Add subnet tag to node config hash generation (#250)

* Add subnet tag to node config hash generation; fix GPU dockerfile for volumens fix

* Adjust Dockefile so it installs ray-on-golem globally

Dockerfile

@@ -31,11 +31,13 @@ COPY ray_on_golem/__init__.py /app/ray_on_golem/__init__.py
 
 RUN pip install poetry && \
 	poetry config virtualenvs.create false
+RUN poetry install --no-interaction --no-ansi --only ray
 
 RUN pip config set global.index-url https://pypi.dev.golem.network/simple
 RUN pip install pillow
+
 RUN python -m venv --system-site-packages /root/venv
-RUN /bin/bash -c "source /root/venv/bin/activate && poetry install --no-interaction --no-ansi --only ray"
+RUN bash -c "echo source /root/venv/bin/activate >> /root/.bashrc"
 
 COPY ray_on_golem /app/ray_on_golem/
 

examples/GPU.Dockerfile

@@ -43,6 +43,9 @@ RUN pip install pillow
 RUN pip install numpy numba
 RUN pip config set global.index-url https://pypi.dev.golem.network/simple
 
+RUN python -m venv --system-site-packages /root/venv
+RUN bash -c "echo source /root/venv/bin/activate >> /root/.bashrc"
+
 COPY ray_on_golem /app/ray_on_golem/
 
 RUN rm -r /root/.cache

ray_on_golem/server/cluster/cluster.py

@@ -251,7 +251,7 @@ async def _get_or_create_manager_stack(
         node_config: NodeConfigData,
         subnet_tag: str,
     ) -> ManagerStack:
-        stack_hash = node_config.get_hash()
+        stack_hash = node_config.get_hash(subnet_tag)
 
         async with self._manager_stacks_locks[stack_hash]:
             stack = self._manager_stacks.get(stack_hash)

ray_on_golem/server/cluster/nodes.py

@@ -333,7 +333,6 @@ async def _upload_node_configuration(
         await self._run_command(
             context, f"echo 'export PATH=$PATH:/root/.local/bin' >> /root/.bashrc"
         )
-        await self._run_command(context, f"echo 'source /root/venv/bin/activate' >> /root/.bashrc")
 
         await self._run_command(context, "mkdir -p /root/.ssh")
         await self._run_command(

ray_on_golem/server/models.py

@@ -87,8 +87,8 @@ class NodeConfigData(BaseModel):
     class Config:
         extra = "forbid"
 
-    def get_hash(self) -> str:
-        return hashlib.md5(self.json().encode()).hexdigest()
+    def get_hash(self, subnet_tag: str) -> str:
+        return hashlib.md5(self.json().encode()).hexdigest() + "-" + subnet_tag
 
 
 class ProviderParametersData(BaseModel):