Add iteration time test as part of regression testing

regression/cloud_providers.yaml

@@ -6,17 +6,19 @@ command: |-
   pip uninstall -y mosaicml-streaming
   cd streaming
   pip install -e '.[dev]'
-  composer -n 1 regression/synthetic_dataset.py --create --cloud_url $GS_CLOUD_URL --size_limit 4194304 # 4mb
-  composer -n 2 regression/iterate_data.py --cloud_url $GS_CLOUD_URL --check_download
-  composer -n 1 regression/synthetic_dataset.py --delete --cloud_url $GS_CLOUD_URL
-  composer -n 1 regression/synthetic_dataset.py --create --cloud_url $S3_URL --size_limit 134217728 # 128 mb
-  composer -n 4 regression/iterate_data.py --cloud_url $S3_URL --check_download --local
-  composer -n 1 regression/synthetic_dataset.py --delete --cloud_url $S3_URL
-  composer -n 1 regression/synthetic_dataset.py --create --cloud_url $OCI_URL --size_limit 268435456 # 256 mb
-  composer -n 8 regression/iterate_data.py --cloud_url $OCI_URL --check_download
-  composer -n 1 regression/synthetic_dataset.py --delete --cloud_url $OCI_URL
+  python regression/synthetic_dataset.py --create --name imagedataset --out $GCS_URL --size_limit 4194304 # 4mb
+  composer -n 2 regression/iterate_data.py --remote $GCS_URL --validate-files
+  python regression/synthetic_dataset.py --delete --out $GCS_URL
 
-image: mosaicml/composer:0.15.0
+  python regression/synthetic_dataset.py --create --name imagedataset --out $S3_URL --num_samples 500 --size_limit 134217728 # 128 mb
+  composer -n 4 regression/iterate_data.py --remote $S3_URL --local /tmp/local_dataset/ --validate-files
+  python regression/synthetic_dataset.py --delete --out $S3_URL
+
+  python regression/synthetic_dataset.py --create --name imagedataset --out $OCI_URL --num_samples 500 --size_limit 268435456 # 256 mb
+  composer -n 8 regression/iterate_data.py --remote $OCI_URL --validate-files
+  python regression/synthetic_dataset.py --delete --out $OCI_URL
+
+image: mosaicml/composer:latest
 scheduling:
   resumable: true
   priority: medium

regression/iterate_data.py

@@ -4,29 +4,16 @@
 """Create a streaming dataset from toy data with various options for regression testing."""
 
 import os
-import tempfile
+import time
 import urllib.parse
 from argparse import ArgumentParser, Namespace
+from typing import Union
 
-import utils
 from torch.utils.data import DataLoader
+from utils import get_dataloader_params, get_kwargs, get_streaming_dataset_params
 
 from streaming import StreamingDataset
 
-_TRAIN_EPOCHS = 2
-
-
-def get_kwargs(key: str) -> str:
-    """Parse key of named command-line arguments.
-
-    Returns:
-        str: Key of named arguments.
-    """
-    if key.startswith('--'):
-        key = key[2:]
-    key = key.replace('-', '_')
-    return key
-
 
 def parse_args() -> tuple[Namespace, dict[str, str]]:
     """Parse command-line arguments.
@@ -35,38 +22,33 @@ def parse_args() -> tuple[Namespace, dict[str, str]]:
         tuple(Namespace, dict[str, str]): Command-line arguments and named arguments.
     """
     args = ArgumentParser()
-    args.add_argument('--cloud_url', type=str)
-    args.add_argument('--check_download', default=False, action='store_true')
-    args.add_argument('--local', default=False, action='store_true')
-    args.add_argument(
-        '--keep_zip',
-        default=False,
-        action='store_true',
-        help=('Whether to keep or delete the compressed form when decompressing'
-              ' downloaded shards for StreamingDataset.'),
-    )
-    args.add_argument(
-        '--shuffle',
-        default=False,
-        action='store_true',
-        help=('Whether to iterate over the samples in randomized order for'
-              ' StreamingDataset.'),
-    )
+    args.add_argument('--epochs',
+                      type=int,
+                      default=2,
+                      help='Number of epochs to iterate over the data')
+    args.add_argument('--validate-files', default=False, action='store_true', help='Verify files')
+    args.add_argument('--validate-iter-time',
+                      default=False,
+                      action='store_true',
+                      help='Test iter time')
 
     args, runtime_args = args.parse_known_args()
     kwargs = {get_kwargs(k): v for k, v in zip(runtime_args[::2], runtime_args[1::2])}
     return args, kwargs
 
 
-def get_file_count(cloud_url: str) -> int:
+def get_file_count(remote: str) -> Union[int, None]:
     """Get the number of files in a remote directory.
 
     Args:
-        cloud_url (str): Cloud provider url.
+        remote (str): Remote directory URL.
     """
-    obj = urllib.parse.urlparse(cloud_url)
+    obj = urllib.parse.urlparse(remote)
     cloud = obj.scheme
     files = []
+    if cloud == '':
+        return len(
+            [name for name in os.listdir(remote) if os.path.isfile(os.path.join(remote, name))])
     if cloud == 'gs':
         from google.cloud.storage import Bucket, Client
 
@@ -75,12 +57,14 @@ def get_file_count(cloud_url: str) -> int:
 
         bucket = Bucket(gcs_client, obj.netloc)
         files = bucket.list_blobs(prefix=obj.path.lstrip('/'))
+        return sum(1 for f in files if f.key[-1] != '/')
     elif cloud == 's3':
         import boto3
 
         s3 = boto3.resource('s3')
         bucket = s3.Bucket(obj.netloc)
         files = bucket.objects.filter(Prefix=obj.path.lstrip('/'))
+        return sum(1 for f in files if f.key[-1] != '/')
     elif cloud == 'oci':
         import oci
 
@@ -91,8 +75,9 @@ def get_file_count(cloud_url: str) -> int:
         objects = oci_client.list_objects(namespace, obj.netloc, prefix=obj.path.lstrip('/'))
 
         files = objects.data.objects
-
-    return sum(1 for _ in files)
+        return sum(1 for _ in files)
+    else:
+        raise ValueError(f'Unsupported remote directory prefix {cloud} in {remote}')
 
 
 def main(args: Namespace, kwargs: dict[str, str]) -> None:
@@ -102,41 +87,31 @@ def main(args: Namespace, kwargs: dict[str, str]) -> None:
         args (Namespace): Command-line arguments.
         kwargs (dict): Named arguments.
     """
-    tmp_dir = tempfile.gettempdir()
-    tmp_download_dir = os.path.join(tmp_dir, 'test_iterate_data_download')
-    dataset = StreamingDataset(
-        remote=args.cloud_url if args.cloud_url is not None else utils.get_local_remote_dir(),
-        local=tmp_download_dir if args.local else None,
-        split=kwargs.get('split'),
-        download_retry=int(kwargs.get('download_retry', 2)),
-        download_timeout=float(kwargs.get('download_timeout', 60)),
-        validate_hash=kwargs.get('validate_hash'),
-        keep_zip=args.keep_zip,
-        epoch_size=int(kwargs['epoch_size']) if 'epoch_size' in kwargs else None,
-        predownload=int(kwargs['predownload']) if 'predownload' in kwargs else None,
-        cache_limit=kwargs.get('cache_limit'),
-        partition_algo=kwargs.get('partition_algo', 'orig'),
-        num_canonical_nodes=int(kwargs['num_canonical_nodes'])
-        if 'num_canonical_nodes' in kwargs else None,
-        batch_size=int(kwargs['batch_size']) if 'batch_size' in kwargs else None,
-        shuffle=args.shuffle,
-        shuffle_algo=kwargs.get('shuffle_algo', 'py1s'),
-        shuffle_seed=int(kwargs.get('shuffle_seed', 9176)),
-        shuffle_block_size=int(kwargs.get('shuffle_block_size', 1 << 18)),
-    )
-
-    dataloader = DataLoader(dataset)
-    for _ in range(_TRAIN_EPOCHS):
+    dataset_params = get_streaming_dataset_params(kwargs)
+    dataloader_params = get_dataloader_params(kwargs)
+    dataset = StreamingDataset(**dataset_params)
+    dataloader = DataLoader(dataset=dataset, **dataloader_params)
+    iter_time = []
+    start_time = 0
+    for epoch in range(args.epochs):
+        if epoch > 0:
+            start_time = time.time()
         for _ in dataloader:
             pass
+        if epoch > 0:
+            iter_time.append(time.time() - start_time)
 
-    if args.check_download and args.cloud_url is not None:
-        num_cloud_files = get_file_count(args.cloud_url)
+    if args.validate_files and dataset.streams[0].remote is not None:
+        num_remote_files = get_file_count(dataset.streams[0].remote)
         local_dir = dataset.streams[0].local
         num_local_files = len([
             name for name in os.listdir(local_dir) if os.path.isfile(os.path.join(local_dir, name))
         ])
-        assert num_cloud_files == num_local_files
+        assert num_remote_files == num_local_files, f'Expected {num_remote_files} files, got {num_local_files}'
+
+    # TODO: Assert the iteration time is within a certain threshold
+    if args.validate_iter_time:
+        print(f'Average iter time: {sum(iter_time) / len(iter_time)} secs.')
 
 
 if __name__ == '__main__':

regression/iterate_data.yaml

@@ -6,25 +6,41 @@ command: |-
   pip uninstall -y mosaicml-streaming
   cd streaming
   pip install -e '.[dev]'
-  composer -n 1 regression/synthetic_dataset.py --create --compression gz
-  composer -n 2 regression/iterate_data.py
-  composer -n 1 regression/synthetic_dataset.py --delete
-  composer -n 1 regression/synthetic_dataset.py --create --hashes sha1 xxh128
-  composer -n 2 regression/iterate_data.py
-  composer -n 4 regression/iterate_data.py --local
-  composer -n 8 regression/iterate_data.py --download_retry 10
-  composer -n 8 regression/iterate_data.py --download_timeout 120
-  composer -n 8 regression/iterate_data.py --validate_hash sha1
-  composer -n 8 regression/iterate_data.py --validate_hash xxh128
-  composer -n 8 regression/iterate_data.py --keep_zip
-  composer -n 8 regression/iterate_data.py --predownload 1000
-  composer -n 8 regression/iterate_data.py --cache_limit 64mb
-  composer -n 8 regression/iterate_data.py --num_canonical_nodes 16
-  composer -n 8 regression/iterate_data.py --batch_size 1000
-  composer -n 8 regression/iterate_data.py --shuffle --shuffle_algo py1b --shuffle_seed 12 --shuffle_block_size 10000
-  composer -n 1 regression/synthetic_dataset.py --delete
-
-image: mosaicml/composer:0.15.0
+  python regression/synthetic_dataset.py --create --name numberandsaydataset --out /tmp/streaming_dataset/ --num_samples 10_000 --size_limit 10240
+  composer -n 2 regression/iterate_data.py --remote /tmp/streaming_dataset/ --batch_size 16
+  composer -n 2 regression/iterate_data.py --remote /tmp/streaming_dataset/ --local /tmp/local_dataset/ --batch_size 32
+  composer -n 4 regression/iterate_data.py --local /tmp/streaming_dataset/
+  python regression/synthetic_dataset.py --delete --out /tmp/streaming_dataset/
+  rm -rf /tmp/local_dataset/
+
+  python regression/synthetic_dataset.py --create --name numberandsaydataset --out /tmp/streaming_dataset_gz/ --hashes sha1,xxh128 --compression gz --num_samples 10_000 --size_limit 10240
+  composer -n 2 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --batch_size 4
+  composer -n 2 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --batch_size 1
+  composer -n 4 regression/iterate_data.py --local /tmp/streaming_dataset_gz/
+  rm -rf /tmp/local_dataset/
+
+  composer -n 8 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --download_retry 4
+  rm -rf /tmp/local_dataset/
+
+  composer -n 8 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --download_timeout 120
+  rm -rf /tmp/local_dataset/
+
+  composer -n 8 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --validate_hash sha1
+  rm -rf /tmp/local_dataset/
+
+  composer -n 8 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --keep_zip
+  rm -rf /tmp/local_dataset/
+
+  composer -n 8 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --predownload 1000
+  rm -rf /tmp/local_dataset/
+
+  composer -n 8 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --num_canonical_nodes 16
+  rm -rf /tmp/local_dataset/
+
+  composer -n 8 regression/iterate_data.py --remote /tmp/streaming_dataset_gz/ --local /tmp/local_dataset/ --shuffle_algo py1b --shuffle_seed 12 --shuffle_block_size 10000
+  rm -rf /tmp/local_dataset/
+
+image: mosaicml/composer:latest
 scheduling:
   resumable: true
   priority: medium

regression/iteration_time.yaml

@@ -0,0 +1,22 @@
+name: streaming-regression-test-iteration-time
+compute:
+  gpus: 8 # Number of GPUs to use
+  # cluster: TODO # Name of the cluster to use for this run
+command: |-
+  pip uninstall -y mosaicml-streaming
+  cd streaming
+  pip install -e '.[dev]'
+  python regression/synthetic_dataset.py --create --name imagedataset --out /tmp/streaming_dataset/ --num_samples 300
+  composer -n 2 regression/iterate_data.py --remote /tmp/streaming_dataset/ --local /tmp/local_dataset/ --batch_size 32
+  python regression/synthetic_dataset.py --delete --out /tmp/streaming_dataset/
+  rm -rf /tmp/local_dataset/
+
+image: mosaicml/composer:latest
+scheduling:
+  resumable: true
+  priority: medium
+integrations:
+  - integration_type: git_repo
+    git_repo: mosaicml/streaming
+    git_branch: main
+    ssh_clone: false