amina-cli 0.4.1__tar.gz → 0.4.2__tar.gz

{amina_cli-0.4.1 → amina_cli-0.4.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: amina-cli
-Version: 0.4.1
+Version: 0.4.2
 Summary: CLI for AminoAnalytica protein engineering platform
 Project-URL: Homepage, https://aminoanalytica.com
 Project-URL: Documentation, https://docs.aminoanalytica.com

{amina_cli-0.4.1 → amina_cli-0.4.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "amina-cli"
-version = "0.4.1"
+version = "0.4.2"
 description = "CLI for AminoAnalytica protein engineering platform"
 readme = "README.md"
 license = {text = "Apache-2.0"}

{amina_cli-0.4.1 → amina_cli-0.4.2}/src/amina_cli/__init__.py

@@ -9,4 +9,4 @@ Quick start:
     amina run esmfold --sequence "MKFLILLFNILCLFPVLAADNH"
 """
 
-__version__ = "0.4.1"
+__version__ = "0.4.2"

{amina_cli-0.4.1 → amina_cli-0.4.2}/src/amina_cli/commands/jobs_cmd.py

@@ -958,19 +958,20 @@ def reconcile(
             per_job.append(entry)
             continue
 
-        local_status = job_info.get("status", "")
-        if local_status not in TERMINAL_STATUSES:
-            # Local cache (now fresh) says this job hasn't terminated yet.
-            # Don't waste a per-job HTTP roundtrip just to confirm.
-            summary["still_running"] += 1
-            entry["action"] = "still_running"
-            entry["status"] = local_status
-            per_job.append(entry)
-            continue
-
-        # Terminal: fetch the full payload (signed_urls, output_files) so the
-        # response.json we write matches the shape `amina jobs download`
-        # produces.
+        # Always go to the server for the truth, even if the local cache says
+        # non-terminal. Reason: the bulk /list_jobs refresh above can return
+        # stale data when the worker's completion callback dropped (Supabase
+        # blip, gateway crash mid-handler) — a 300-job campaign reported 197
+        # jobs still_running while all 300 had finished on Modal. The CLI
+        # used to early-return on local non-terminal status, which made this
+        # bug unrecoverable from the client. Now reconcile mirrors what
+        # `amina jobs status` does: ask Modal directly via call_id.
+        #
+        # Cost: O(pending) HTTP calls per cycle. Bounded by jobs the user is
+        # actively waiting on (response.json-present dirs short-circuit
+        # above), and the gateway's heal-on-read upgrade makes the post-merge
+        # local cache mostly fresh anyway, so this fall-through is rarely hit
+        # in practice. It exists as a defence-in-depth for gateway brownouts.
         try:
             status_result = _resolve_job_status(job_info)
         except ToolExecutionError as fetch_err:
@@ -981,6 +982,15 @@ def reconcile(
             continue
         _persist_terminal_status(job_info, status_result)
 
+        if not status_result.get("terminal", False) and status_result.get("status") not in TERMINAL_STATUSES:
+            # Genuinely still running on the server. Count, record the
+            # FRESH status (not the stale-cache one we started with), move on.
+            summary["still_running"] += 1
+            entry["action"] = "still_running"
+            entry["status"] = status_result.get("status", "running")
+            per_job.append(entry)
+            continue
+
         # Build the response payload. For completed jobs the gateway hands us
         # a full `result` dict (output_files, signed_urls, data, ...). For
         # failed/cancelled jobs Modal often returns only an error string with