Remove repository config caching logic

.github/workflows/requirements/unit-tests.txt

@@ -1,6 +1,5 @@
 aiohttp==3.14.3
 build==1.5.0
-cachetools==7.1.7
 coverage==7.15.4
 gidgethub==5.4.0
 gidgetlab==2.1.2

pyproject.toml

@@ -8,9 +8,8 @@ version = "0.0.1"
 dependencies = [
     "aiohttp",
     "aiojobs",
-    "cachetools",
-    "gidgethub",
-    "gidgetlab>=2.1.2",
+    "gidgethub[aiohttp]",
+    "gidgetlab[aiohttp]>=2.1.2",
     "pydantic",
     "pydantic-settings",
     "pyjwt",

spack/repos/spack_repo/hubcast/packages/py_hubcast/package.py

@@ -26,7 +26,7 @@ class PyHubcast(PythonPackage):
     depends_on("py-aiohttp", type=("build", "run"))
     depends_on("py-aiojobs", type=("build", "run"))
     depends_on("py-pyjwt", type=("build", "run"))
-    depends_on("py-gidgethub", type=("build", "run"))
+    depends_on("py-gidgethub+aiohttp", type=("build", "run"))
     depends_on("py-gidgetlab@2.1.2:+aiohttp", type=("build", "run"))
     depends_on("py-repligit", type=("build", "run"))
     depends_on("py-pyyaml", type=("build", "run"))

src/hubcast/web/github/routes.py

@@ -279,14 +279,11 @@ async def sync_branch(
         log.info("Skipped branch sync - branch has open PR")
         return
 
-    # only refresh config when a default branch push touches .github/hubcast.yml
     default_branch = event.data["repository"]["default_branch"]
     is_default_branch = sync_ref == f"refs/heads/{default_branch}"
     config_changed = gh.repo_config_path in changed_files_from_push(event.data)
     try:
-        repo_config = await get_repo_config(
-            gh, src_fullname, refresh=is_default_branch and config_changed
-        )
+        repo_config = await get_repo_config(gh)
     except RepoConfigError as exc:
         await report_config_error(gh, want_sha, exc)
         return
@@ -296,13 +293,10 @@ async def sync_branch(
     head_commit = event.data.get("head_commit")
     commit_msg = head_commit["message"] if head_commit else ""
 
-    # only set/update webhook on default branch pushes when the push actually
-    # touched the config file (config_changed above); this avoids spurious
-    # permission errors when the config merely aged out of the cache
-    # we also give maintainers the option to force-set the webhook: if the
-    # commit message contains [hubcast config], we'll set the webhook
+    # only set/update the webhook on default branch pushes that touch the config
+    # file (avoids spurious permission errors on unrelated pushes); maintainers
+    # can also force it by including [hubcast config] in the commit message
     if is_default_branch and (config_changed or "[hubcast config]" in commit_msg):
-        # setup callback webhook on GitLab
         try:
             await gl.set_webhook(
                 dest_org=repo_config.dest_org,
@@ -369,12 +363,11 @@ async def remove_branch(
     *arg,
     **kwargs,
 ) -> None:
-    src_fullname = event.data["repository"]["full_name"]
     sync_ref = event.data["ref"]
 
     update_log_context(ref=sync_ref)
 
-    repo_config = await get_repo_config(gh, src_fullname)
+    repo_config = await get_repo_config(gh)
 
     dest_fullname = repo_config.dest_fullname
     dest_remote_url = f"{gl.instance_url}/{dest_fullname}.git"
@@ -420,7 +413,7 @@ async def sync_pr(
 
     # get the repository configuration from .github/hubcast.yml
     try:
-        repo_config = await get_repo_config(gh, base_fullname)
+        repo_config = await get_repo_config(gh)
     except RepoConfigError as exc:
         await report_config_error(gh, want_sha, exc)
         return
@@ -531,7 +524,7 @@ async def remove_pr(
     base_fullname = pull_request["base"]["repo"]["full_name"]
 
     # get the repository configuration from .github/hubcast.yml
-    repo_config = await get_repo_config(gh, base_fullname)
+    repo_config = await get_repo_config(gh)
 
     if not repo_config.delete_closed:
         log.info("Skipped PR branch removal - delete_closed disabled")
@@ -640,13 +633,11 @@ async def respond_comment(
         pull_request = await gh.get_pr(pr_number)
 
         # get the branch this PR belongs to
-        base_fullname = pull_request["base"]["repo"]["full_name"]
         branch = _pr_branch_name(pull_request)
-
         update_log_context(branch=branch)
 
         # get the gitlab repo information and run the pipeline
-        repo_config = await get_repo_config(gh, base_fullname)
+        repo_config = await get_repo_config(gh)
         dest_fullname = repo_config.dest_fullname
 
         try:
@@ -676,13 +667,11 @@ async def respond_comment(
         pull_request = await gh.get_pr(pr_number)
 
         # get the branch this PR belongs to
-        base_fullname = pull_request["base"]["repo"]["full_name"]
         branch = _pr_branch_name(pull_request)
-
         update_log_context(branch=branch)
 
         # get the gitlab repo information and run the pipeline
-        repo_config = await get_repo_config(gh, base_fullname)
+        repo_config = await get_repo_config(gh)
         dest_fullname = repo_config.dest_fullname
 
         try:
@@ -766,7 +755,6 @@ async def rerun_check(
     Handles a user re-running a check run by retrying the specific GitLab job or pipeline it's attached to.
     See https://docs.github.com/en/webhooks/webhook-events-and-payloads?actionType=rerequested#check_run.
     """
-    src_fullname = event.data["repository"]["full_name"]
     check_run_commit = event.data["check_run"]["head_sha"]
     details_url = event.data["check_run"]["details_url"]
     update_log_context(
@@ -783,7 +771,7 @@ async def rerun_check(
         return
 
     try:
-        repo_config = await get_repo_config(gh, src_fullname)
+        repo_config = await get_repo_config(gh)
     except RepoConfigError as exc:
         await report_config_error(gh, check_run_commit, exc)
         return

src/hubcast/web/github/utils.py

@@ -2,7 +2,6 @@
 from typing import Any
 
 import yaml
-from cachetools import TTLCache
 
 from hubcast.clients.github import GitHubClient
 from hubcast.exceptions import HubcastError, RepoConfigError
@@ -14,9 +13,6 @@
 
 log = logging.getLogger(__name__)
 
-# Shared cache for repository configs with 30-minute TTL
-config_cache: TTLCache[str, RepoConfig | None] = TTLCache(maxsize=1000, ttl=1800)
-
 
 def changed_files_from_push(payload: dict[str, Any]) -> set[str]:
     """Collect all file paths touched by the commits in a push payload."""
@@ -28,15 +24,15 @@ def changed_files_from_push(payload: dict[str, Any]) -> set[str]:
     }
 
 
-async def get_repo_config(
-    gh: GitHubClient, fullname: str, refresh: bool = False
-) -> RepoConfig:
-    """Get repository configuration from cache or fetch from GitHub.
+async def get_repo_config(gh: GitHubClient) -> RepoConfig:
+    """Fetch and validate the repository configuration from GitHub.
+
+    Fetched fresh on every event so all replicas always see the current
+    destination repo; a single contents-API call is well within App
+    installation rate limits.
 
     Args:
         gh: GitHub client instance
-        fullname: Full repository name (e.g., "owner/repo")
-        refresh: Whether to force refresh from GitHub
 
     Returns:
         RepoConfig instance
@@ -45,24 +41,12 @@ async def get_repo_config(
         HubcastError: If config file contains invalid YAML, is missing required keys,
             or has validation errors
     """
-    # check cache first unless refresh is requested
-    if fullname in config_cache and not refresh:
-        config = config_cache[fullname]
-        log.info("Repo config retrieved from cache")
-        if config is None:
-            # raise so route handlers can't continue
-            # we don't want to raise this as a RepoConfigError because telling users
-            # about the absence of the config will create noise and confusion
-            raise HubcastError("Repo config file not found", log_level="INFO")
-        return config
-
-    # cache miss or refresh requested, fetch from GH
     fetched_config = await gh.get_repo_config()
 
     if fetched_config is None:  # 404
-        config_cache[fullname] = None
-        log.info("Cached absence of repo config")
         # raise so route handlers can't continue
+        # we don't want to raise this as a RepoConfigError because telling users
+        # about the absence of the config will create noise and confusion
         raise HubcastError("Repo config file not found", log_level="INFO")
 
     # parse and validate YAML
@@ -86,6 +70,5 @@ async def get_repo_config(
             error=str(e),
         )
 
-    config_cache[fullname] = config
     log.info("Repo config fetched from source forge")
     return config

tests/test_github_routes.py

@@ -414,38 +414,6 @@ async def test_sync_branch_object_present_but_ref_missing(
     mock_repligit_ops["send_pack"].assert_awaited_once()
 
 
-@pytest.mark.asyncio
-@pytest.mark.parametrize(
-    "ref,modified,expected_refresh",
-    [
-        ("refs/heads/main", [".github/hubcast.yml"], True),
-        ("refs/heads/main", ["src/app.py"], False),
-        ("refs/heads/feature", [".github/hubcast.yml"], False),
-    ],
-)
-async def test_sync_branch_config_refresh(
-    ref,
-    modified,
-    expected_refresh,
-    mock_push_event,
-    mock_gh,
-    mock_gl,
-    mock_repligit_ops,
-):
-    """Config should only be refreshed when a default branch push touches the config file."""
-
-    mock_push_event.data["ref"] = ref
-    mock_push_event.data["commits"] = [
-        {"added": [], "modified": modified, "removed": []}
-    ]
-
-    await sync_branch(event=mock_push_event, gh=mock_gh, gl=mock_gl, gl_user="gl-user")
-
-    mock_repligit_ops["get_repo_config"].assert_awaited_once_with(
-        mock_gh, "owner/repo", refresh=expected_refresh
-    )
-
-
 @pytest.mark.asyncio
 @pytest.mark.parametrize(
     "ref,modified,commit_msg,webhook_expected",