Chore/pull in upstream

.github/workflows/nightly-release.yml

@@ -40,7 +40,7 @@ jobs:
       - name: setup python
         uses: actions/setup-python@v2
         with:
-          python-version: '3.7'
+          python-version: '3.8'
           architecture: x64
 
       - name: install deps

.github/workflows/pull-request.yml

@@ -24,14 +24,14 @@ on:
       - reopened
       - edited
 jobs:
-  validate-naming-convention:
-    name: Pull Request's title matches naming convention
-    runs-on: ubuntu-latest
-    steps:
-      - uses: deepakputhraya/action-pr-title@master
-        with:
-          regex: '^\[(?:feat|fix|doc|refactor|deprecation)\]\s[A-Z].*(?<!\.)$'
-          github_token: ${{ github.token }}
+#  validate-naming-convention:
+#    name: Pull Request's title matches naming convention
+#    runs-on: ubuntu-latest
+#    steps:
+#      - uses: deepakputhraya/action-pr-title@master
+#        with:
+#          regex: '^\[(?:feat|fix|doc|refactor|deprecation)\]\s[A-Z].*(?<!\.)$'
+#          github_token: ${{ github.token }}
   run-checks:
     if: github.event.pull_request.draft == false && github.event.action != 'edited'
     runs-on: ubuntu-latest
@@ -68,8 +68,8 @@ jobs:
 
   storage-performance-checks:
     needs: run-checks
-    concurrency: perf-tests
-    runs-on: [self-hosted, performance-tests]
+    concurrency: storage-performance-checks
+    runs-on: [self-hosted, perf-tests]
     name: Performance tests
     steps:
       - name: checkout

.github/workflows/python-package.yml

@@ -96,6 +96,7 @@ jobs:
           python -m pip install -r requirements.txt
 
       - name: Build bdist wheels for 'cp37-cp37m'
+        if: matrix.manylinux-version == 'manylinux_2_24_x86_64'
         uses: nick-fields/retry@v2
         with:
           max_attempts: 3

CHANGELOG.md

@@ -1,5 +1,44 @@
 # Changelog
 
+## Unreleased:
+
+### Fixes: 
+- Fix issues with tag false reassignment (mihran113)
+
+## 3.29.1 May 8, 2025:
+
+### Enhancements:
+- Constant indexing of in-progress runs (alberttorosyan)
+- Fallback to union view if index db is missing (alberttorosyan, mihran113)
+
+
+### Fixes:
+- Fix min/max calculation for single point metrics (mihran113)
+- Aim web ui integration in jupyter/colab (larissapoghosyan)
+- Package publishing for Linux/Python 3.7 (alberttorosyan)
+
+## 3.29.0 May 8, 2025 (Yanked)
+
+## 3.28.0 Mar 21, 2025
+
+### Enhancements:
+- Skip metrics check when run is known to yield false result (alberttorosyan)
+- Remove metric version check to improve performance of metric retrieval (mihran113)
+- Move indexing thread to main process of `aim up` (alberttorosyan)
+- Add AimCallback implementation for hugging face distributed runs (VassilisVassiliadis)
+- Add py.typed marker to allow usage of existing type annotations (bluenote10)
+
+
+### Fixes:
+- Decrease client resources keep-alive time (mihran113)
+- Fix connection of data points on epoch alignment (mihran113)
+- Resolve issue with adding duplicate tags to the same run (mihran113)
+- Improve error messages for remote tracking server (mihran113)
+- Fix spurious assertion error in message stream parsing (qzed)
+- Correct indentation on query proxy object return statement (alberttorosyan)
+- Fix typing issues in S3ArtifactStorage implementation (sbatchelder)
+
+
 ## 3.27.0 Dec 18, 2024
 
 ### Enhancements:

aim/VERSION

@@ -1 +1 @@
-3.27.0
+3.29.1

aim/acme.py

@@ -1,2 +1,3 @@
 # Alias to SDK acme interface
-from aim.sdk.adapters.acme import AimCallback, AimWriter  # noqa F401
+from aim.sdk.adapters.acme import AimCallback as AimCallback
+from aim.sdk.adapters.acme import AimWriter as AimWriter

aim/cli/convert/commands.py

@@ -40,7 +40,7 @@ def convert_tensorboard(ctx, logdir, flat, no_cache):
 @click.option('--flat', '-f', required=False, is_flag=True, default=False)
 def convert_tensorflow(ctx, logdir, flat):
     click.secho(
-        "WARN: Command 'tf' is deprecated and will be removed in future releases," " please use 'tensorboard' instead.",
+        "WARN: Command 'tf' is deprecated and will be removed in future releases, please use 'tensorboard' instead.",
         fg='red',
     )
     repo_inst = ctx.obj['repo_inst']

aim/cli/init/commands.py

@@ -20,15 +20,15 @@ def init(repo, yes, skip_if_exists):
     re_init = False
     if Repo.exists(repo_path):
         if yes and skip_if_exists:
-            raise click.BadParameter('Conflicting init options.' 'Either specify -y/--yes or -s/--skip-if-exists')
+            raise click.BadParameter('Conflicting init options.Either specify -y/--yes or -s/--skip-if-exists')
         elif yes:
             re_init = True
         elif skip_if_exists:
             click.echo('Repo exists at {}. Skipped initialization.'.format(repo_path))
             return
         else:
             re_init = click.confirm(
-                'Aim repository is already initialized. ' 'Do you want to re-initialize to empty Aim repository?'
+                'Aim repository is already initialized. Do you want to re-initialize to empty Aim repository?'
             )
         if not re_init:
             return

aim/cli/manager/manager.py

@@ -33,18 +33,18 @@ def check_startup_success():
         import requests
 
         server_path = 'http://{}:{}{}'.format(args['--host'], args['--port'], args['--base-path'])
-        status_api = f'{server_path}/api/projects/status'
-        retry_count = 5
-        sleep_interval = 1
+        status_api = f'{server_path}/api/projects/'
+        retry_count = 10
+        sleep_interval = 0.1
         for _ in range(retry_count):
+            time.sleep(sleep_interval)
+            sleep_interval *= 2
             try:
                 response = requests.get(status_api)
                 if response.status_code == 200:
                     return True
             except Exception:
                 pass
-            sleep_interval += 1
-            time.sleep(sleep_interval)
 
         return False
 

aim/cli/runs/commands.py

@@ -195,7 +195,7 @@ def update_metrics(ctx, yes):
     index_manager = RepoIndexManager.get_index_manager(repo)
     hashes = repo.list_all_runs()
     for run_hash in tqdm.tqdm(hashes, desc='Updating runs', total=len(hashes)):
-        meta_tree = repo.request_tree('meta', run_hash, read_only=False, from_union=False)
+        meta_tree = repo.request_tree('meta', run_hash, read_only=False)
         meta_run_tree = meta_tree.subtree(('meta', 'chunks', run_hash))
         try:
             # check if the Run has already been updated.

aim/cli/runs/utils.py

@@ -48,7 +48,7 @@ def upload_repo_runs(buffer: io.BytesIO, bucket_name: str) -> Tuple[bool, str]:
         import boto3
     except ImportError:
         raise RuntimeError(
-            "This command requires 'boto3' to be installed. " 'Please install it with command: \n pip install boto3'
+            "This command requires 'boto3' to be installed. Please install it with command: \n pip install boto3"
         )
 
     try:

aim/cli/server/commands.py

@@ -95,5 +95,5 @@ def server(host, port, repo, ssl_keyfile, ssl_certfile, base_path, log_level, de
         )
         exec_cmd(cmd, stream_output=True)
     except ShellCommandException:
-        click.echo('Failed to run Aim Tracking Server. ' 'Please see the logs above for details.')
+        click.echo('Failed to run Aim Tracking Server. Please see the logs above for details.')
         exit(1)

aim/cli/storage/commands.py

@@ -56,7 +56,7 @@ def to_3_11(ctx, hashes, yes):
         try:
             run = Run(run_hash, repo=repo)
             if run.check_metrics_version():
-                backup_run(run)
+                backup_run(repo, run.hash)
                 run.update_metrics()
                 index_manager.index(run_hash)
             else:

aim/cli/up/commands.py

@@ -11,7 +11,9 @@
     get_repo_instance,
     set_log_level,
 )
+from aim.sdk.index_manager import RepoIndexManager
 from aim.sdk.repo import Repo
+from aim.sdk.run_status_manager import RunStatusManager
 from aim.sdk.utils import clean_repo_path
 from aim.web.configs import (
     AIM_ENV_MODE_KEY,
@@ -29,7 +31,7 @@
 @click.command()
 @click.option('-h', '--host', default=AIM_UI_DEFAULT_HOST, type=str)
 @click.option('-p', '--port', default=AIM_UI_DEFAULT_PORT, type=int)
-@click.option('-w', '--workers', default=1, type=int)
+@click.option('-w', '--workers', default=2, type=int)
 @click.option('--uds', required=False, type=click.Path(exists=False, file_okay=True, dir_okay=False, readable=True))
 @click.option('--repo', required=False, type=click.Path(exists=True, file_okay=False, dir_okay=True, writable=True))
 @click.option('--tf_logs', type=click.Path(exists=True, readable=True))
@@ -96,7 +98,7 @@ def up(
         db_cmd = build_db_upgrade_command()
         exec_cmd(db_cmd, stream_output=True)
     except ShellCommandException:
-        click.echo('Failed to initialize Aim DB. ' 'Please see the logs above for details.')
+        click.echo('Failed to initialize Aim DB. Please see the logs above for details.')
         return
 
     if port == 0:
@@ -122,6 +124,11 @@ def up(
     if profiler:
         os.environ[AIM_PROFILER_KEY] = '1'
 
+    index_mng = RepoIndexManager.get_index_manager(repo_inst)
+    index_mng.start()
+
+    run_status_mng = RunStatusManager(repo_inst)
+    run_status_mng.start()
     try:
         server_cmd = build_uvicorn_command(
             'aim.web.run:app',

aim/distributed_hugging_face.py

@@ -0,0 +1,2 @@
+# Alias to SDK distributed hugging face interface
+from aim.sdk.adapters.distributed_hugging_face import AimCallback  # noqa: F401

aim/ext/notifier/notifier.py

@@ -34,7 +34,7 @@ def notify(self, message: Optional[str] = None, **kwargs):
                 except Exception as e:
                     attempt += 1
                     if attempt == self.MAX_RETRIES:
-                        logger.error(f'Notifier {sub} failed to send message "{message}". ' f'No retries left.')
+                        logger.error(f'Notifier {sub} failed to send message "{message}". No retries left.')
                         raise NotificationSendError(e)
                     else:
                         logger.error(

aim/ext/sshfs/utils.py

@@ -197,7 +197,7 @@ def unmount_remote_repo(mount_point: str, mount_root: str):
     if exit_code != 0:
         # in case of failure log warning so the user can unmount manually if needed
         logger.warning(
-            f'Could not unmount path: {mount_point}.\n' f'Please unmount manually using command:\n' f'{" ".join(cmd)}'
+            f'Could not unmount path: {mount_point}.\nPlease unmount manually using command:\n{" ".join(cmd)}'
         )
     else:
         shutil.rmtree(mount_root)

aim/ext/tensorboard_tracker/tracker.py

@@ -31,7 +31,7 @@ def _decode_histogram(value):
 
     # This is a bit weird but it seems the histogram counts is usually padded by 0 as tensorboard
     # only stores the right limits?
-    # See https://github.com/pytorch/pytorch/blob/7d2a18da0b3427fcbe44b461a0aa508194535885/torch/utils/tensorboard/summary.py#L390 # noqa
+    # See https://github.com/pytorch/pytorch/blob/7d2a18da0b3427fcbe44b461a0aa508194535885/torch/utils/tensorboard/summary.py#L390
     bin_counts = bin_counts[1:]
 
     bin_range = (bucket_limits[0], bucket_limits[-1])

aim/ext/transport/handlers.py

@@ -51,14 +51,12 @@ def get_tree(**kwargs):
     name = kwargs['name']
     sub = kwargs['sub']
     read_only = kwargs['read_only']
-    from_union = kwargs['from_union']
     index = kwargs['index']
     timeout = kwargs['timeout']
-    no_cache = kwargs.get('no_cache', False)
     if index:
         return ResourceRef(repo._get_index_tree(name, timeout))
     else:
-        return ResourceRef(repo.request_tree(name, sub, read_only=read_only, from_union=from_union, no_cache=no_cache))
+        return ResourceRef(repo.request_tree(name, sub, read_only=read_only))
 
 
 def get_structured_run(hash_, read_only, created_at, **kwargs):

aim/ext/transport/heartbeat.py

@@ -15,10 +15,8 @@ class HeartbeatSender(object):
     HEARTBEAT_INTERVAL_DEFAULT = 10
     NETWORK_CHECK_INTERVAL = 180
 
-    NETWORK_UNSTABLE_WARNING_TEMPLATE = (
-        'Network connection between client `{}` ' 'and server `{}` appears to be unstable.'
-    )
-    NETWORK_ABSENT_WARNING_TEMPLATE = 'Network connection between client `{}` ' 'and server `{}` appears to be absent.'
+    NETWORK_UNSTABLE_WARNING_TEMPLATE = 'Network connection between client `{}` and server `{}` appears to be unstable.'
+    NETWORK_ABSENT_WARNING_TEMPLATE = 'Network connection between client `{}` and server `{}` appears to be absent.'
 
     def __init__(
         self,
@@ -118,7 +116,7 @@ def reset_responses():
 
 
 class HeartbeatWatcher:
-    CLIENT_KEEP_ALIVE_TIME_DEFAULT = 30 * 60  # 30 minutes
+    CLIENT_KEEP_ALIVE_TIME_DEFAULT = 5 * 60  # 5 minutes
 
     def __init__(self, heartbeat_pool, keep_alive_time: Union[int, float] = CLIENT_KEEP_ALIVE_TIME_DEFAULT):
         self._heartbeat_pool = heartbeat_pool

aim/ext/transport/message_utils.py

@@ -5,7 +5,8 @@
 from typing import Iterator, Tuple
 
 from aim.storage.object import CustomObject
-from aim.storage.treeutils import decode_tree, encode_tree  # noqa
+from aim.storage.treeutils import decode_tree as decode_tree
+from aim.storage.treeutils import encode_tree as encode_tree
 from aim.storage.types import BLOB
 
 
@@ -45,28 +46,23 @@ def pack_stream(tree: Iterator[Tuple[bytes, bytes]]) -> bytes:
             yield struct.pack('I', len(key)) + key + struct.pack('?', True) + struct.pack('I', len(val)) + val
 
 
-def unpack_helper(msg: bytes) -> Tuple[bytes, bytes]:
-    (key_size,), tail = struct.unpack('I', msg[:4]), msg[4:]
-    key, tail = tail[:key_size], tail[key_size:]
-    (is_blob,), tail = struct.unpack('?', tail[:1]), tail[1:]
-    (value_size,), tail = struct.unpack('I', tail[:4]), tail[4:]
-    value, tail = tail[:value_size], tail[value_size:]
-    assert len(tail) == 0
-    if is_blob:
-        yield key, BLOB(data=value)
-    else:
-        yield key, value
-
-
 def unpack_stream(stream) -> Tuple[bytes, bytes]:
     for msg in stream:
-        yield from unpack_helper(msg)
+        yield from unpack_args(msg)
 
 
 def raise_exception(server_exception):
+    from filelock import Timeout
+
     module = importlib.import_module(server_exception.get('module_name'))
     exception = getattr(module, server_exception.get('class_name'))
     args = json.loads(server_exception.get('args') or [])
+    message = server_exception.get('message')
+
+    # special handling for lock timeouts as they require lock argument which can't be passed over the network
+    if exception == Timeout:
+        raise Exception(message)
+
     raise exception(*args) if args else exception()
 
 
@@ -75,6 +71,7 @@ def build_exception(exception: Exception):
         'module_name': exception.__class__.__module__,
         'class_name': exception.__class__.__name__,
         'args': json.dumps(exception.args),
+        'message': str(exception),
     }
 
 

aim/ext/transport/utils.py

@@ -13,7 +13,7 @@ def inner(func):
         def wrapper(*args, **kwargs):
             try:
                 return func(*args, **kwargs)
-            except exc_type as e:  # noqa
+            except exc_type:
                 if error_message is not None:
                     logger.error(error_message)
                     raise RuntimeError(error_message)

aim/fastai.py

@@ -1,2 +1,2 @@
 # Alias to SDK fast.ai interface
-from aim.sdk.adapters.fastai import AimCallback  # noqa F401
+from aim.sdk.adapters.fastai import AimCallback as AimCallback

aim/hf_dataset.py

@@ -1,2 +1,2 @@
 # Alias to SDK Hugging Face Datasets interface
-from aim.sdk.objects.plugins.hf_datasets_metadata import HFDataset  # noqa F401
+from aim.sdk.objects.plugins.hf_datasets_metadata import HFDataset as HFDataset

aim/hugging_face.py

@@ -1,2 +1,2 @@
 # Alias to SDK Hugging Face interface
-from aim.sdk.adapters.hugging_face import AimCallback  # noqa F401
+from aim.sdk.adapters.hugging_face import AimCallback as AimCallback

aim/keras.py

@@ -1,2 +1,3 @@
 # Alias to SDK Keras interface
-from aim.sdk.adapters.keras import AimCallback, AimTracker  # noqa F401
+from aim.sdk.adapters.keras import AimCallback as AimCallback
+from aim.sdk.adapters.keras import AimTracker as AimTracker

aim/keras_tuner.py

@@ -1,2 +1,2 @@
 # Alias to SDK Keras-Tuner interface
-from aim.sdk.adapters.keras_tuner import AimCallback  # noqa F401
+from aim.sdk.adapters.keras_tuner import AimCallback as AimCallback

aim/mxnet.py

@@ -1,2 +1,2 @@
 # Alias to SDK mxnet interface
-from aim.sdk.adapters.mxnet import AimLoggingHandler  # noqa F401
+from aim.sdk.adapters.mxnet import AimLoggingHandler as AimLoggingHandler

aim/optuna.py

@@ -1,2 +1,2 @@
 # Alias to SDK Optuna interface
-from aim.sdk.adapters.optuna import AimCallback  # noqa F401
+from aim.sdk.adapters.optuna import AimCallback as AimCallback

aim/paddle.py

@@ -1,2 +1,2 @@
 # Alias to SDK PaddlePaddle interface
-from aim.sdk.adapters.paddle import AimCallback  # noqa F401
+from aim.sdk.adapters.paddle import AimCallback as AimCallback

aim/prophet.py

@@ -1,2 +1,2 @@
 # Alias to SDK Prophet interface
-from aim.sdk.adapters.prophet import AimLogger  # noqa F401
+from aim.sdk.adapters.prophet import AimLogger as AimLogger