sagemaker-ops-cli 0.1.1__tar.gz → 0.2.0__tar.gz

{sagemaker_ops_cli-0.1.1 → sagemaker_ops_cli-0.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sagemaker-ops-cli
-Version: 0.1.1
+Version: 0.2.0
 Summary: CLI and TUI for submitting and monitoring Amazon SageMaker Processing Jobs and Pipelines.
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
@@ -54,7 +54,7 @@ pip install git+https://github.com/southpolemonkey/smops.git
 Install from a local wheel:
 
 ```bash
-pip install dist/sagemaker_ops_cli-0.1.1-py3-none-any.whl
+pip install dist/sagemaker_ops_cli-0.2.0-py3-none-any.whl
 ```
 
 Install with Homebrew:
@@ -85,6 +85,29 @@ To enable YAML config files:
 pip install -e '.[yaml]'
 ```
 
+## Defaults
+
+Set a default AWS region once so you do not need to pass `--region` on every command:
+
+```bash
+smops config set-region ap-southeast-2
+smops config get-region
+```
+
+The config file is stored at `~/.config/smops/config.json` by default. You can inspect it with:
+
+```bash
+smops config show
+smops config path
+```
+
+Region resolution order is:
+
+1. `--region`
+2. `SMOPS_DEFAULT_REGION`
+3. `smops config set-region ...`
+4. Region configured on the selected AWS profile
+
 ## Build The Python Package
 
 ```bash
@@ -94,8 +117,8 @@ python -m build
 
 Build artifacts are written to `dist/`:
 
-- `sagemaker_ops_cli-0.1.1-py3-none-any.whl`
-- `sagemaker_ops_cli-0.1.1.tar.gz`
+- `sagemaker_ops_cli-0.2.0-py3-none-any.whl`
+- `sagemaker_ops_cli-0.2.0.tar.gz`
 
 ## Submit A Processing Job
 
@@ -126,6 +149,23 @@ smops pipeline start \
   --parameter Mode=prod
 ```
 
+## Interactive TUI
+
+Open the TUI selector and choose between Pipelines and Processing Jobs:
+
+```bash
+smops tui --profile dev
+```
+
+Inside the TUI:
+
+- `p` / `P`: switch to the next AWS profile from your local AWS config
+- `s`: start a pipeline or submit a processing job from the current TUI
+- `r`: refresh
+- `q`: quit
+
+For pipeline starts, enter the pipeline name, optional display name, and optional comma-separated parameters such as `InputDate=2026-07-01,Mode=test`. For processing job submits, enter the path to a JSON/YAML config file using the same structure as boto3 `create_processing_job`.
+
 ## Processing Jobs TUI
 
 ```bash
@@ -147,6 +187,8 @@ smops tui processing --all-profiles
 Keyboard shortcuts:
 
 - `Up` / `Down` or `Left` / `Right`: switch jobs
+- `p` / `P`: switch to the next AWS profile
+- `s`: submit a Processing Job from a JSON/YAML config file
 - `r`: refresh
 - `q`: quit
 
@@ -172,6 +214,8 @@ Keyboard shortcuts:
 
 - `Left` / `Right`: switch focus between the executions and steps panels
 - `Up` / `Down`: move within the focused panel
+- `p` / `P`: switch to the next AWS profile
+- `s`: start a Pipeline execution
 - `l`: load the CloudWatch log tail for the selected failed step
 - `r`: refresh
 - `q`: quit
@@ -186,11 +230,32 @@ Log discovery is currently supported for these step job types:
 
 ```bash
 smops processing list --profile dev --region us-east-1
+smops processing wait --profile dev --region us-east-1 --name my-processing-job
 smops pipeline list --profile dev --region us-east-1
 smops pipeline list --profile dev --region us-east-1 --name my-pipeline --hours 6
 smops pipeline steps --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
+smops pipeline wait --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
+smops pipeline inspect --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
+smops pipeline diagnose --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
 ```
 
+Most non-interactive commands support `--json` for agents and automation:
+
+```bash
+smops processing list --profile dev --region us-east-1 --json
+smops processing wait --profile dev --region us-east-1 --name my-processing-job --json
+smops pipeline start --profile dev --region us-east-1 --name my-pipeline --json
+smops pipeline list --profile dev --region us-east-1 --json
+smops pipeline steps --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+smops pipeline wait --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+smops pipeline inspect --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+smops pipeline diagnose --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+```
+
+JSON responses use a stable envelope. Successful commands return `status: "ok"`; errors return `status: "error"` and a user-facing `error` message. List commands return `items`, `count`, and `next_token`.
+
+`pipeline inspect` returns execution details, all steps, and failed steps. `pipeline diagnose` extends that with the first failed step, inferred SageMaker job type/name, CloudWatch log group and stream prefix, log tail, and suggested next actions.
+
 `processing list` reads 20 running jobs per page by default. If the output includes `Next token`, pass it to fetch the next page:
 
 ```bash

{sagemaker_ops_cli-0.1.1 → sagemaker_ops_cli-0.2.0}/README.md

@@ -34,7 +34,7 @@ pip install git+https://github.com/southpolemonkey/smops.git
 Install from a local wheel:
 
 ```bash
-pip install dist/sagemaker_ops_cli-0.1.1-py3-none-any.whl
+pip install dist/sagemaker_ops_cli-0.2.0-py3-none-any.whl
 ```
 
 Install with Homebrew:
@@ -65,6 +65,29 @@ To enable YAML config files:
 pip install -e '.[yaml]'
 ```
 
+## Defaults
+
+Set a default AWS region once so you do not need to pass `--region` on every command:
+
+```bash
+smops config set-region ap-southeast-2
+smops config get-region
+```
+
+The config file is stored at `~/.config/smops/config.json` by default. You can inspect it with:
+
+```bash
+smops config show
+smops config path
+```
+
+Region resolution order is:
+
+1. `--region`
+2. `SMOPS_DEFAULT_REGION`
+3. `smops config set-region ...`
+4. Region configured on the selected AWS profile
+
 ## Build The Python Package
 
 ```bash
@@ -74,8 +97,8 @@ python -m build
 
 Build artifacts are written to `dist/`:
 
-- `sagemaker_ops_cli-0.1.1-py3-none-any.whl`
-- `sagemaker_ops_cli-0.1.1.tar.gz`
+- `sagemaker_ops_cli-0.2.0-py3-none-any.whl`
+- `sagemaker_ops_cli-0.2.0.tar.gz`
 
 ## Submit A Processing Job
 
@@ -106,6 +129,23 @@ smops pipeline start \
   --parameter Mode=prod
 ```
 
+## Interactive TUI
+
+Open the TUI selector and choose between Pipelines and Processing Jobs:
+
+```bash
+smops tui --profile dev
+```
+
+Inside the TUI:
+
+- `p` / `P`: switch to the next AWS profile from your local AWS config
+- `s`: start a pipeline or submit a processing job from the current TUI
+- `r`: refresh
+- `q`: quit
+
+For pipeline starts, enter the pipeline name, optional display name, and optional comma-separated parameters such as `InputDate=2026-07-01,Mode=test`. For processing job submits, enter the path to a JSON/YAML config file using the same structure as boto3 `create_processing_job`.
+
 ## Processing Jobs TUI
 
 ```bash
@@ -127,6 +167,8 @@ smops tui processing --all-profiles
 Keyboard shortcuts:
 
 - `Up` / `Down` or `Left` / `Right`: switch jobs
+- `p` / `P`: switch to the next AWS profile
+- `s`: submit a Processing Job from a JSON/YAML config file
 - `r`: refresh
 - `q`: quit
 
@@ -152,6 +194,8 @@ Keyboard shortcuts:
 
 - `Left` / `Right`: switch focus between the executions and steps panels
 - `Up` / `Down`: move within the focused panel
+- `p` / `P`: switch to the next AWS profile
+- `s`: start a Pipeline execution
 - `l`: load the CloudWatch log tail for the selected failed step
 - `r`: refresh
 - `q`: quit
@@ -166,11 +210,32 @@ Log discovery is currently supported for these step job types:
 
 ```bash
 smops processing list --profile dev --region us-east-1
+smops processing wait --profile dev --region us-east-1 --name my-processing-job
 smops pipeline list --profile dev --region us-east-1
 smops pipeline list --profile dev --region us-east-1 --name my-pipeline --hours 6
 smops pipeline steps --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
+smops pipeline wait --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
+smops pipeline inspect --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
+smops pipeline diagnose --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:...
 ```
 
+Most non-interactive commands support `--json` for agents and automation:
+
+```bash
+smops processing list --profile dev --region us-east-1 --json
+smops processing wait --profile dev --region us-east-1 --name my-processing-job --json
+smops pipeline start --profile dev --region us-east-1 --name my-pipeline --json
+smops pipeline list --profile dev --region us-east-1 --json
+smops pipeline steps --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+smops pipeline wait --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+smops pipeline inspect --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+smops pipeline diagnose --profile dev --region us-east-1 --execution-arn arn:aws:sagemaker:... --json
+```
+
+JSON responses use a stable envelope. Successful commands return `status: "ok"`; errors return `status: "error"` and a user-facing `error` message. List commands return `items`, `count`, and `next_token`.
+
+`pipeline inspect` returns execution details, all steps, and failed steps. `pipeline diagnose` extends that with the first failed step, inferred SageMaker job type/name, CloudWatch log group and stream prefix, log tail, and suggested next actions.
+
 `processing list` reads 20 running jobs per page by default. If the output includes `Next token`, pass it to fetch the next page:
 
 ```bash

{sagemaker_ops_cli-0.1.1 → sagemaker_ops_cli-0.2.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "sagemaker-ops-cli"
-version = "0.1.1"
+version = "0.2.0"
 description = "CLI and TUI for submitting and monitoring Amazon SageMaker Processing Jobs and Pipelines."
 readme = "README.md"
 requires-python = ">=3.10"

{sagemaker_ops_cli-0.1.1 → sagemaker_ops_cli-0.2.0}/sagemaker_ops/__init__.py

@@ -1,4 +1,4 @@
 """SageMaker operations CLI."""
 
-__version__ = "0.1.1"
+__version__ = "0.2.0"
 

{sagemaker_ops_cli-0.1.1 → sagemaker_ops_cli-0.2.0}/sagemaker_ops/aws.py

@@ -3,6 +3,8 @@ from __future__ import annotations
 import base64
 import binascii
 import json
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
 from dataclasses import dataclass
 from datetime import datetime, timedelta, timezone
 from pathlib import Path
@@ -14,6 +16,8 @@ from botocore.exceptions import BotoCoreError, ClientError
 
 ACTIVE_PROCESSING_STATUSES = ("InProgress", "Stopping")
 ACTIVE_PIPELINE_STATUSES = ("Executing", "Stopping")
+TERMINAL_PROCESSING_STATUSES = ("Completed", "Failed", "Stopped")
+TERMINAL_PIPELINE_STATUSES = ("Succeeded", "Failed", "Stopped")
 
 
 class AwsCliError(RuntimeError):
@@ -102,6 +106,13 @@ def parse_parameters(items: Iterable[str]) -> list[dict[str, str]]:
     return parameters
 
 
+def available_profiles() -> list[str]:
+    try:
+        return list(boto3.Session().available_profiles)
+    except BotoCoreError as exc:
+        raise AwsCliError(f"读取 AWS profiles 失败: {exc}") from exc
+
+
 def build_contexts(
     profiles: tuple[str, ...],
     region: str | None,
@@ -143,6 +154,32 @@ def submit_processing_job(ctx: AwsContext, spec: dict[str, Any]) -> dict[str, An
         raise AwsCliError(f"提交 processing job 失败: {exc}") from exc
 
 
+def describe_processing_job(ctx: AwsContext, job_name: str) -> ProcessingJobView:
+    try:
+        detail = ctx.sagemaker.describe_processing_job(ProcessingJobName=job_name)
+    except (BotoCoreError, ClientError) as exc:
+        raise AwsCliError(f"读取 processing job 失败 job={job_name}: {exc}") from exc
+    return _processing_job_view_from_detail(ctx, detail)
+
+
+def wait_processing_job(
+    ctx: AwsContext,
+    job_name: str,
+    timeout_seconds: int = 3600,
+    poll_seconds: int = 30,
+) -> ProcessingJobView:
+    deadline = time.monotonic() + max(0, timeout_seconds)
+    poll_seconds = max(1, poll_seconds)
+
+    while True:
+        job = describe_processing_job(ctx, job_name)
+        if job.status in TERMINAL_PROCESSING_STATUSES:
+            return job
+        if time.monotonic() >= deadline:
+            raise AwsCliError(f"等待 processing job 超时 job={job_name} status={job.status}")
+        time.sleep(min(poll_seconds, max(0.0, deadline - time.monotonic())))
+
+
 def start_pipeline_execution(
     ctx: AwsContext,
     pipeline_name: str,
@@ -214,13 +251,22 @@ def _processing_job_view_from_summary(ctx: AwsContext, summary: dict[str, Any])
         detail = ctx.sagemaker.describe_processing_job(ProcessingJobName=name)
     except (BotoCoreError, ClientError):
         detail = summary
+    return _processing_job_view_from_detail(ctx, detail, fallback=summary)
+
+
+def _processing_job_view_from_detail(
+    ctx: AwsContext,
+    detail: dict[str, Any],
+    fallback: dict[str, Any] | None = None,
+) -> ProcessingJobView:
+    fallback = fallback or {}
     cluster = detail.get("ProcessingResources", {}).get("ClusterConfig", {})
     return ProcessingJobView(
         profile=ctx.profile,
         region=ctx.region,
-        name=name,
-        status=detail.get("ProcessingJobStatus", summary.get("ProcessingJobStatus", "")),
-        creation_time=detail.get("CreationTime", summary.get("CreationTime")),
+        name=detail.get("ProcessingJobName", fallback.get("ProcessingJobName", "")),
+        status=detail.get("ProcessingJobStatus", fallback.get("ProcessingJobStatus", "")),
+        creation_time=detail.get("CreationTime", fallback.get("CreationTime")),
         last_modified_time=detail.get("LastModifiedTime"),
         started_time=detail.get("ProcessingStartTime"),
         ended_time=detail.get("ProcessingEndTime"),
@@ -228,7 +274,7 @@ def _processing_job_view_from_summary(ctx: AwsContext, summary: dict[str, Any])
         instance_count=cluster.get("InstanceCount"),
         role_arn=detail.get("RoleArn", ""),
         failure_reason=detail.get("FailureReason", ""),
-        arn=detail.get("ProcessingJobArn", summary.get("ProcessingJobArn", "")),
+        arn=detail.get("ProcessingJobArn", fallback.get("ProcessingJobArn", "")),
     )
 
 
@@ -281,10 +327,7 @@ def list_pipeline_executions_page(
         next_token=next_token,
     )
     cutoff = datetime.now(timezone.utc) - timedelta(hours=recent_hours)
-    executions: list[PipelineExecutionView] = []
-
-    for name in names:
-        executions.extend(_list_recent_pipeline_executions_for_name(ctx, name, per_pipeline, cutoff))
+    executions = _list_recent_pipeline_executions(ctx, names, per_pipeline, cutoff)
 
     return PipelineExecutionsPage(
         executions=sorted(
@@ -296,6 +339,29 @@ def list_pipeline_executions_page(
     )
 
 
+def _list_recent_pipeline_executions(
+    ctx: AwsContext,
+    pipeline_names: list[str],
+    per_pipeline: int,
+    cutoff: datetime,
+) -> list[PipelineExecutionView]:
+    if not pipeline_names:
+        return []
+    if len(pipeline_names) == 1:
+        return _list_recent_pipeline_executions_for_name(ctx, pipeline_names[0], per_pipeline, cutoff)
+
+    executions: list[PipelineExecutionView] = []
+    workers = min(8, len(pipeline_names))
+    with ThreadPoolExecutor(max_workers=workers) as executor:
+        futures = [
+            executor.submit(_list_recent_pipeline_executions_for_name, ctx, name, per_pipeline, cutoff)
+            for name in pipeline_names
+        ]
+        for future in as_completed(futures):
+            executions.extend(future.result())
+    return executions
+
+
 def _list_recent_pipeline_executions_for_name(
     ctx: AwsContext,
     pipeline_name: str,
@@ -317,18 +383,17 @@ def _list_recent_pipeline_executions_for_name(
     for summary in response.get("PipelineExecutionSummaries", []):
         status = summary.get("PipelineExecutionStatus", "")
         execution_arn = summary.get("PipelineExecutionArn", "")
-        detail = _describe_pipeline_execution_safely(ctx, execution_arn) if execution_arn else {}
-        start_time = detail.get("StartTime", summary.get("StartTime"))
-        last_modified_time = detail.get("LastModifiedTime", summary.get("LastModifiedTime"))
+        start_time = summary.get("StartTime")
+        last_modified_time = summary.get("LastModifiedTime")
         if not _should_show_pipeline_execution(status, start_time, last_modified_time, cutoff):
             continue
         executions.append(
             PipelineExecutionView(
                 profile=ctx.profile,
                 region=ctx.region,
-                pipeline_name=detail.get("PipelineName", pipeline_name),
+                pipeline_name=pipeline_name,
                 execution_arn=execution_arn,
-                display_name=summary.get("PipelineExecutionDisplayName", detail.get("PipelineExecutionDisplayName", "")),
+                display_name=summary.get("PipelineExecutionDisplayName", ""),
                 status=status,
                 start_time=start_time,
                 last_modified_time=last_modified_time,
@@ -405,6 +470,71 @@ def describe_pipeline_execution(ctx: AwsContext, execution_arn: str) -> dict[str
         raise AwsCliError(f"读取 pipeline execution 失败: {exc}") from exc
 
 
+def wait_pipeline_execution(
+    ctx: AwsContext,
+    execution_arn: str,
+    timeout_seconds: int = 3600,
+    poll_seconds: int = 30,
+) -> dict[str, Any]:
+    deadline = time.monotonic() + max(0, timeout_seconds)
+    poll_seconds = max(1, poll_seconds)
+
+    while True:
+        detail = describe_pipeline_execution(ctx, execution_arn)
+        status = detail.get("PipelineExecutionStatus", "")
+        if status in TERMINAL_PIPELINE_STATUSES:
+            return detail
+        if time.monotonic() >= deadline:
+            raise AwsCliError(f"等待 pipeline execution 超时 execution={execution_arn} status={status}")
+        time.sleep(min(poll_seconds, max(0.0, deadline - time.monotonic())))
+
+
+def inspect_pipeline_execution(ctx: AwsContext, execution_arn: str) -> dict[str, Any]:
+    detail = describe_pipeline_execution(ctx, execution_arn)
+    detail = {
+        **detail,
+        "PipelineExecutionArn": detail.get("PipelineExecutionArn") or execution_arn,
+        "PipelineName": detail.get("PipelineName") or _pipeline_name_from_execution_arn(execution_arn),
+    }
+    steps = list_pipeline_steps(ctx, execution_arn)
+    failed_steps = [step for step in steps if step.get("StepStatus") == "Failed"]
+    return {
+        "profile": ctx.profile,
+        "region": ctx.region,
+        "execution": detail,
+        "steps": steps,
+        "failed_steps": failed_steps,
+    }
+
+
+def diagnose_pipeline_execution(
+    ctx: AwsContext,
+    execution_arn: str,
+    log_limit: int = 80,
+) -> dict[str, Any]:
+    inspection = inspect_pipeline_execution(ctx, execution_arn)
+    failed_steps = inspection["failed_steps"]
+    failed_step = failed_steps[0] if failed_steps else None
+    log_source = infer_log_source(failed_step) if failed_step else None
+    log_tail = tail_step_logs(ctx, failed_step, limit=log_limit) if failed_step else []
+    job_type = None
+    job_name = None
+    if log_source and failed_step:
+        job_type = _step_job_type(failed_step)
+        job_name = log_source[1]
+
+    return {
+        **inspection,
+        "failed_step": failed_step,
+        "job_type": job_type,
+        "job_name": job_name,
+        "log_group": log_source[0] if log_source else None,
+        "log_stream_prefix": log_source[1] if log_source else None,
+        "log_tail": log_tail,
+        "next_actions": _diagnostic_next_actions(execution_arn, failed_step),
+    }
+
+
 def tail_step_logs(ctx: AwsContext, step: dict[str, Any], limit: int = 80) -> list[str]:
     source = infer_log_source(step)
     if source is None:
@@ -413,6 +543,10 @@ def tail_step_logs(ctx: AwsContext, step: dict[str, Any], limit: int = 80) -> li
     return tail_cloudwatch_logs(ctx, log_group, stream_prefix, limit=limit)
 
 
+def tail_processing_job_logs(ctx: AwsContext, job_name: str, limit: int = 80) -> list[str]:
+    return tail_cloudwatch_logs(ctx, "/aws/sagemaker/ProcessingJobs", job_name, limit=limit)
+
+
 def tail_cloudwatch_logs(
     ctx: AwsContext,
     log_group: str,
@@ -467,6 +601,39 @@ def infer_log_source(step: dict[str, Any]) -> tuple[str, str] | None:
     return None
 
 
+def _step_job_type(step: dict[str, Any]) -> str | None:
+    metadata = step.get("Metadata") or {}
+    for key in ("ProcessingJob", "TrainingJob", "TransformJob"):
+        if isinstance(metadata.get(key), dict):
+            return key
+    return None
+
+
+def _pipeline_name_from_execution_arn(execution_arn: str) -> str:
+    marker = ":pipeline/"
+    if marker not in execution_arn:
+        return ""
+    tail = execution_arn.split(marker, 1)[1]
+    return tail.split("/", 1)[0]
+
+
+def _diagnostic_next_actions(execution_arn: str, failed_step: dict[str, Any] | None) -> list[dict[str, str]]:
+    actions = [
+        {
+            "type": "inspect",
+            "command": f"smops pipeline inspect --execution-arn {execution_arn} --json",
+        }
+    ]
+    if failed_step:
+        actions.append(
+            {
+                "type": "diagnose",
+                "command": f"smops pipeline diagnose --execution-arn {execution_arn} --json",
+            }
+        )
+    return actions
+
+
 def format_dt(value: datetime | None) -> str:
     if value is None:
         return ""