feat: preemption

README.md

@@ -3,34 +3,13 @@
 This is the [Google Batch](https://cloud.google.com/batch/docs/get-started) external executor plugin for snakemake.
 For documentation, see the [Snakemake plugin catalog](https://snakemake.github.io/snakemake-plugin-catalog/plugins/executor/googlebatch.html).
 
-###  TODO
-
-- Add bash strict mode (should default to true)
-- Integrate snakemake MPI support (needs to work with snippet)
-
-
-### Questions
-
-- What leads to STATE_UNSPECIFIED?
-- For Google: what is the source of truth (listing) for batch? I see different answers in different places.
-- For Johannes: Why can't we use debug logging for executor plugins? I instead need to use info and make it very verbose.
-- For All: How do we want to use [COS](https://cloud.google.com/container-optimized-os/docs/concepts/features-and-benefits)? It would allow a container base to be used instead I think?
-
 ### Notes
 
 - Conda is used to install Snakemake and dependencies.
 - The COS (container OS) uses the default Snakemake container, unless you specify differently.
 
-### Feedback
-
-- Debugging batch is impossible (and slow). A "hello world" workflow takes 10 minutes to run and debug once.
-- The jobs table is slow to load and sometimes does not load / shows old jobs at the top (without me touching anything)
-- The logs directly in batch are so much better! Having the stream option there would still be nice (vs. having to refresh.)
-- The batch UI (jobs table) is very slow to load and often just doesn't even after button or page refresh.
-
 For examples, look into the [examples](examples) directory.
 
-
 ## Developer
 
 The instructions for creating and scaffolding this plugin are [here](https://github.com/snakemake/poetry-snakemake-plugin#scaffolding-an-executor-plugin).

docs/further.md

@@ -331,4 +331,4 @@ rule hello_world:
 		googlebatch_snippets="mpi,myscript.sh"
 	shell:
         "..."
-```
+```

example/README.md

@@ -0,0 +1,6 @@
+# Snakemake Google Batch Examples
+
+ - [hello-world](hello-world): The most basic hello world example with multiple langauges
+ - [hello-world-intel-mpi](hello-world-intel-mpi): Run an example MPI job to calculate Pi with intel MPI
+ - [hello-world-cos](hello-world-cos): The same using the container operating system (COS)
+ - [hello-world-preemption](hello-world-preemption): Example asking for preemptible jobs for all rules.

example/hello-world-preemption/README.md

@@ -0,0 +1,9 @@
+# Hello World
+
+> with preemtible instances
+
+Here is an example specifying to use an s3 bucket and asking for preemptible instances.
+
+```bash
+snakemake --jobs 1 --executor googlebatch --googlebatch-region us-central1 --googlebatch-project llnl-flux --default-storage-provider s3 --default-storage-prefix s3://snakemake-testing-llnl --preemptible-rules
+```

example/hello-world-preemption/Snakefile

@@ -0,0 +1,19 @@
+# By convention, the first pseudorule should be called "all"
+# We're using the expand() function to create multiple targets
+rule all:
+	input:
+		expand(
+			"{greeting}/world.txt",
+			greeting = ['hello', 'hola'],
+		),
+
+# First real rule, this is using a wildcard called "greeting"
+rule multilingual_hello_world:
+	output:
+		"{greeting}/world.txt",
+	shell:
+		"""
+		mkdir -p "{wildcards.greeting}"
+		sleep 5
+		echo "{wildcards.greeting}, World!" > {output}
+		"""

scripts/install-snek.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+
+url=$1
+
+tmp=$(mktemp -d -t snek-install-XXXX)
+rm -rf ${tmp}
+
+git clone --depth 1 ${url} ${tmp}
+cd ${tmp}
+/opt/conda/bin/python -m pip install .
+cd -
+rm -rf ${tmp}

snakemake_executor_plugin_googlebatch/__init__.py

@@ -82,7 +82,7 @@ class ExecutorSettings(ExecutorSettingsBase):
         },
     )
 
-    work_tasks: Optional[str] = field(
+    work_tasks: Optional[int] = field(
         default=1,
         metadata={
             "help": "The default number of work tasks (these are NOT MPI ranks)",
@@ -91,7 +91,7 @@ class ExecutorSettings(ExecutorSettingsBase):
         },
     )
 
-    cpu_milli: Optional[str] = field(
+    cpu_milli: Optional[int] = field(
         default=1000,
         metadata={
             "help": "Milliseconds per cpu-second",
@@ -100,7 +100,7 @@ class ExecutorSettings(ExecutorSettingsBase):
         },
     )
 
-    cpu_milli: Optional[str] = field(
+    cpu_milli: Optional[int] = field(
         default=1000,
         metadata={
             "help": "Milliseconds per cpu-second",
@@ -109,7 +109,7 @@ class ExecutorSettings(ExecutorSettingsBase):
         },
     )
 
-    work_tasks_per_node: Optional[str] = field(
+    work_tasks_per_node: Optional[int] = field(
         default=1,
         metadata={
             "help": "The default number of work tasks per node (NOT MPI ranks)",
@@ -118,7 +118,7 @@ class ExecutorSettings(ExecutorSettingsBase):
         },
     )
 
-    memory: Optional[str] = field(
+    memory: Optional[int] = field(
         default=1000,
         metadata={
             "help": "Memory in MiB",
@@ -136,7 +136,7 @@ class ExecutorSettings(ExecutorSettingsBase):
         },
     )
 
-    retry_count: Optional[str] = field(
+    retry_count: Optional[int] = field(
         default=1,
         metadata={
             "help": "Retry count (default to 1)",

snakemake_executor_plugin_googlebatch/command.py

@@ -31,38 +31,31 @@
 )
 
 install_snakemake = """
-# Only the main job should install conda (rest can use it)
 echo "I am batch index ${BATCH_TASK_INDEX}"
 export PATH=/opt/conda/bin:${PATH}
-if [ $BATCH_TASK_INDEX = 0 ] && [ ! -d "/opt/conda" ] ; then
-    workdir=$(pwd)
-    url=https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
-    wget ${url} -O ./miniconda.sh
-    chmod +x ./miniconda.sh
-    bash ./miniconda.sh -b -u -p /opt/conda
-    rm -rf ./miniconda.sh
-    conda config --system --set channel_priority strict
-    which python
-    /opt/conda/bin/python --version
-    url=https://github.com/snakemake/snakemake-interface-common
-    git clone --depth 1 ${url} /tmp/snakemake-common
-    cd /tmp/snakemake-common
-    /opt/conda/bin/python -m pip install .
-    url=https://github.com/snakemake/snakemake-interface-executor-plugins
-    git clone --depth 1 ${url} /tmp/snakemake-plugin
-    cd /tmp/snakemake-plugin
-    /opt/conda/bin/python -m pip install .
-    git clone --depth 1 https://github.com/snakemake/snakemake /tmp/snakemake
-    cd /tmp/snakemake
-    /opt/conda/bin/python -m pip install .
-    cd ${workdir}
-fi
+repo=https://raw.githubusercontent.com/snakemake/snakemake-executor-plugin-googlebatch
+path=add-preemtible/scripts/install-snek.sh
+wget ${repo}/${path}
+chmod +x ./install-snek.sh
+workdir=$(pwd)
+url=https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
+wget ${url} -O ./miniconda.sh
+chmod +x ./miniconda.sh
+bash ./miniconda.sh -b -u -p /opt/conda
+rm -rf ./miniconda.sh
+
+which python
+/opt/conda/bin/python --version
+./install-snek.sh https://github.com/snakemake/snakemake-storage-plugin-gcs
+./install-snek.sh https://github.com/snakemake/snakemake
+
+cd ${workdir}
 """
 
 check_for_snakemake = (
     snakemake_base_environment
     + """
-$(pwd)
+echo $(pwd)
 ls
 which snakemake || whereis snakemake
 """

snakemake_executor_plugin_googlebatch/executor.py

@@ -110,6 +110,19 @@ def validate(self, job):
                 f"Job {job} has container without image_family batch-cos*."
             )
 
+    def is_preemptible(self, job):
+        """
+        Determine if a job is preemptible.
+
+        The logic for determining if the set is valid should belong upstream.
+        """
+        is_p = self.workflow.remote_execution_settings.preemptible_rules.is_preemptible
+        if job.is_group():
+            preemptible = all(is_p(rule) for rule in job.rules)
+        else:
+            preemptible = is_p(job.rule.name)
+        return preemptible
+
     def get_command_writer(self, job):
         """
         Get a command writer for a job.
@@ -178,6 +191,9 @@ def run_job(self, job: JobExecutorInterface):
         setup.script = batch_v1.Runnable.Script()
         setup.script.text = setup_command
 
+        # Placement policy
+        # https://cloud.google.com/python/docs/reference/batch/latest/google.cloud.batch_v1.types.AllocationPolicy.PlacementPolicy
+
         # This will ensure all nodes finish first
         barrier = batch_v1.Runnable()
         barrier.barrier = batch_v1.Runnable.Barrier()
@@ -202,8 +218,15 @@ def run_job(self, job: JobExecutorInterface):
         group.permissive_ssh = True
 
         # This includes instances (machine type) boot disk and policy
+        # Also preemtion
         policy = self.get_allocation_policy(job)
 
+        # If we have preemption for the job and retry, update task retries with it
+        retries = self.workflow.remote_execution_settings.preemptible_retries
+        if self.is_preemptible(job) and retries:
+            self.logger.debug(f"Updating preemptible retries to {retries}")
+            task.max_retry_count = retries
+
         batchjob = batch_v1.Job()
         batchjob.task_groups = [group]
         batchjob.allocation_policy = policy
@@ -262,6 +285,12 @@ def get_allocation_policy(self, job):
         policy.machine_type = machine_type
         policy.boot_disk = boot_disk
 
+        # Do we want preemptible?
+        # https://github.com/googleapis/googleapis/blob/master/google/cloud/batch/v1/job.proto#L479 and  # noqa
+        # https://github.com/googleapis/google-cloud-python/blob/main/packages/google-cloud-batch/google/cloud/batch_v1/types/job.py#L672  # noqa
+        if self.is_preemptible(job):
+            policy.provisioning_model = 3
+
         instances = batch_v1.AllocationPolicy.InstancePolicyOrTemplate()
         instances.policy = policy
         allocation_policy.instances = [instances]