Gracefully handle HTTP errors

gantry/clients/gitlab.py

@@ -17,7 +17,7 @@ async def _request(self, url: str, response_type: str) -> dict | str:
         returns: the response from Gitlab in the specified format
         """
 
-        async with aiohttp.ClientSession(raise_for_status=True) as session:
+        async with aiohttp.ClientSession() as session:
             async with session.get(url, headers=self.headers) as resp:
                 if response_type == "json":
                     return await resp.json()

gantry/clients/prometheus/prometheus.py

@@ -68,17 +68,17 @@ async def query_range(self, query: str | dict, start: int, end: int) -> list:
 
     async def _query(self, url: str) -> list:
         """Query Prometheus with a query string"""
-        async with aiohttp.ClientSession(raise_for_status=True) as session:
+        async with aiohttp.ClientSession() as session:
             # submit cookie with request
             async with session.get(url, cookies=self.cookies) as resp:
                 try:
                     return await resp.json()
                 except aiohttp.ContentTypeError:
-                    logger.error(
-                        """Prometheus query failed with unexpected response.
-                        The cookie may have expired."""
+                    # this will get caught in collection.py and fetch_job won't continue
+                    raise aiohttp.ClientError(
+                        """Prometheus query failed with unexpected
+                        response, cookie may have expired."""
                     )
-                    return {}
 
     def prettify_res(self, response: dict) -> list:
         """Process Prometheus response into a list of dicts with {label: value}"""

gantry/routes/collection.py

@@ -1,6 +1,7 @@
 import logging
 import re
 
+import aiohttp
 import aiosqlite
 
 from gantry.clients import db
@@ -56,20 +57,25 @@ async def fetch_job(
     ):
         return
 
-    # check if the job is a ghost
-    job_log = await gitlab.job_log(job.gl_id)
-    is_ghost = "No need to rebuild" in job_log
-    if is_ghost:
-        logger.warning(f"job {job.gl_id} is a ghost, skipping")
-        return
-
     try:
+        # all code that makes HTTP requests should be in this try block
+
+        # check if the job is a ghost
+        job_log = await gitlab.job_log(job.gl_id)
+        is_ghost = "No need to rebuild" in job_log
+        if is_ghost:
+            logger.warning(f"job {job.gl_id} is a ghost, skipping")
+            return
+
         annotations = await prometheus.job.get_annotations(job.gl_id, job.midpoint)
         resources, node_hostname = await prometheus.job.get_resources(
             annotations["pod"], job.midpoint
         )
         usage = await prometheus.job.get_usage(annotations["pod"], job.start, job.end)
         node_id = await fetch_node(db_conn, prometheus, node_hostname, job.midpoint)
+    except aiohttp.ClientError as e:
+        logger.error(f"Request failed: {e}")
+        return
     except IncompleteData as e:
         # missing data, skip this job
         logger.error(f"{e} job={job.gl_id}")