datatorch 0.6.4__tar.gz → 0.7.2__tar.gz

{datatorch-0.6.4 → datatorch-0.7.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datatorch
-Version: 0.6.4
+Version: 0.7.2
 Summary: A CLI and library for interacting with DataTorch.
 Home-page: https://github.com/datatorch/python
 Author: DataTorch
@@ -95,4 +95,4 @@ python3 -m venv env
 
 rm -rf env
 
-source env/bin/activate && pip3 install dist/datatorch-0.6.1-py3-none-any.whl && pip3 install pycocotools
+source env/bin/activate && pip3 install dist/datatorch-0.6.3-py3-none-any.whl && pip3 install pycocotools

{datatorch-0.6.4 → datatorch-0.7.2}/README.md

@@ -45,4 +45,4 @@ python3 -m venv env
 
 rm -rf env
 
-source env/bin/activate && pip3 install dist/datatorch-0.6.1-py3-none-any.whl && pip3 install pycocotools
+source env/bin/activate && pip3 install dist/datatorch-0.6.3-py3-none-any.whl && pip3 install pycocotools

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/agent/__init__.py

@@ -10,10 +10,9 @@ from .directory import agent_directory
 from .agent import Agent, tasks
 
 from gql import Client as GqlClient
-from gql.transport.exceptions import TransportClosed, TransportServerError
+from gql.transport.exceptions import TransportError, TransportQueryError
 from gql.transport.websockets import WebsocketsTransport
-from websockets.exceptions import InvalidMessage, InvalidURI
-from websockets import ConnectionClosedError
+from websockets.exceptions import ConnectionClosed, InvalidMessage, InvalidURI
 
 from datatorch.api import Client as DtClient
 from datatorch.utils.package import get_version
@@ -90,6 +89,22 @@ async def _exit_tasks() -> None:
         task.cancel()
 
 
+async def _restart_after_failure(transport, backoff_wait: float) -> float:
+    """Tears down jobs, transport and tasks, then sleeps with backoff.
+
+    Returns the next backoff wait."""
+    await _exit_jobs()
+    await _close_transport(transport)
+    await _exit_tasks()
+
+    backoff_wait = min(BACKOFF_MAX, backoff_wait * BACKOFF_FACTOR)
+    logger.debug(
+        f"Sleeping for {round(backoff_wait)} seconds and then attempting restart."
+    )
+    await asyncio.sleep(backoff_wait)
+    return backoff_wait
+
+
 async def start() -> None:
     """Creates and runs an agent."""
     setup_logging()
@@ -109,27 +124,36 @@ async def start() -> None:
                 backoff_wait = BACKOFF_INIT_WAIT
                 await Agent.run(session)
 
+        except (asyncio.CancelledError, InvalidURI):
+            break
+
         except (
-            ConnectionClosedError,
-            ConnectionRefusedError,
+            # ConnectionClosed covers both the error and the "OK" close
+            # (a server-side restart closes cleanly); OSError covers the
+            # socket-level failures seen right after a laptop wakes
+            # (DNS gaierror, ConnectionReset, ConnectionRefused, timeouts).
+            # TransportQueryError is what gql raises when the socket drops
+            # while a query is in flight ("Query completed without any
+            # answer received from the server"); it is NOT a TransportError
+            # subclass in the installed gql, so it is listed on its own.
+            ConnectionClosed,
+            OSError,
             IncompleteReadError,
-            TransportServerError,
-            TransportClosed,
+            TransportError,
+            TransportQueryError,
             InvalidMessage,
+            asyncio.TimeoutError,
         ) as e:
-            await _exit_jobs()
-            await _close_transport(transport)
-            await _exit_tasks()
-
             logger.error(e)
-            backoff_wait = min(BACKOFF_MAX, backoff_wait * BACKOFF_FACTOR)
-            logger.debug(
-                f"Sleeping for {round(backoff_wait)} seconds and then attempting restart."
-            )
-            await asyncio.sleep(backoff_wait)
+            backoff_wait = await _restart_after_failure(transport, backoff_wait)
 
-        except (asyncio.CancelledError, InvalidURI):
-            break
+        except Exception:
+            # Last resort: the loop must never die. On 2026-09-12 an
+            # uncaught exception here left the process alive for weeks
+            # with no agent loop, so the runner showed OFFLINE while its
+            # window looked fine.
+            logger.exception("Unexpected error in the agent loop; restarting.")
+            backoff_wait = await _restart_after_failure(transport, backoff_wait)
 
     logger.info("Exiting job processing task.")
 
@@ -139,12 +163,12 @@ async def stop() -> None:
 
     print(" ")
 
-    logger.warning("Gracefully exiting agent.")
+    logger.warning("Gracefully exiting runner.")
 
-    logger.info("Closing agent jobs.")
+    logger.info("Closing runner jobs.")
     await _exit_jobs()
 
-    logger.info("Closing agent jobs.")
+    logger.info("Closing runner jobs.")
     await _exit_jobs()
 
     logger.info("Closing all other tasks.")

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/agent/client.py

@@ -4,6 +4,9 @@ from .directory import agent_directory
 from gql.client import AsyncClientSession
 
 from typing_extensions import TypedDict
+import logging
+
+logger = logging.getLogger(__name__)
 
 
 class Log(TypedDict):
@@ -236,6 +239,17 @@ class AgentApiClient(object):
         # fmt: on
         return self.execute(mutate, params={"id": step_id, "logs": logs})
 
+    async def force_reconnect(self) -> None:
+        """Closes the websocket so the steps subscription raises and the
+        outer loop in `agent.start()` reconnects with backoff."""
+        transport = getattr(self.session, "transport", None)
+        if transport is None or getattr(transport, "websocket", None) is None:
+            return
+        try:
+            await transport.close()
+        except Exception as e:
+            logger.debug(f"Transport close raised while forcing reconnect: {e}")
+
     async def execute(self, query, *args, params: dict = {}, **kwargs) -> dict:
         """Wrapper around execute"""
         removed_none = dict((k, v) for k, v in params.items() if v is not None)

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/agent/monitoring.py

@@ -67,10 +67,17 @@ class AgentSystemStats(object):
         print_stats(stats)
         return stats
 
+    # Consecutive failed reports before the connection is considered dead.
+    # The steps subscription can sit silently on a half-open socket (e.g.
+    # after sleep); the report failing is the signal that reaches us.
+    MAX_REPORT_FAILURES = 3
+    REPORT_TIMEOUT = 30
+
     def __init__(self, agent, sample_rate=60):
         self.agent = agent
         self.sample_rate = sample_rate
         self.sample = 0
+        self.report_failures = 0
 
     async def start(self):
         try:
@@ -90,7 +97,25 @@ class AgentSystemStats(object):
         while True:
             self.sample += 1
             stats = self.stats()
-            loop = asyncio.get_running_loop()
             if self.sample != 1:
-                loop.create_task(self.agent.api.metrics(stats))
+                await self._report(stats)
             await asyncio.sleep(self.sample_rate)
+
+    async def _report(self, stats):
+        """Sends one metric report; repeated failures force a reconnect."""
+        try:
+            await asyncio.wait_for(
+                self.agent.api.metrics(stats), timeout=self.REPORT_TIMEOUT
+            )
+            self.report_failures = 0
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:
+            self.report_failures += 1
+            logger.warning(
+                f"Metric report failed ({self.report_failures}/{self.MAX_REPORT_FAILURES}): {e}"
+            )
+            if self.report_failures >= self.MAX_REPORT_FAILURES:
+                logger.error("Connection looks dead; forcing a reconnect.")
+                self.report_failures = 0
+                await self.agent.api.force_reconnect()

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/api/scripts/import_coco.py

@@ -133,10 +133,10 @@ def import_coco(
         api = ApiClient()
 
     _LOGGER.debug("Loading Project Information.")
-    if "/" in project_string:
-        project: Project = api.project(*project_string.split("/", 1))
-    else:
-        project: Project = api.project(project_string)
+    # api.project() accepts a single "login/slug" or UUID string and does
+    # its own splitting; the previous branch splatted the split into two
+    # positional args and crashed on every namespace/slug import.
+    project: Project = api.project(project_string)
 
     labels = project.labels()
     _LOGGER.debug("Project ID: %s", project.id)

datatorch-0.7.2/datatorch/cli/agent/__init__.py

@@ -0,0 +1,32 @@
+import click
+
+from .start import start
+from .create import create
+from .dir import dir
+
+
+@click.group(
+    name="runner",
+    help=(
+        "Commands for managing runners — self-hosted compute that "
+        "executes pipeline jobs. (Formerly 'agent'.)"
+    ),
+)
+def runner():
+    pass
+
+
+runner.add_command(start)
+runner.add_command(create)
+runner.add_command(dir)
+
+# Deprecated alias: `datatorch agent <cmd>` keeps working for existing
+# installs/scripts, sharing the exact command objects; hidden from
+# --help so new users learn `runner`. The wire protocol (agent token,
+# GraphQL mutations) and the on-disk agent directory are unchanged.
+agent = click.Group(
+    name="agent",
+    commands=runner.commands,
+    hidden=True,
+    help="Deprecated alias for 'runner'.",
+)

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/cli/agent/create.py

@@ -35,8 +35,10 @@ def create(ctx):
     agent_settings = agent_directory.settings
 
     if agent_settings.agent_id:
-        click.echo("An agent is already installed.")
-        confirmed = click.confirm("Would you like to create a new agent?", default=True)
+        click.echo("A runner is already installed.")
+        confirmed = click.confirm(
+            "Would you like to create a new runner?", default=True
+        )
         if not confirmed:
             return
 
@@ -48,8 +50,8 @@ def create(ctx):
     if not user_settings.api_key:
         ctx.invoke(login, host=user_settings.api_url)
 
-    name = click.prompt("Enter agents name", default=platform.node(), show_default=True)
-    spinner = Spinner("Creating agent")
+    name = click.prompt("Enter runner name", default=platform.node(), show_default=True)
+    spinner = Spinner("Creating runner")
 
     try:
         agent = create_agent(name)["createAgent"]
@@ -57,8 +59,8 @@ def create(ctx):
         agent_settings.agent_token = agent["token"]
         agent_settings.api_url = user_settings.api_url
     except Exception as ex:
-        spinner.done(click.style("Failed to new create agent.", fg="red"))
+        spinner.done(click.style("Failed to create runner.", fg="red"))
         click.echo(ex)
         return
 
-    spinner.done("Successfully created agent.")
+    spinner.done("Successfully created runner.")

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/cli/agent/dir.py

@@ -5,7 +5,7 @@ import subprocess
 from datatorch.agent import agent_directory
 
 
-@click.command(help="Opens agent directory where settings and files are stored.")
+@click.command(help="Opens the runner directory where settings and files are stored.")
 def dir():
     path = agent_directory.dir
     if platform.system() == "Windows":

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/cli/agent/start.py

@@ -15,7 +15,7 @@ def add_signal_handlers(loop: AbstractEventLoop):
         pass  # Ignore if not implemented. Means this program is running in windows.
 
 
-@click.command(help="Run an agent")
+@click.command(help="Start the runner (executes pipeline jobs on this machine)")
 def start():
     loop = asyncio.new_event_loop()
     asyncio.set_event_loop(loop)

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch/cli/groups.py

@@ -9,7 +9,7 @@ from .main.pull import pull
 from .main.upgrade import package_upgrade
 from .main.version import version
 
-from .agent import agent
+from .agent import agent, runner
 from .pipeline import pipeline
 from .action import action
 from .import_cmds import import_cmd
@@ -29,6 +29,7 @@ main.add_command(version)
 main.add_command(package_upgrade)
 
 main.add_command(pipeline)
+main.add_command(runner)
 main.add_command(agent)
 main.add_command(action)
 main.add_command(import_cmd)

{datatorch-0.6.4 → datatorch-0.7.2}/datatorch.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datatorch
-Version: 0.6.4
+Version: 0.7.2
 Summary: A CLI and library for interacting with DataTorch.
 Home-page: https://github.com/datatorch/python
 Author: DataTorch
@@ -95,4 +95,4 @@ python3 -m venv env
 
 rm -rf env
 
-source env/bin/activate && pip3 install dist/datatorch-0.6.1-py3-none-any.whl && pip3 install pycocotools
+source env/bin/activate && pip3 install dist/datatorch-0.6.3-py3-none-any.whl && pip3 install pycocotools

{datatorch-0.6.4 → datatorch-0.7.2}/setup.py

@@ -33,7 +33,7 @@ requirements = [
 
 setup(
     name="datatorch",
-    version="0.6.4",
+    version="0.7.2",
     description="A CLI and library for interacting with DataTorch.",
     author="DataTorch",
     author_email="support@datatorch.io",

datatorch-0.6.4/datatorch/cli/agent/__init__.py

@@ -1,15 +0,0 @@
-import click
-
-from .start import start
-from .create import create
-from .dir import dir
-
-
-@click.group(help="Commands for managing agents.")
-def agent():
-    pass
-
-
-agent.add_command(start)
-agent.add_command(create)
-agent.add_command(dir)