ellf-cli 10.0.1__tar.gz → 10.0.3__tar.gz

{ellf_cli-10.0.1/ellf_cli.egg-info → ellf_cli-10.0.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ellf-cli
-Version: 10.0.1
+Version: 10.0.3
 Summary: Ellf Command Line Interface
 Home-page: https://prodi.gy
 Author: ExplosionAI GmbH

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/about.json

@@ -1,7 +1,7 @@
 {
   "title": "Ellf CLI",
   "name": "ellf-cli",
-  "version": "10.0.1",
+  "version": "10.0.3",
   "summary": "Ellf Command Line Interface",
   "uri": "https://prodi.gy",
   "prog": "ellf",

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/commands/actions.py

@@ -9,7 +9,13 @@ from wasabi import msg
 from ellf_pam_sdk.models import ActionDetail, ActionSummary
 
 from ..cli import cli
-from ..errors import BrokerError, CLIError, EllfError, HTTPXErrors
+from ..errors import (
+    BrokerError,
+    CLIError,
+    EllfError,
+    HTTPXErrors,
+    JobStartRejectedError,
+)
 from ..messages import Messages
 from ..query import delete_job, resolve_action, resolve_recipe, start_job, stop_job
 from ..ui import (
@@ -108,9 +114,17 @@ def create(
     if not no_start:
         try:
             start(action_id, no_wait=no_wait, as_json=as_json)
-        except Exception:
+        except JobStartRejectedError:
+            # The broker rejected the start, so no workload exists and the
+            # PAM record can be safely rolled back.
             auth.client.action.delete(id=action_id)
             raise
+        except Exception:
+            # The job may have been submitted, so keep the record.
+            msg.warn(
+                Messages.E060.format(noun="action", name=action_id, plural="actions")
+            )
+            raise
     return action_id
 
 

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/commands/agents.py

@@ -9,7 +9,13 @@ from wasabi import msg
 from ellf_pam_sdk.models import AgentDetail, AgentSummary
 
 from ..cli import cli
-from ..errors import BrokerError, CLIError, EllfError, HTTPXErrors
+from ..errors import (
+    BrokerError,
+    CLIError,
+    EllfError,
+    HTTPXErrors,
+    JobStartRejectedError,
+)
 from ..messages import Messages
 from ..query import (
     delete_job,
@@ -115,9 +121,15 @@ def create(
     if not no_start:
         try:
             start(agent_id, no_wait=no_wait, as_json=as_json)
-        except Exception:
+        except JobStartRejectedError:
+            # The broker rejected the start, so no workload exists and the
+            # PAM record can be safely rolled back.
             auth.client.agent.delete(id=agent_id)
             raise
+        except Exception:
+            # The job may have been submitted, so keep the record.
+            msg.warn(Messages.E060.format(noun="agent", name=agent_id, plural="agents"))
+            raise
     return agent_id
 
 

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/commands/services.py

@@ -9,7 +9,13 @@ from wasabi import msg
 from ellf_pam_sdk.models import ServiceDetail, ServiceSummary
 
 from ..cli import cli
-from ..errors import BrokerError, CLIError, EllfError, HTTPXErrors
+from ..errors import (
+    BrokerError,
+    CLIError,
+    EllfError,
+    HTTPXErrors,
+    JobStartRejectedError,
+)
 from ..messages import Messages
 from ..query import (
     delete_job,
@@ -114,9 +120,17 @@ def create(
     if not no_start:
         try:
             start(service_id, no_wait=no_wait, as_json=as_json)
-        except Exception:
+        except JobStartRejectedError:
+            # The broker rejected the start, so no workload exists and the
+            # PAM record can be safely rolled back.
             auth.client.service.delete(id=service_id)
             raise
+        except Exception:
+            # The job may have been submitted, so keep the record.
+            msg.warn(
+                Messages.E060.format(noun="service", name=service_id, plural="services")
+            )
+            raise
     return service_id
 
 

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/commands/tasks.py

@@ -11,7 +11,13 @@ from wasabi import msg
 from ellf_pam_sdk.models import TaskDetail, TaskSummary
 
 from ..cli import cli
-from ..errors import BrokerError, CLIError, EllfError, HTTPXErrors
+from ..errors import (
+    BrokerError,
+    CLIError,
+    EllfError,
+    HTTPXErrors,
+    JobStartRejectedError,
+)
 from ..messages import Messages
 from ..query import delete_job, resolve_recipe, resolve_task, start_job, stop_job
 from ..ui import (
@@ -116,9 +122,15 @@ def create(
     if not no_start:
         try:
             start(task_id, no_wait=no_wait, as_json=as_json)
-        except Exception:
+        except JobStartRejectedError:
+            # The broker rejected the start, so no workload exists and the
+            # PAM record can be safely rolled back.
             auth.client.task.delete(id=task_id)
             raise
+        except Exception:
+            # The job may have been submitted, so keep the record.
+            msg.warn(Messages.E060.format(noun="task", name=task_id, plural="tasks"))
+            raise
     return task_id
 
 

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/errors.py

@@ -61,6 +61,16 @@ class CLIError(Exception):
         super().__init__(self.message)
 
 
+class JobStartRejectedError(CLIError):
+    """The broker rejected a job start outright, so no workload was created.
+
+    Raised only when the broker's start response reports a cluster or
+    validation error. Other start failures (e.g. status polling after the
+    job was submitted) must not use this class: the workload may be running,
+    so it is not safe to roll back the PAM record.
+    """
+
+
 class EllfParseSecretsError(EllfError):
     def __init__(
         self, secrets_file: Path | None, error: Exception | None = None

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/messages.py

@@ -63,6 +63,7 @@ ellf secret create my-credentials OPENAI_API_KEY="sk-..." CUSTOM_SECRET=-
     E057 = "Could not import recipes from {package}"
     E058 = "Could not get a Prodigy login token from the broker for task {name}"
     E059 = "Could not fetch questions from the task server for {name}"
+    E060 = "Could not confirm that {noun} {name} started. The {noun} was created and still exists: check it with `ellf {plural} info {name}` or start it with `ellf {plural} start {name}`"
 
     T001 = "Not creating {noun} {name}: {noun} already exists"
     T002 = "Successfully created {noun} {name}"

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/ellf_cli/query.py

@@ -22,7 +22,7 @@ from ellf_pam_sdk.errors import ELLF_ERRORS, EllfError
 
 from .auth import AuthState
 from .commands._state import get_auth_state, get_root_cfg, get_saved_settings
-from .errors import CLIError, EllfErrors
+from .errors import CLIError, EllfErrors, JobStartRejectedError
 from .messages import Messages
 from .ty import Page
 from .util import URL
@@ -603,10 +603,10 @@ class JobOperations:
             return
         elif change_response.cluster_error:
             err = Messages.E025.format(noun=self.job_type.value, name=job_spec.job_id)
-            raise CLIError(err, change_response.cluster_error)
+            raise JobStartRejectedError(err, change_response.cluster_error)
         elif change_response.validation_error:
             err = Messages.E044.format(noun=self.job_type.value, name=job_spec.job_id)
-            raise CLIError(err, change_response.validation_error)
+            raise JobStartRejectedError(err, change_response.validation_error)
         elif not quiet:
             msg.info(
                 Messages.T006.format(noun=self.job_type.value, name=job_spec.job_id)
@@ -826,7 +826,11 @@ def delete_job(job: _JobT, auth: AuthState, quiet: bool = False) -> _JobT:
 
 
 def check_job_started(
-    job: _JobT, auth: AuthState, *, wait_seconds: int = 60 * 2
+    job: _JobT,
+    auth: AuthState,
+    *,
+    wait_seconds: int = 60 * 2,
+    not_found_grace_seconds: int = 10,
 ) -> _JobT | None:
     job_type = _get_job_type(job)
     started_states = {
@@ -846,7 +850,7 @@ def check_job_started(
     from .ui import loading
 
     with loading(Messages.T034.format(noun=job_type)):
-        for _ in range(wait_seconds):
+        for attempt in range(wait_seconds):
             try:
                 status = auth.broker_client.jobs.get_status(job.id)
             except (httpx.HTTPError, BrokerError):
@@ -858,6 +862,16 @@ def check_job_started(
                 and getattr(status, "health", None) not in failed_healths
             ):
                 break
+            if (
+                status.state == broker_models.WorkloadState.not_found
+                and attempt < not_found_grace_seconds
+                and getattr(status, "health", None) not in failed_healths
+            ):
+                # A just-submitted job can report not_found before its
+                # resources appear in the cluster listing, so an early
+                # not_found is not a failure: keep polling.
+                time.sleep(1)
+                continue
             if status.state in failed_states or (
                 getattr(status, "health", None) in failed_healths
             ):

{ellf_cli-10.0.1 → ellf_cli-10.0.3/ellf_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ellf-cli
-Version: 10.0.1
+Version: 10.0.3
 Summary: Ellf Command Line Interface
 Home-page: https://prodi.gy
 Author: ExplosionAI GmbH

{ellf_cli-10.0.1 → ellf_cli-10.0.3}/tests/test_query.py

@@ -8,7 +8,9 @@ from uuid import UUID, uuid4
 
 import pytest
 
+from ellf_broker_sdk import models as broker_models
 from ellf_cli import query
+from ellf_cli.errors import CLIError
 from ellf_cli.query import _normalise_scope_params, get_not_found_error
 from ellf_pam_sdk.errors import ClusterNotFound, ProjectNotFound
 
@@ -119,3 +121,104 @@ class TestNormaliseScopeParams:
     def test_does_not_touch_unrelated_keys(self, fake_scope_resolution):
         out = _normalise_scope_params({"name": "foo", "after": "cursor-1"})
         assert out == {"name": "foo", "after": "cursor-1"}
+
+
+# ---------------------------------------------------------------------------
+# check_job_started: a just-submitted job can report ``not_found`` before its
+# resources appear in the cluster listing (the broker derives ``not_found``
+# from an empty label-selector list). The first polls must therefore tolerate
+# ``not_found`` instead of treating it as a terminal failure -- this race
+# caused freshly created actions to be reported as failed-to-start (and, before
+# the JobStartRejectedError split, rolled back and deleted) while the workload
+# was in fact running.
+
+
+class _FakeJobsEndpoint:
+    """Broker jobs stub returning a scripted sequence of statuses."""
+
+    def __init__(self, statuses: list[Any]) -> None:
+        self._statuses = statuses
+        self.calls = 0
+
+    def get_status(self, job_id: UUID) -> Any:
+        index = min(self.calls, len(self._statuses) - 1)
+        self.calls += 1
+        return self._statuses[index]
+
+
+class TestCheckJobStarted:
+    def _status(
+        self,
+        state: broker_models.WorkloadState,
+        health: broker_models.WorkloadHealth,
+        *,
+        is_ready: bool = False,
+    ) -> broker_models.JobStatusResponse:
+        return broker_models.JobStatusResponse(
+            job_id=uuid4(), state=state, health=health, is_ready=is_ready
+        )
+
+    def _run(
+        self,
+        statuses: list[broker_models.JobStatusResponse],
+        monkeypatch: pytest.MonkeyPatch,
+        **kwargs: Any,
+    ) -> tuple[Any, _FakeJobsEndpoint]:
+        monkeypatch.setattr(query.time, "sleep", lambda seconds: None)
+        jobs = _FakeJobsEndpoint(statuses)
+        auth = SimpleNamespace(broker_client=SimpleNamespace(jobs=jobs))
+        job = SimpleNamespace(id=uuid4(), url_logs="https://example.com/logs")
+        result = query.check_job_started(job, auth, **kwargs)  # type: ignore[arg-type]
+        return result, jobs
+
+    def test_not_found_within_grace_then_running(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        statuses = [
+            self._status(
+                broker_models.WorkloadState.not_found,
+                broker_models.WorkloadHealth.inactive,
+            )
+        ] * 3 + [
+            self._status(
+                broker_models.WorkloadState.running,
+                broker_models.WorkloadHealth.progressing,
+            )
+        ]
+        result, jobs = self._run(statuses, monkeypatch)
+        assert result is not None
+        assert jobs.calls == 4
+
+    def test_not_found_past_grace_raises(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        statuses = [
+            self._status(
+                broker_models.WorkloadState.not_found,
+                broker_models.WorkloadHealth.inactive,
+            )
+        ]
+        with pytest.raises(CLIError):
+            self._run(statuses, monkeypatch, not_found_grace_seconds=5)
+        # Polls through the grace window, then fails on the next poll.
+
+    def test_failed_state_raises_without_grace(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        statuses = [
+            self._status(
+                broker_models.WorkloadState.failed,
+                broker_models.WorkloadHealth.terminal,
+            )
+        ]
+        with pytest.raises(CLIError):
+            self._run(statuses, monkeypatch)
+
+    def test_succeeded_returns_job(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        statuses = [
+            self._status(
+                broker_models.WorkloadState.succeeded,
+                broker_models.WorkloadHealth.inactive,
+            )
+        ]
+        result, jobs = self._run(statuses, monkeypatch)
+        assert result is not None
+        assert jobs.calls == 1