wt-runner 0.3.2__tar.gz → 0.3.3__tar.gz

{wt_runner-0.3.2 → wt_runner-0.3.3}/CHANGELOG.md

@@ -1,5 +1,10 @@
 # Changelog
 
+## v0.3.3 — 2026-11-09
+
+- `/run-from-pubsub` skips Pub/Sub redeliveries of a run that is already claimed or finished, via a create-only `lock.json` claim in the results location. A claim is never released, so a run whose attempt died is not retried ([#290](https://github.com/wildlife-dynamics/wt/pull/290))
+- Fix the Pub/Sub wait timeout so it is actually capped at `PUBSUB_ACK_MAX_TIMEOUT` (570s); a larger requested timeout previously passed through unchanged ([#290](https://github.com/wildlife-dynamics/wt/pull/290))
+
 ## v0.3.2 — 2026-11-03
 
 - Release to pick up the new `wt-invokers-gcp>=0.7.0` floor in `wt-runner-gcp`; no changes to the Python package contents

{wt_runner-0.3.2/src/wt_runner.egg-info → wt_runner-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: wt-runner
-Version: 0.3.2
+Version: 0.3.3
 Summary: FastAPI application for workflow execution using wt-invokers
 License: BSD-3-Clause
 Classifier: Development Status :: 3 - Alpha

{wt_runner-0.3.2 → wt_runner-0.3.3}/pyproject.toml

@@ -169,7 +169,7 @@ extend-immutable-calls = [
 
 [tool.pixi.package]
 name = "wt-runner"
-version = "0.3.2"
+version = "0.3.3"
 
 [tool.pixi.package.build]
 backend = { name = "pixi-build-python", version = "*", channels = ["conda-forge", "https://prefix.dev/pixi-build-backends"] }

{wt_runner-0.3.2 → wt_runner-0.3.3}/src/wt_runner/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.3.2'
-__version_tuple__ = version_tuple = (0, 3, 2)
+__version__ = version = '0.3.3'
+__version_tuple__ = version_tuple = (0, 3, 3)
 
-__commit_id__ = commit_id = 'gf28c604'
+__commit_id__ = commit_id = 'g43bf668'

{wt_runner-0.3.2 → wt_runner-0.3.3}/src/wt_runner/app.py

@@ -77,6 +77,7 @@ except ImportError:
     ecoscope_get_results_json = None
 
 import obstore
+from obstore.exceptions import AlreadyExistsError
 
 # Invoker registry mapping invoker names to classes
 INVOKERS: dict[str, type[AbstractInvoker]] = {
@@ -92,6 +93,7 @@ TIMEOUT_EXPIRED_ERROR_MSG = (
     "The workflow timed out. Consider reducing the amount of data being processed."
 )
 PUBSUB_ACK_MAX_TIMEOUT = 570  # seconds
+RUN_LOCK_FILENAME = "lock.json"
 
 
 async def get_results_json(results_url: str) -> dict[str, Any]:
@@ -484,6 +486,12 @@ async def run_from_pubsub(
         "run-from-pubsub-endpoint",
         attributes=asdict(span_attributes),
     ):
+        if not await claim_run(invoker_params["results_url"]):
+            logging.info(  # noqa: LOG015  # routed through root logger by design
+                "Skipping redelivered run %s: it already has a claim or a result.",
+                span_attributes.workflow_run_id,
+            )
+            return {"status": "skipped"}
         trace_context = build_context_headers()
         invoker_params["extra_env"] |= {k.upper(): v for k, v in trace_context.items()}
         invoker_params["otel_exporter"] = get_otel_exporter()
@@ -507,7 +515,7 @@ async def run_from_pubsub(
                 timeout = command_payload.invoker_kwargs.get("timeout", PUBSUB_ACK_MAX_TIMEOUT)
                 # Maximum timeout when running from PubSub is 10 minutes.
                 # It's set to a little bit less to have time to cancel and handle the error
-                timeout = min(timeout, max(timeout, PUBSUB_ACK_MAX_TIMEOUT))
+                timeout = min(timeout, PUBSUB_ACK_MAX_TIMEOUT)
                 exit_code = await invoker.wait(timeout=timeout, error_msg=TIMEOUT_EXPIRED_ERROR_MSG)
                 if exit_code != 0:
                     raise RuntimeError(f"Workflow invoker failed with exit code {exit_code}.")
@@ -590,6 +598,28 @@ def prepare_invoker_parameters(
     )  # Extra kwargs are passed to the invoker
 
 
+async def claim_run(results_url: str) -> bool:
+    """Claim a run for this delivery, or return False if it was already claimed or finished.
+
+    Pub/Sub push delivers at least once, so a run can arrive again after a deploy or a lost ack.
+    A create-only put (GCS ``ifGenerationMatch=0``) lets exactly one delivery win without
+    needing delete permission on the results bucket. A claim is never released, so a delivery
+    whose attempt died is not retried; the server's run timeout covers that case.
+    """
+    store = obstore.store.from_url(results_url)
+    try:
+        await store.head_async("result.json")
+    except FileNotFoundError:
+        pass
+    else:
+        return False  # finished before claims existed, or the claim's attempt wrote it
+    try:
+        await store.put_async(RUN_LOCK_FILENAME, b"{}", mode="create")
+    except AlreadyExistsError:
+        return False
+    return True
+
+
 async def upload_error_to_gcs(error_details: dict[str, Any], results_url: str) -> None:
     """Upload error details to Google Cloud Storage.
 

{wt_runner-0.3.2 → wt_runner-0.3.3/src/wt_runner.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: wt-runner
-Version: 0.3.2
+Version: 0.3.3
 Summary: FastAPI application for workflow execution using wt-invokers
 License: BSD-3-Clause
 Classifier: Development Status :: 3 - Alpha

wt_runner-0.3.3/src/wt_runner.egg-info/scm_version.json

@@ -0,0 +1,8 @@
+{
+  "tag": "0.3.3",
+  "distance": 0,
+  "node": "g43bf668",
+  "dirty": false,
+  "branch": "HEAD",
+  "node_date": "2026-11-09"
+}

wt_runner-0.3.2/src/wt_runner.egg-info/scm_version.json

@@ -1,8 +0,0 @@
-{
-  "tag": "0.3.2",
-  "distance": 0,
-  "node": "gf28c604",
-  "dirty": false,
-  "branch": "HEAD",
-  "node_date": "2026-11-03"
-}