konduktor-nightly 0.1.0.dev20250519104943__py3-none-any.whl → 0.1.0.dev20250521104900__py3-none-any.whl

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.dev20250519104943'
+__version__ = '1.0.0.dev0.1.0.dev20250521104900'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

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.dev20250519104943.dist-info → konduktor_nightly-0.1.0.dev20250521104900.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250519104943
+Version: 0.1.0.dev20250521104900
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250519104943.dist-info → konduktor_nightly-0.1.0.dev20250521104900.dist-info}/RECORD

@@ -1,4 +1,4 @@
-konduktor/__init__.py,sha256=R1b8oEBavaV7tqJx9KqKstRMivzsCoXfSblRpMwSmc4,1540
+konduktor/__init__.py,sha256=BpU070cC1XqK9i4AWlWvYixgyNGfMsI40UCLdAsI8us,1540
 konduktor/adaptors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 konduktor/adaptors/aws.py,sha256=s47Ra-GaqCQibzVfmD0pmwEWHif1EGO5opMbwkLxTCU,8244
 konduktor/adaptors/common.py,sha256=ZIqzjx77PIHUwpjfAQ1uX8B2aX78YMuGj4Bppd-MdyM,4183
@@ -71,7 +71,7 @@ konduktor/manifests/pod_cleanup_controller.yaml,sha256=hziL1Ka1kCAEL9R7Tjvpb80iw
 konduktor/resource.py,sha256=w2PdIrmQaJWA-GLSmVBcg4lxwuxvPulz35_YSKa5o24,19254
 konduktor/task.py,sha256=ofwd8WIhfD6C3ThLcv6X3GUzQHyZ6ddjUagE-umF4K0,35207
 konduktor/templates/jobset.yaml.j2,sha256=onYiHtXAgk-XBtji994hPu_g0hxnLzvmfxwjbdKdeZc,960
-konduktor/templates/pod.yaml.j2,sha256=TNwVnadENnLmYa753yNlvyjmfkibHviaOu3OkCmz6mo,17778
+konduktor/templates/pod.yaml.j2,sha256=v0s_gLmr7bBMVtdiElHbBvt36bJzAxBptqWVLzvdvE4,15520
 konduktor/usage/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 konduktor/usage/constants.py,sha256=gCL8afIHZhO0dcxbJGpESE9sCC1cBSbeRnQ8GwNOY4M,612
 konduktor/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -91,8 +91,8 @@ konduktor/utils/schemas.py,sha256=2fHsTi3t9q3LXqOPrcpkmPsMbaoJBnuJstd6ULmDiUo,16
 konduktor/utils/subprocess_utils.py,sha256=WoFkoFhGecPR8-rF8WJxbIe-YtV94LXz9UG64SDhCY4,9448
 konduktor/utils/ux_utils.py,sha256=czCwiS1bDqgeKtzAJctczpLwFZzAse7WuozdvzEFYJ4,7437
 konduktor/utils/validator.py,sha256=tgBghVyedyzGx84-U2Qfoh_cJBE3oUk9gclMW90ORks,691
-konduktor_nightly-0.1.0.dev20250519104943.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
-konduktor_nightly-0.1.0.dev20250519104943.dist-info/METADATA,sha256=8K22_WDfcgSJrA_5XvaXrjj6PzsAT0cKetHf1jgTTMA,4366
-konduktor_nightly-0.1.0.dev20250519104943.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
-konduktor_nightly-0.1.0.dev20250519104943.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
-konduktor_nightly-0.1.0.dev20250519104943.dist-info/RECORD,,
+konduktor_nightly-0.1.0.dev20250521104900.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
+konduktor_nightly-0.1.0.dev20250521104900.dist-info/METADATA,sha256=Zy1uJzy9wocXmmbkhqofIoykrQ_Re5xwCHvnj6jxJGM,4366
+konduktor_nightly-0.1.0.dev20250521104900.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+konduktor_nightly-0.1.0.dev20250521104900.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
+konduktor_nightly-0.1.0.dev20250521104900.dist-info/RECORD,,