 @@ spec:
  resources:
  limits:
  cpu: 1
+ volumes:
+ - name: shm-volume
+ emptyDir:
+ medium: Memory
+ sizeLimit: 1Gi

examples/v2beta1/charm/charm-2.yaml

@@ -8,7 +8,7 @@ spec:
  priority: 3
  runPolicy:
  cleanPodPolicy: Running
- ttlSecondsAfterFinished: 60
+ ttlSecondsAfterFinished: 600
  sshAuthMountPath: /home/mpiuser/.ssh
  mpiReplicaSpecs:
  Launcher:
@@ -36,7 +36,6 @@ spec:
  resources:
  limits:
  cpu: 1
- memory: 1Gi
  Worker:
  minReplicas: 1
  maxReplicas: 10