chore: import upstream snapshot with attribution

release/tune_tests/cloud_tests/tpl_aws_1x4.yaml

@@ -0,0 +1,19 @@
+cloud: {{env["ANYSCALE_CLOUD_NAME"]}}
+
+advanced_instance_config:
+  TagSpecifications:
+    - ResourceType: "instance"
+      Tags:
+        - Key: ttl-hours
+          Value: '72'
+
+  BlockDeviceMappings:
+    - DeviceName: /dev/sda1
+      Ebs:
+        VolumeSize: 300
+        DeleteOnTermination: true
+
+head_node:
+    instance_type: m5.xlarge
+
+worker_nodes: []

release/tune_tests/cloud_tests/tpl_gce_1x4.yaml

@@ -0,0 +1,8 @@
+cloud: {{env["ANYSCALE_CLOUD_NAME"]}}
+zones:
+    - us-west1-b
+
+head_node:
+    instance_type: n1-standard-4
+
+worker_nodes: []

release/tune_tests/cloud_tests/tpl_gce_4x2.yaml

@@ -0,0 +1,17 @@
+cloud_id: {{env["ANYSCALE_CLOUD_ID"]}}
+region: us-west1
+allowed_azs:
+    - us-west1-b
+
+max_workers: 3
+
+head_node_type:
+    name: head_node
+    instance_type: n1-standard-2
+
+worker_node_types:
+    - name: worker_node
+      instance_type: n1-standard-2
+      min_workers: 3
+      max_workers: 3
+      use_spot: false

release/tune_tests/cloud_tests/tpl_gcp_k8s_4x8.yaml

@@ -0,0 +1,15 @@
+cloud_id: cld_k8WcxPgjUtSE8RVmfZpTLuKM  # anyscale_k8s_gcp_cloud
+region: us-central1
+
+max_workers: 3
+
+head_node_type:
+    name: head_node
+    instance_type: n2-standard-8
+
+worker_node_types:
+    - name: worker_node
+      instance_type: n2-standard-8
+      min_workers: 3
+      max_workers: 3
+      use_spot: false

release/tune_tests/cloud_tests/workloads/long_running_cloud_storage.py

@@ -0,0 +1,94 @@
+import json
+import os
+import pickle
+import tempfile
+import time
+from collections import Counter
+
+import click
+import numpy as np
+
+from ray.tune import Checkpoint, CheckpointConfig, RunConfig, Callback, report, Tuner
+
+
+class ProgressCallback(Callback):
+    def __init__(self):
+        self.last_update = 0
+        self.update_interval = 60
+
+    def on_step_end(self, iteration, trials, **kwargs):
+        if time.time() - self.last_update > self.update_interval:
+            now = time.time()
+            result = {
+                "last_update": now,
+                "iteration": iteration,
+                "trial_states": dict(Counter([trial.status for trial in trials])),
+            }
+
+            path = "/tmp/release_test_out.json"
+            test_output_json_tmp = path + ".tmp"
+            with open(test_output_json_tmp, "wt") as f:
+                json.dump(result, f)
+            os.replace(test_output_json_tmp, path)
+            print(f"Wrote results to {test_output_json_tmp}")
+            print(json.dumps(result))
+
+            self.last_update = now
+
+
+def function_trainable(config):
+    sleep_time = config["sleep_time"]
+    score = config["score"]
+
+    checkpoint_num_files = config["checkpoint_num_files"]
+    checkpoint_iters = config["checkpoint_iters"]
+    checkpoint_size_b = config["checkpoint_size_b"] // checkpoint_num_files
+    checkpoint_num_items = checkpoint_size_b // 8  # np.float64
+
+    for i in range(int(10e12)):
+        metrics = {"score": i + score}
+        if (
+            checkpoint_iters >= 0
+            and checkpoint_size_b > 0
+            and i % checkpoint_iters == 0
+        ):
+            with tempfile.TemporaryDirectory() as directory:
+                for i in range(checkpoint_num_files):
+                    checkpoint_file = os.path.join(directory, f"bogus_{i:02d}.ckpt")
+                    checkpoint_data = np.random.uniform(0, 1, size=checkpoint_num_items)
+                    with open(checkpoint_file, "wb") as fp:
+                        pickle.dump(checkpoint_data, fp)
+
+                checkpoint = Checkpoint.from_directory(directory)
+                report(metrics, checkpoint=checkpoint)
+        else:
+            report(metrics)
+        time.sleep(sleep_time)
+
+
+@click.command()
+@click.argument("bucket", type=str)
+@click.option("--smoke-test", is_flag=True, default=False)
+def main(bucket, smoke_test):
+    # Note: smoke_test is ignored as we just adjust the timeout.
+    # The parameter is passed by the release test pipeline.
+    tuner = Tuner(
+        function_trainable,
+        param_space={
+            "sleep_time": 30,
+            "score": 0.5,
+            "checkpoint_num_files": 16,
+            "checkpoint_size_b": int(10**8),
+            "checkpoint_iters": 10,
+        },
+        run_config=RunConfig(
+            storage_path=bucket,
+            callbacks=[ProgressCallback()],
+            checkpoint_config=CheckpointConfig(num_to_keep=2),
+        ),
+    )
+    tuner.fit()
+
+
+if __name__ == "__main__":
+    main()