Update link to console, conditionally display studio link, update link color to blue

Author: mollyheamazon

.gitignore

@@ -40,3 +40,4 @@ env/
 sagemaker_train/src/**/container_drivers/sm_train.sh
 sagemaker_train/src/**/container_drivers/sourcecode.json
 sagemaker_train/src/**/container_drivers/distributed.json
+.kiro

sagemaker-train/src/sagemaker/train/common_utils/metrics_visualizer.py

@@ -2,12 +2,85 @@
 
 import logging
 from typing import Optional, List, Dict, Any
-import boto3
 from sagemaker.core.resources import TrainingJob
 
 logger = logging.getLogger(__name__)
 
 
+def _is_in_studio() -> bool:
+    """Check if running inside SageMaker Studio."""
+    from sagemaker.train.common_utils.finetune_utils import _read_domain_id_from_metadata
+    return _read_domain_id_from_metadata() is not None
+
+
+def _get_studio_base_url(region: str) -> str:
+    """Get Studio base URL, or empty string if domain not resolvable."""
+    from sagemaker.train.common_utils.finetune_utils import _read_domain_id_from_metadata
+    domain_id = _read_domain_id_from_metadata()
+    if not domain_id or not region:
+        return ""
+    return f"https://studio-{domain_id}.studio.{region}.sagemaker.aws"
+
+
+def _parse_job_arn(job_arn: str):
+    """Parse a SageMaker job ARN into (region, resource) or None."""
+    import re
+    m = re.match(r'arn:aws(?:-[a-z]+)?:sagemaker:([a-z0-9-]+):\d+:(\S+)', job_arn)
+    return (m.group(1), m.group(2)) if m else None
+
+
+def get_console_job_url(job_arn: str) -> str:
+    """Get AWS Console URL for a SageMaker job ARN.
+    
+    Args:
+        job_arn: Full ARN like arn:aws:sagemaker:us-east-1:123:training-job/my-job
+        
+    Returns:
+        Console URL or empty string.
+    """
+    parsed = _parse_job_arn(job_arn)
+    if not parsed:
+        return ""
+    region, resource = parsed
+    job_type_map = {
+        "training-job/": "#/jobs/",
+        "processing-job/": "#/processing-jobs/",
+        "transform-job/": "#/transform-jobs/",
+    }
+    for prefix, fragment in job_type_map.items():
+        if resource.startswith(prefix):
+            job_name = resource.split("/", 1)[1]
+            return f"https://{region}.console.aws.amazon.com/sagemaker/home?region={region}{fragment}{job_name}"
+    return ""
+
+
+def get_cloudwatch_logs_url(job_arn: str) -> str:
+    """Get CloudWatch Logs console URL for a SageMaker job ARN.
+    
+    Returns:
+        CloudWatch console URL or empty string.
+    """
+    parsed = _parse_job_arn(job_arn)
+    if not parsed:
+        return ""
+    region, resource = parsed
+    log_group_map = {
+        "training-job/": "/aws/sagemaker/TrainingJobs",
+        "processing-job/": "/aws/sagemaker/ProcessingJobs",
+        "transform-job/": "/aws/sagemaker/TransformJobs",
+    }
+    for prefix, log_group in log_group_map.items():
+        if resource.startswith(prefix):
+            job_name = resource.split("/", 1)[1]
+            encoded_group = log_group.replace("/", "$252F")
+            return (
+                f"https://{region}.console.aws.amazon.com/cloudwatch/home?region={region}"
+                f"#logsV2:log-groups/log-group/{encoded_group}"
+                f"$3FlogStreamNameFilter$3D{job_name}"
+            )
+    return ""
+
+
 def get_studio_url(training_job, domain_id: str = None) -> str:
     """Get SageMaker Studio URL for training job logs.
     
@@ -43,34 +116,10 @@ def get_studio_url(training_job, domain_id: str = None) -> str:
         region = training_job.region if hasattr(training_job, 'region') and training_job.region else 'us-east-1'
         job_name = training_job.training_job_name
 
-    # Auto-detect domain if not provided
-    if not domain_id:
-        # First try Studio metadata (when running inside Studio)
-        try:
-            import os, json as _json
-            metadata_path = '/opt/ml/metadata/resource-metadata.json'
-            if os.path.exists(metadata_path):
-                with open(metadata_path, 'r') as f:
-                    domain_id = _json.load(f).get('DomainId')
-        except Exception:
-            pass
-
-    if not domain_id:
-        # Fall back to list_domains, sorted by creation time for deterministic results
-        try:
-            sm_client = boto3.client('sagemaker', region_name=region)
-            domains = sm_client.list_domains()['Domains']
-            if domains:
-                domains.sort(key=lambda d: d.get('CreationTime', ''))
-                domain_id = domains[0]['DomainId']
-        except Exception:
-            pass
-    
-    if not domain_id:
+    base = _get_studio_base_url(region)
+    if not base:
         return ""
-    
-    # Studio URL format: https://studio-{domain_id}.studio.{region}.sagemaker.aws/jobs/train/{job_name}
-    return f"https://studio-{domain_id}.studio.{region}.sagemaker.aws/jobs/train/{job_name}"
+    return f"{base}/jobs/train/{job_name}"
 
 
 def display_job_links_html(rows: list, as_html: bool = False):

sagemaker-train/src/sagemaker/train/common_utils/trainer_wait.py

@@ -299,34 +299,45 @@ def get_cached_mlflow_url():
 
                     clear_output(wait=True)
 
-                    # Header section with training job name
+                    # Header section with training job info
                     header_table = Table(show_header=False, box=None, padding=(0, 1))
                     header_table.add_column("Property", style="cyan bold", width=20)
                     header_table.add_column("Value", style="dim", overflow="fold")
 
-                    # Add Studio job link
-                    try:
-                        from sagemaker.train.common_utils.metrics_visualizer import get_studio_url
-                        studio_url = get_studio_url(training_job)
-                        header_table.add_row("TrainingJob Name", f"[underline][link={studio_url}]🔗 {training_job.training_job_name}[/link][/underline]")
-                    except Exception:
-                        header_table.add_row("TrainingJob Name", f"[bold green]{training_job.training_job_name}[/bold green]")
-                    
+                    header_table.add_row("TrainingJob Name", f"[bold green]{training_job.training_job_name}[/bold green]")
                     header_table.add_row("TrainingJob ARN", f"[dim]{training_job.training_job_arn}[/dim]")
 
-                    # Add MLflow link to header if available
+                    # Build links row
+                    links = []
+                    try:
+                        from sagemaker.train.common_utils.metrics_visualizer import (
+                            _is_in_studio, get_console_job_url, get_cloudwatch_logs_url, get_studio_url
+                        )
+                        if _is_in_studio():
+                            studio_url = get_studio_url(training_job)
+                            if studio_url:
+                                links.append(f"[bright_blue underline][link={studio_url}]🔗 Training Job (Studio)[/link][/bright_blue underline]")
+                        else:
+                            console_url = get_console_job_url(training_job.training_job_arn)
+                            if console_url:
+                                links.append(f"[bright_blue underline][link={console_url}]🔗 Training Job[/link][/bright_blue underline]")
+                        cw_url = get_cloudwatch_logs_url(training_job.training_job_arn)
+                        if cw_url:
+                            links.append(f"[bright_blue underline][link={cw_url}]🔗 CloudWatch Logs[/link][/bright_blue underline]")
+                    except Exception:
+                        pass
                     if has_mlflow_config:
                         cached_url = get_cached_mlflow_url()
                         if cached_url:
-                            exp_name = training_job.mlflow_config.mlflow_experiment_name if hasattr(training_job, 'mlflow_config') else None
-                            if exp_name and not _is_unassigned_attribute(exp_name):
-                                link_text = exp_name
-                            else:
-                                link_text = "MLflow Experiment"
-                            
-                            header_table.add_row("MLflow Experiment", f"[underline][link={cached_url}]🔗 {link_text}[/link][/underline]")
+                            links.append(f"[bright_blue underline][link={cached_url}]🔗 MLflow Experiment[/link][/bright_blue underline]")
                         elif mlflow_link_cache['error']:
                             header_table.add_row("MLflow Experiment", f"[red]{mlflow_link_cache['error']}[/red]")
+                    if has_mlflow_config:
+                        exp_name = training_job.mlflow_config.mlflow_experiment_name if hasattr(training_job, 'mlflow_config') else None
+                        if exp_name and not _is_unassigned_attribute(exp_name):
+                            header_table.add_row("MLflow Experiment", f"{exp_name}")
+                    if links:
+                        header_table.add_row("Links", " | ".join(links))
 
                     status_table = Table(show_header=False, box=None, padding=(0, 1))
                     status_table.add_column("Property", style="cyan bold", width=20)

sagemaker-train/src/sagemaker/train/evaluate/execution.py

@@ -931,31 +931,37 @@ def wait(
                 header_table.add_column("Property", style="cyan bold", width=20)
                 header_table.add_column("Value", style="dim", overflow="fold")
 
-                # Extract pipeline name from execution ARN and build Studio link
+                # Extract pipeline name and region from execution ARN
                 pipeline_name = None
                 exec_id = ''
+                region = None
                 if self.arn:
                     arn_parts = self.arn.split('/')
                     if len(arn_parts) >= 4:
                         pipeline_name = arn_parts[-3]
                         exec_id = arn_parts[-1]
+                    region = self.arn.split(":")[3] if len(self.arn.split(":")) > 3 else None
                 # Use execution display name if available, fall back to self.name
                 display_name = self.name
                 if self._pipeline_execution:
                     dn = getattr(self._pipeline_execution, 'pipeline_execution_display_name', None)
                     if dn and not (hasattr(dn, '__class__') and 'Unassigned' in dn.__class__.__name__):
                         display_name = dn
+                header_table.add_row("Evaluation Job", str(display_name))
+
+                # Build links row
+                links = []
                 try:
-                    from sagemaker.train.common_utils.metrics_visualizer import get_studio_url
-                    dummy_url = get_studio_url(self.arn.split('/')[0].replace(':pipeline', ':training-job') + '/dummy' if self.arn else 'dummy')
-                    if dummy_url and pipeline_name:
-                        base = dummy_url.rsplit('/jobs/train/', 1)[0]
-                        pipeline_url = f"{base}/jobs/evaluation/detail?pipeline_name={pipeline_name}&execution_id={exec_id}"
-                        header_table.add_row("Evaluation Job", f"[underline][link={pipeline_url}]🔗 {display_name}[/link][/underline]")
-                    else:
-                        header_table.add_row("Evaluation Job", str(display_name))
+                    from sagemaker.train.common_utils.metrics_visualizer import _is_in_studio, _get_studio_base_url
+                    if region and pipeline_name and _is_in_studio():
+                        base = _get_studio_base_url(region)
+                        if base:
+                            pipeline_url = f"{base}/jobs/evaluation/detail?pipeline_name={pipeline_name}&execution_id={exec_id}"
+                            links.append(f"[bright_blue underline][link={pipeline_url}]🔗 Pipeline Execution (Studio)[/link][/bright_blue underline]")
                 except Exception:
-                    header_table.add_row("Evaluation Job", str(display_name))
+                    pass
+                if links:
+                    header_table.add_row("Links", " | ".join(links))
 
                 # Create main status table
                 status_table = Table(show_header=False, box=None, padding=(0, 1))
@@ -1045,16 +1051,45 @@ def wait(
                     if job_arn_entries:
                         links_table = Table(show_header=True, header_style="bold magenta", box=None, padding=(0, 1))
                         links_table.add_column("Step", style="cyan", width=20)
-                        links_table.add_column("Job Link", width=12)
+                        links_table.add_column("Job Link", style="dim")
+                        links_table.add_column("Logs", style="dim")
                         links_table.add_column("Job ARN", style="dim", overflow="fold")
+                        from sagemaker.train.common_utils.metrics_visualizer import (
+                            _is_in_studio, _parse_job_arn, _get_studio_base_url,
+                            get_console_job_url, get_cloudwatch_logs_url,
+                        )
+                        in_studio = _is_in_studio()
+                        studio_base = _get_studio_base_url(region) if in_studio else ""
+                        studio_path_map = {
+                            "training-job/": "jobs/train/",
+                            "processing-job/": "jobs/processing/",
+                            "transform-job/": "jobs/transform/",
+                        }
                         for entry in job_arn_entries:
+                            job_link = ""
+                            logs_link = ""
                             try:
-                                from sagemaker.train.common_utils.metrics_visualizer import get_studio_url
-                                url = get_studio_url(entry['job_arn'])
-                                link_col = f"[underline][link={url}]🔗 link[/link][/underline]" if url else ""
+                                arn = entry['job_arn']
+                                if in_studio and studio_base:
+                                    parsed = _parse_job_arn(arn)
+                                    if parsed:
+                                        _, resource = parsed
+                                        for prefix, path in studio_path_map.items():
+                                            if resource.startswith(prefix):
+                                                job_name = resource.split("/", 1)[1]
+                                                url = f"{studio_base}/{path}{job_name}"
+                                                job_link = f"[bright_blue underline][link={url}]🔗 link[/link][/bright_blue underline]"
+                                                break
+                                else:
+                                    url = get_console_job_url(arn)
+                                    if url:
+                                        job_link = f"[bright_blue underline][link={url}]🔗 link[/link][/bright_blue underline]"
+                                cw_url = get_cloudwatch_logs_url(arn)
+                                if cw_url:
+                                    logs_link = f"[bright_blue underline][link={cw_url}]🔗 logs[/link][/bright_blue underline]"
                             except Exception:
-                                link_col = ""
-                            links_table.add_row(entry['step_name'], link_col, entry['job_arn'])
+                                pass
+                            links_table.add_row(entry['step_name'], job_link, logs_link, entry['job_arn'])
                         content_parts.append(Text(""))
                         content_parts.append(Text("Job ARNs", style="bold magenta"))
                         content_parts.append(links_table)