dayhoff-tools 1.14.11__tar.gz → 1.14.13__tar.gz

{dayhoff_tools-1.14.11 → dayhoff_tools-1.14.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dayhoff-tools
-Version: 1.14.11
+Version: 1.14.13
 Summary: Common tools for all the repos at Dayhoff Labs
 Author: Daniel Martin-Alarcon
 Author-email: dma@dayhofflabs.com

{dayhoff_tools-1.14.11 → dayhoff_tools-1.14.13}/dayhoff_tools/cli/batch/commands/cancel.py

@@ -1,5 +1,7 @@
 """Cancel command for stopping running jobs."""
 
+import re
+
 import click
 
 from ..aws_batch import BatchClient, BatchError
@@ -11,6 +13,14 @@ from ..manifest import (
 )
 
 
+def _parse_retry_job_id(job_id: str) -> tuple[str, str | None]:
+    """Parse a job ID to extract parent job ID and retry suffix."""
+    match = re.match(r"^(.+)(-r\d+)$", job_id)
+    if match:
+        return match.group(1), job_id
+    return job_id, None
+
+
 @click.command()
 @click.argument("job_id")
 @click.option("--force", is_flag=True, help="Force termination of running containers")
@@ -19,19 +29,29 @@ def cancel(job_id, force, base_path):
     """Cancel a running batch job.
 
     Cancels the job in AWS Batch and updates the manifest status.
+    Also supports cancelling retry jobs by their ID (e.g., job-id-r1).
 
     \b
     Examples:
       dh batch cancel dma-embed-20260109-a3f2
+      dh batch cancel dma-embed-20260109-a3f2-r1  # Cancel specific retry
       dh batch cancel dma-embed-20260109-a3f2 --force
     """
+    # Check if this is a retry job ID
+    parent_job_id, retry_id = _parse_retry_job_id(job_id)
+
     # Load manifest
     try:
-        manifest = load_manifest(job_id, base_path)
+        manifest = load_manifest(parent_job_id, base_path)
     except FileNotFoundError:
-        click.echo(f"Job not found: {job_id}", err=True)
+        click.echo(f"Job not found: {parent_job_id}", err=True)
         raise SystemExit(1)
 
+    # If cancelling a specific retry job
+    if retry_id:
+        _cancel_retry_job(manifest, retry_id, force, base_path)
+        return
+
     # Check if job can be cancelled
     if manifest.status in (
         JobStatus.SUCCEEDED,
@@ -94,3 +114,47 @@ def cancel(job_id, force, base_path):
     except BatchError as e:
         click.echo(click.style(f"✗ Failed to cancel job: {e}", fg="red"), err=True)
         raise SystemExit(1)
+
+
+def _cancel_retry_job(manifest, retry_id: str, force: bool, base_path: str):
+    """Cancel a specific retry job."""
+    # Find the retry info
+    retry_info = None
+    for retry in manifest.retries:
+        if retry.retry_id == retry_id:
+            retry_info = retry
+            break
+
+    if not retry_info:
+        click.echo(f"Retry job not found: {retry_id}", err=True)
+        click.echo(f"Known retries: {[r.retry_id for r in manifest.retries]}", err=True)
+        raise SystemExit(1)
+
+    if not retry_info.batch_job_id:
+        click.echo(f"Retry job {retry_id} has no AWS Batch job ID.", err=True)
+        raise SystemExit(1)
+
+    # Cancel in AWS Batch
+    try:
+        client = BatchClient()
+
+        if force:
+            click.echo(f"Terminating retry job {retry_info.batch_job_id}...")
+            client.terminate_job(
+                retry_info.batch_job_id,
+                reason="Terminated by user via dh batch cancel --force",
+            )
+        else:
+            click.echo(f"Cancelling retry job {retry_info.batch_job_id}...")
+            client.cancel_job(
+                retry_info.batch_job_id,
+                reason="Cancelled by user via dh batch cancel",
+            )
+
+        click.echo()
+        click.echo(click.style(f"✓ Retry job {retry_id} cancelled successfully", fg="green"))
+        click.echo(f"Parent job: {manifest.job_id}")
+
+    except BatchError as e:
+        click.echo(click.style(f"✗ Failed to cancel retry job: {e}", fg="red"), err=True)
+        raise SystemExit(1)

{dayhoff_tools-1.14.11 → dayhoff_tools-1.14.13}/dayhoff_tools/cli/batch/commands/status.py

@@ -174,15 +174,41 @@ def _show_job_list(user, status_filter, pipeline, base_path):
     click.echo("Use 'dh batch status <job-id>' for details.")
 
 
+def _parse_retry_job_id(job_id: str) -> tuple[str, str | None]:
+    """Parse a job ID to extract parent job ID and retry suffix.
+
+    Args:
+        job_id: Job ID like 'dma-embed-20260120-63ec' or 'dma-embed-20260120-63ec-r1'
+
+    Returns:
+        Tuple of (parent_job_id, retry_id or None)
+    """
+    import re
+
+    # Check for retry suffix like -r1, -r2, etc.
+    match = re.match(r"^(.+)(-r\d+)$", job_id)
+    if match:
+        return match.group(1), job_id
+    return job_id, None
+
+
 def _show_job_details(job_id: str, base_path: str):
     """Show detailed status for a specific job."""
+    # Check if this is a retry job ID
+    parent_job_id, retry_id = _parse_retry_job_id(job_id)
+
     try:
-        manifest = load_manifest(job_id, base_path)
+        manifest = load_manifest(parent_job_id, base_path)
     except FileNotFoundError:
         click.echo(f"Job not found: {job_id}", err=True)
-        click.echo(f"Looking in: {base_path}/{job_id}/manifest.json", err=True)
+        click.echo(f"Looking in: {base_path}/{parent_job_id}/manifest.json", err=True)
         raise SystemExit(1)
 
+    # If showing a retry job, show retry-specific details
+    if retry_id:
+        _show_retry_details(manifest, retry_id)
+        return
+
     click.echo()
     click.echo(f"Job ID:    {manifest.job_id}")
     click.echo(f"Status:    {format_status(manifest.status)}")
@@ -235,7 +261,31 @@ def _show_job_details(job_id: str, base_path: str):
         click.echo()
         click.echo("Retries:")
         for retry in manifest.retries:
-            click.echo(f"  - {retry.retry_id}: indices {retry.indices}")
+            reslice_info = ""
+            if retry.reslice_prefix:
+                reslice_info = f" (resliced to {retry.reslice_count} chunks)"
+            click.echo(f"  - {retry.retry_id}: {len(retry.indices)} indices{reslice_info}")
+            click.echo(f"    Indices: {retry.indices}")
+            if retry.batch_job_id:
+                # Show brief status for retry job
+                try:
+                    client = BatchClient()
+                    array_status = client.get_array_job_status(retry.batch_job_id)
+                    if array_status.is_complete:
+                        pct = array_status.success_rate * 100
+                        color = "green" if pct == 100 else "yellow" if pct > 90 else "red"
+                        click.echo(
+                            f"    Status: Complete - {click.style(f'{pct:.0f}%', fg=color)} "
+                            f"({array_status.succeeded}/{array_status.total} succeeded)"
+                        )
+                    else:
+                        click.echo(
+                            f"    Status: Running - {array_status.succeeded}/{array_status.total} done, "
+                            f"{array_status.running} running"
+                        )
+                except BatchError:
+                    click.echo(f"    Status: (could not fetch)")
+            click.echo(f"    Details: dh batch status {retry.retry_id}")
 
     # Suggest next steps
     click.echo()
@@ -285,3 +335,48 @@ def _show_array_status(batch_job_id: str):
 
     except BatchError as e:
         click.echo(f"    (Could not fetch live status: {e})")
+
+
+def _show_retry_details(manifest, retry_id: str):
+    """Show detailed status for a retry job."""
+    # Find the retry info
+    retry_info = None
+    for retry in manifest.retries:
+        if retry.retry_id == retry_id:
+            retry_info = retry
+            break
+
+    if not retry_info:
+        click.echo(f"Retry job not found: {retry_id}", err=True)
+        click.echo(f"Known retries: {[r.retry_id for r in manifest.retries]}", err=True)
+        raise SystemExit(1)
+
+    click.echo()
+    click.echo(f"Retry Job:   {retry_id}")
+    click.echo(f"Parent Job:  {manifest.job_id}")
+    click.echo(f"Pipeline:    {manifest.pipeline}")
+    click.echo(f"User:        {manifest.user}")
+    click.echo(
+        f"Created:     {retry_info.created.isoformat()} ({format_time_ago(retry_info.created)})"
+    )
+
+    click.echo()
+    click.echo("Retry Config:")
+    click.echo(f"  Indices:   {retry_info.indices}")
+    if retry_info.reslice_prefix:
+        click.echo(f"  Reslice:   {retry_info.reslice_prefix} ({retry_info.reslice_count} chunks)")
+    else:
+        click.echo(f"  Reslice:   No (retrying original chunks)")
+
+    if retry_info.batch_job_id:
+        click.echo()
+        click.echo("Batch:")
+        click.echo(f"  AWS Job ID: {retry_info.batch_job_id}")
+
+        # Get live status from AWS Batch
+        _show_array_status(retry_info.batch_job_id)
+
+    click.echo()
+    click.echo("Next steps:")
+    click.echo(f"  View logs:      dh batch logs {manifest.job_id}")
+    click.echo(f"  Parent status:  dh batch status {manifest.job_id}")

{dayhoff_tools-1.14.11 → dayhoff_tools-1.14.13}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "poetry.core.masonry.api"
 
 [project]
 name = "dayhoff-tools"
-version = "1.14.11"
+version = "1.14.13"
 description = "Common tools for all the repos at Dayhoff Labs"
 authors = [
     {name = "Daniel Martin-Alarcon", email = "dma@dayhofflabs.com"}