fix: limit dagger-data volume growth by pruning named caches (#193)

The plain `prune` operation only removes unreferenced execution-cache
entries; named cache volumes (gradle-cache, go-build-cache, go-mod-cache,
pip-cache) are never reclaimed and grow without bound, causing the
dagger-data Docker volume to reach 200 GB over time.

Pass maxUsedSpace/targetSpace arguments to `prune` so Dagger also
reclaims named volumes when the total cache exceeds 75 GB, targeting
50 GB. Use an aggressive targetSpace of 20 GB for disk-space emergency
recovery in check-dagger. Add a dagger-prune task for manual use.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.forgejo/workflows/ci.yml

@@ -55,7 +55,10 @@ jobs:
       - name: Prune Dagger cache before check
         env:
           DAGGER_NO_NAG: "1"
-        run: dagger query '{ engine { localCache { prune } } }' 2>/dev/null || true
+        # prune(maxUsedSpace) also reclaims named cache volumes (gradle-cache, go-build-cache, etc.)
+        # when total cache exceeds the limit; without args only unreferenced entries are removed.
+        run: |
+          dagger query '{ engine { localCache { prune(maxUsedSpace: "75gb", targetSpace: "50gb") } } }' || true
 
       - name: Run Full Check Suite
         env:
@@ -66,7 +69,8 @@ jobs:
         if: always()
         env:
           DAGGER_NO_NAG: "1"
-        run: dagger query '{ engine { localCache { prune } } }' 2>/dev/null || true
+        run: |
+          dagger query '{ engine { localCache { prune(maxUsedSpace: "75gb", targetSpace: "50gb") } } }' || true
 
       - name: Cleanup TLS credentials
         if: always()

Taskfile.yml

@@ -288,7 +288,7 @@ tasks:
               echo "$(_ts) dagger: network error on attempt $attempt/3, retrying..." >&2
             elif [ "$attempt" -lt 3 ] && grep -q "No space left on device" "$DAGGER_OUT"; then
               echo "$(_ts) dagger: disk space error on attempt $attempt/3, pruning Dagger cache..." >&2
-              dagger query '{ engine { localCache { prune } } }' 2>/dev/null || true
+              dagger query '{ engine { localCache { prune(targetSpace: "20gb") } } }' 2>/dev/null || true
               echo "$(_ts) dagger: waiting 90s for freed space to settle..." >&2
               sleep 90
             else
@@ -320,6 +320,12 @@ tasks:
         wait "$RECV_PID" 2>/dev/null || true
         exit $RC
 
+  dagger-prune:
+    desc: Prune the Dagger engine cache (keeps named volumes unless total exceeds 75 GB, then targets 50 GB)
+    cmds:
+      - |
+        dagger query '{ engine { localCache { prune(maxUsedSpace: "75gb", targetSpace: "50gb") } } }'
+
   integration-android:
     desc: UI integration tests on a connected Android emulator (Stalwart on host, emulator reaches it via 10.0.2.2)
     deps: [_preflight, _android-sdk-check, _android-avd-setup]

scripts/deploy_playstore.py

@@ -4,6 +4,7 @@
 import json
 import os
 import sys
+import time
 
 import google_auth_httplib2
 import httplib2
@@ -15,6 +16,7 @@ PACKAGE_NAME = "de.sharedinbox.mua"
 AAB_PATH = "build/app/outputs/bundle/release/app-release.aab"
 TRACK = "internal"
 _TIMEOUT = 300  # seconds — AAB uploads can be large
+_MAX_UPLOAD_ATTEMPTS = 3
 
 
 def main():
@@ -40,14 +42,38 @@ def main():
     edit = service.edits().insert(body={}, packageName=PACKAGE_NAME).execute(num_retries=3)
     edit_id = edit["id"]
 
-    media = MediaFileUpload(AAB_PATH, mimetype="application/octet-stream", resumable=True)
-    bundle = (
-        service.edits()
-        .bundles()
-        .upload(packageName=PACKAGE_NAME, editId=edit_id, media_body=media)
-        .execute(num_retries=3)
-    )
-    version_code = bundle["versionCode"]
+    # The resumable upload can fail with RedirectMissingLocation on transient
+    # network hiccups. Retry with a fresh MediaFileUpload each time (resumable
+    # uploads can't reuse the same object) using exponential backoff.
+    version_code = None
+    last_exc = None
+    for attempt in range(_MAX_UPLOAD_ATTEMPTS):
+        try:
+            media = MediaFileUpload(
+                AAB_PATH, mimetype="application/octet-stream", resumable=True
+            )
+            bundle = (
+                service.edits()
+                .bundles()
+                .upload(packageName=PACKAGE_NAME, editId=edit_id, media_body=media)
+                .execute(num_retries=3)
+            )
+            version_code = bundle["versionCode"]
+            break
+        except httplib2.error.RedirectMissingLocation as exc:
+            last_exc = exc
+            if attempt < _MAX_UPLOAD_ATTEMPTS - 1:
+                delay = 10 * (2 ** attempt)
+                print(
+                    f"Upload attempt {attempt + 1} failed (redirect error), "
+                    f"retrying in {delay}s…"
+                )
+                time.sleep(delay)
+    else:
+        raise RuntimeError(
+            f"AAB upload failed after {_MAX_UPLOAD_ATTEMPTS} attempts"
+        ) from last_exc
+
     print(f"Uploaded AAB, version code: {version_code}")
 
     service.edits().tracks().update(

scripts/test_deploy_playstore.py

@@ -88,5 +88,112 @@ class TestMainHappyPath(unittest.TestCase):
         mock_http_cls.assert_called_once_with(timeout=deploy_playstore._TIMEOUT)
 
 
+def _redirect_error():
+    import httplib2
+    return httplib2.error.RedirectMissingLocation("redirect missing", {}, b"")
+
+
+class TestUploadRetry(unittest.TestCase):
+    def _make_mock_service(self, upload_side_effects):
+        mock_service = MagicMock()
+        mock_edits = mock_service.edits.return_value
+        mock_edits.insert.return_value.execute.return_value = {"id": "edit-1"}
+        mock_edits.bundles.return_value.upload.return_value.execute.side_effect = (
+            upload_side_effects
+        )
+        mock_edits.tracks.return_value.update.return_value.execute.return_value = {}
+        mock_edits.commit.return_value.execute.return_value = {}
+        return mock_service, mock_edits
+
+    def _run_with_service(self, mock_service):
+        fake_config = '{"type":"service_account"}'
+        with patch.dict(os.environ, {"PLAY_STORE_CONFIG_JSON": fake_config}):
+            with patch("deploy_playstore.os.path.exists", return_value=True):
+                with patch("deploy_playstore.service_account.Credentials.from_service_account_info"):
+                    with patch("deploy_playstore.google_auth_httplib2.AuthorizedHttp"):
+                        with patch("deploy_playstore.build", return_value=mock_service):
+                            with patch("deploy_playstore.MediaFileUpload"):
+                                with patch("deploy_playstore.time.sleep"):
+                                    deploy_playstore.main()
+
+    def test_succeeds_on_first_attempt(self):
+        mock_service, mock_edits = self._make_mock_service([{"versionCode": 5}])
+        self._run_with_service(mock_service)
+        mock_edits.bundles.return_value.upload.return_value.execute.assert_called_once_with(
+            num_retries=3
+        )
+
+    def test_retries_once_on_redirect_error_then_succeeds(self):
+        mock_service, mock_edits = self._make_mock_service(
+            [_redirect_error(), {"versionCode": 9}]
+        )
+
+        with patch.dict(os.environ, {"PLAY_STORE_CONFIG_JSON": '{"type":"service_account"}'}):
+            with patch("deploy_playstore.os.path.exists", return_value=True):
+                with patch("deploy_playstore.service_account.Credentials.from_service_account_info"):
+                    with patch("deploy_playstore.google_auth_httplib2.AuthorizedHttp"):
+                        with patch("deploy_playstore.build", return_value=mock_service):
+                            with patch("deploy_playstore.MediaFileUpload") as mock_media_cls:
+                                with patch("deploy_playstore.time.sleep") as mock_sleep:
+                                    deploy_playstore.main()
+
+        self.assertEqual(
+            mock_edits.bundles.return_value.upload.return_value.execute.call_count, 2
+        )
+        mock_sleep.assert_called_once_with(10)
+        self.assertEqual(mock_media_cls.call_count, 2)
+
+    def test_raises_after_all_attempts_exhausted(self):
+        mock_service, _ = self._make_mock_service(
+            [_redirect_error(), _redirect_error(), _redirect_error()]
+        )
+
+        with patch.dict(os.environ, {"PLAY_STORE_CONFIG_JSON": '{"type":"service_account"}'}):
+            with patch("deploy_playstore.os.path.exists", return_value=True):
+                with patch("deploy_playstore.service_account.Credentials.from_service_account_info"):
+                    with patch("deploy_playstore.google_auth_httplib2.AuthorizedHttp"):
+                        with patch("deploy_playstore.build", return_value=mock_service):
+                            with patch("deploy_playstore.MediaFileUpload"):
+                                with patch("deploy_playstore.time.sleep"):
+                                    with self.assertRaises(RuntimeError) as ctx:
+                                        deploy_playstore.main()
+
+        self.assertIn(
+            str(deploy_playstore._MAX_UPLOAD_ATTEMPTS), str(ctx.exception)
+        )
+
+    def test_backoff_delays_are_10s_then_20s(self):
+        mock_service, _ = self._make_mock_service(
+            [_redirect_error(), _redirect_error(), {"versionCode": 3}]
+        )
+
+        with patch.dict(os.environ, {"PLAY_STORE_CONFIG_JSON": '{"type":"service_account"}'}):
+            with patch("deploy_playstore.os.path.exists", return_value=True):
+                with patch("deploy_playstore.service_account.Credentials.from_service_account_info"):
+                    with patch("deploy_playstore.google_auth_httplib2.AuthorizedHttp"):
+                        with patch("deploy_playstore.build", return_value=mock_service):
+                            with patch("deploy_playstore.MediaFileUpload"):
+                                with patch("deploy_playstore.time.sleep") as mock_sleep:
+                                    deploy_playstore.main()
+
+        mock_sleep.assert_has_calls([call(10), call(20)])
+
+    def test_fresh_media_upload_created_on_each_attempt(self):
+        mock_service, _ = self._make_mock_service(
+            [_redirect_error(), {"versionCode": 2}]
+        )
+
+        with patch.dict(os.environ, {"PLAY_STORE_CONFIG_JSON": '{"type":"service_account"}'}):
+            with patch("deploy_playstore.os.path.exists", return_value=True):
+                with patch("deploy_playstore.service_account.Credentials.from_service_account_info"):
+                    with patch("deploy_playstore.google_auth_httplib2.AuthorizedHttp"):
+                        with patch("deploy_playstore.build", return_value=mock_service):
+                            with patch("deploy_playstore.MediaFileUpload") as mock_media_cls:
+                                with patch("deploy_playstore.time.sleep"):
+                                    deploy_playstore.main()
+
+        self.assertEqual(mock_media_cls.call_count, 2)
+
+
 if __name__ == "__main__":
     unittest.main()