konduktor-nightly 0.1.0.dev20250520104912__tar.gz → 0.1.0.dev20250521104900__tar.gz

{konduktor_nightly-0.1.0.dev20250520104912 → konduktor_nightly-0.1.0.dev20250521104900}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250520104912
+Version: 0.1.0.dev20250521104900
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250520104912 → konduktor_nightly-0.1.0.dev20250521104900}/konduktor/__init__.py

@@ -14,7 +14,7 @@ __all__ = [
 ]
 
 # Replaced with the current commit when building the wheels.
-_KONDUKTOR_COMMIT_SHA = '709675de56c5bb11cf4131c48e0fdfc96c67decf'
+_KONDUKTOR_COMMIT_SHA = '46626c8c0df4bbbd5a8fc164b2d2f66c26dbdd33'
 os.makedirs(os.path.expanduser('~/.konduktor'), exist_ok=True)
 
 
@@ -48,5 +48,5 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev0.1.0.dev20250520104912'
+__version__ = '1.0.0.dev0.1.0.dev20250521104900'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

{konduktor_nightly-0.1.0.dev20250520104912 → konduktor_nightly-0.1.0.dev20250521104900}/konduktor/templates/pod.yaml.j2

@@ -15,36 +15,6 @@ kubernetes:
       - key: "nvidia.com/gpu"
         operator: "Exists"
       {% endif %}
-      initContainers:
-        - name: setup-synchronizer
-          image: "alpine:3.19"
-          restartPolicy: Always
-          command: ["/bin/sh", "-c"]
-          args:
-          - |  
-            apk add --no-cache socat 
-            wget https://raw.githubusercontent.com/asaiacai/dumb_barrier/refs/heads/main/dumb_barrier.sh 
-            sh -x dumb_barrier.sh
-          volumeMounts:
-            - name: sync
-              mountPath: /tmp/konduktor
-          env:
-            - name: MASTER_ADDR
-              value: "{{ master_addr }}"
-            - name: RANK
-              valueFrom:
-                fieldRef:
-                  fieldPath: metadata.annotations['batch.kubernetes.io/job-completion-index']
-            - name: WORLD_SIZE
-              value: "{{ num_nodes }}"
-            - name: MASTER_PORT
-              value: "11111"
-            - name: GO_PORT
-              value: "11112"
-            - name: POD_NAMESPACE
-              valueFrom:
-                fieldRef:
-                  fieldPath: metadata.namespace
       containers:
         # TODO(asaiacai): should decide here whether we add the fabric interfaces/containers init etc.        
         - name: konduktor-container
@@ -327,33 +297,6 @@ kubernetes:
               ulimit -Sc 0 && ulimit -Hc 0
               $(prefix_cmd) echo "===== KONDUKTOR: Initialization took $end_setup_time seconds ====="
               set +eo pipefail
-              $(prefix_cmd) cd {{ remote_workdir }}
-              {% if setup_cmd %}
-              # setup task
-              $(prefix_cmd) echo "===== KONDUKTOR: Running setup ======="
-              {{ setup_cmd | indent( width=14 ) }}
-              {% endif %}
-
-              # synchronize workers before executing `run`
-              set -e
-              touch "/tmp/konduktor/SETUP"
-              # TODO(asaiacai): should we make this value tuneable for users?
-              TIMEOUT=300
-              start_sync=$(date +%s);
-              DEADLINE=$(( $(date +%s) + TIMEOUT ))
-
-              echo "[KONDUKTOR: main] Waiting for workers to synchronize"
-              while [ ! -f "/tmp/konduktor/READY" ]; do
-                  if [ "$(date +%s)" -ge "$DEADLINE" ]; then
-                      echo "[KONDUKTOR: main] ERROR: Timed out after 2 minutes of waiting for worker synchronization"
-                      exit 1
-                  fi
-                  sleep 0.5
-              done
-              echo "[KONDUKTOR: main] All workers have joined"
-              end_sync=$(date +%s);
-              echo "[KONDUKTOR: main] Synchronization took $((end_sync - start_sync)) seconds"
-              set +eo pipefail
               # run task  
               $(prefix_cmd) cd {{ remote_workdir }}
               $(prefix_cmd) echo "===== KONDUKTOR: Running task ====="
@@ -395,8 +338,6 @@ kubernetes:
         secret:
           secretName: {{ secret_name }}
       {% endfor %}
-
-        
         
       # TODO(asaiacai): should we add nodeSelectors here or leave to
       # kueue resource flavors. leaning towards defining

{konduktor_nightly-0.1.0.dev20250520104912 → konduktor_nightly-0.1.0.dev20250521104900}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "konduktor-nightly"
-version = "0.1.0.dev20250520104912"
+version = "0.1.0.dev20250521104900"
 description = "GPU Cluster Health Management"
 packages = [
     {include = "konduktor"}