common: fix three bugs in the timeline cache

1. Double-check locking (TOCTOU)
   GetTimeline dropped the read lock, then acquired the write lock to
   fetch fresh data, but never re-checked the cache after the write lock
   was acquired.  A second goroutine could pass the stale-check under the
   read lock at the same time, resulting in both goroutines fetching the
   full timeline independently and overwriting each other's result.  Add
   a re-check immediately after acquiring the write lock so only the
   first writer fetches.

2. LastCachedTime used the wrong item
   The incremental fetch used data[0].Updated as the high-water mark for
   the Since filter, but data is sorted in descending Created order so
   data[0] is the most-recently-created entry, whose Updated timestamp
   may not be the latest across all cached items.  Switch to a linear
   scan over all items to find the true maximum Updated value, matching
   the fix in origin/fix/timeline-cache-race.

3. Cache not invalidated after RequestReviews / UnrequestReview
   After the bot called RequestReviews or UnrequestReview, the timeline
   cache for that PR was left hot for up to 5 seconds.  Any subsequent
   GetTimeline call within that window returned stale data that did not
   include the just-created review_requested / review_request_removed
   entry, causing FindMissingAndExtraReviewers to make decisions based on
   outdated reviewer state.  Call ResetTimelineCache on success in both
   methods so the next read always sees the mutation.

.gitea/workflows/t.yaml

@@ -34,7 +34,9 @@ jobs:
         run: make build
         working-directory: ./autogits
       - name: Prepare images
-        run: make build
+        run: |
+          make build
+          podman rmi $(podman images -f "dangling=true" -q)
         working-directory: ./autogits/integration
       - name: Make sure the pod is down
         run: make down
@@ -43,12 +45,16 @@ jobs:
         run: |
           make up
           make wait_healthy
+          podman ps
+          sleep 5
         working-directory: ./autogits/integration
       - name: Run tests
         run: make pytest
         working-directory: ./autogits/integration
       - name: Make sure the pod is down
         if: always()
-        run: make down
+        run: |
+          podman ps
+          make down
         working-directory: ./autogits/integration
 

Makefile

@@ -1,4 +1,8 @@
 MODULES := devel-importer utils/hujson utils/maintainer-update gitea-events-rabbitmq-publisher gitea_status_proxy group-review obs-forward-bot obs-staging-bot obs-status-service workflow-direct workflow-pr
 
-build:
-	for m in $(MODULES); do go build -C $$m -buildmode=pie || exit 1 ; done
+.PHONY: build $(MODULES)
+
+build: $(MODULES)
+
+$(MODULES):
+	go build -C $@ -buildmode=pie

common/config.go

@@ -23,6 +23,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log"
 	"os"
 	"slices"
 	"strings"
@@ -204,21 +205,16 @@ func ReadWorkflowConfig(gitea GiteaFileContentAndRepoFetcher, git_project string
 
 func ResolveWorkflowConfigs(gitea GiteaFileContentAndRepoFetcher, config *ConfigFile) (AutogitConfigs, error) {
 	configs := make([]*AutogitConfig, 0, len(config.GitProjectNames))
-	var errs []error
 	for _, git_project := range config.GitProjectNames {
 		c, err := ReadWorkflowConfig(gitea, git_project)
 		if err != nil {
 			// can't sync, so ignore for now
-			errs = append(errs, err)
+			log.Println(err)
 		} else {
 			configs = append(configs, c)
 		}
 	}
 
-	if len(errs) > 0 {
-		return configs, errors.Join(errs...)
-	}
-
 	return configs, nil
 }
 

common/gitea_utils.go

@@ -768,6 +768,10 @@ func (gitea *GiteaTransport) RequestReviews(pr *models.PullRequest, reviewers ..
 		return nil, fmt.Errorf("Cannot create pull request reviews: %w", err)
 	}
 
+	// Invalidate the timeline cache so the next GetTimeline call reflects
+	// the newly created review_requested entry.
+	gitea.ResetTimelineCache(pr.Base.Repo.Owner.UserName, pr.Base.Repo.Name, pr.Index)
+
 	return review.GetPayload(), nil
 }
 
@@ -776,6 +780,13 @@ func (gitea *GiteaTransport) UnrequestReview(org, repo string, id int64, reviwer
 		repository.NewRepoDeletePullReviewRequestsParams().WithOwner(org).WithRepo(repo).WithIndex(id).WithBody(&models.PullReviewRequestOptions{
 			Reviewers: reviwers,
 		}), gitea.transport.DefaultAuthentication)
+
+	if err == nil {
+		// Invalidate the timeline cache so the next GetTimeline call reflects
+		// the newly created review_request_removed entry.
+		gitea.ResetTimelineCache(org, repo, id)
+	}
+
 	return err
 }
 
@@ -861,24 +872,31 @@ func (gitea *GiteaTransport) GetTimeline(org, repo string, idx int64) ([]*models
 	prID := fmt.Sprintf("%s/%s!%d", org, repo, idx)
 	giteaTimelineCacheMutex.RLock()
 	TimelineCache, IsCached := giteaTimelineCache[prID]
-	var LastCachedTime strfmt.DateTime
-	if IsCached {
-		l := len(TimelineCache.data)
-		if l > 0 {
-			LastCachedTime = TimelineCache.data[0].Updated
-		}
-
-		// cache data for 5 seconds
-		if TimelineCache.lastCheck.Add(time.Second*5).Compare(time.Now()) > 0 {
-			giteaTimelineCacheMutex.RUnlock()
-			return TimelineCache.data, nil
-		}
+	if IsCached && TimelineCache.lastCheck.Add(time.Second*5).Compare(time.Now()) > 0 {
+		giteaTimelineCacheMutex.RUnlock()
+		return TimelineCache.data, nil
 	}
 	giteaTimelineCacheMutex.RUnlock()
 
 	giteaTimelineCacheMutex.Lock()
 	defer giteaTimelineCacheMutex.Unlock()
 
+	// Re-read after acquiring the write lock: another goroutine may have
+	// already refreshed the cache while we were waiting.
+	TimelineCache, IsCached = giteaTimelineCache[prID]
+	if IsCached && TimelineCache.lastCheck.Add(time.Second*5).Compare(time.Now()) > 0 {
+		return TimelineCache.data, nil
+	}
+
+	// Find the highest Updated timestamp across all cached items so the
+	// incremental fetch picks up both new entries and modified ones.
+	var LastCachedTime strfmt.DateTime
+	for _, d := range TimelineCache.data {
+		if time.Time(d.Updated).Compare(time.Time(LastCachedTime)) > 0 {
+			LastCachedTime = d.Updated
+		}
+	}
+
 	for resCount > 0 {
 		opts := issue.NewIssueGetCommentsAndTimelineParams().WithOwner(org).WithRepo(repo).WithIndex(idx).WithPage(&page)
 		if !LastCachedTime.IsZero() {

group-review/main.go

@@ -327,7 +327,6 @@ func main() {
 	interval := flag.Int64("interval", 10, "Notification polling interval in minutes (min 1 min)")
 	configFile := flag.String("config", "", "PrjGit listing config file")
 	logging := flag.String("logging", "info", "Logging level: [none, error, info, debug]")
-	exitOnConfigError := flag.Bool("exit-on-config-error", false, "Exit if any repository in configuration cannot be resolved")
 	flag.BoolVar(&common.IsDryRun, "dry", false, "Dry run, no effect. For debugging")
 	flag.Parse()
 
@@ -383,10 +382,8 @@ func main() {
 	giteaTransport := common.AllocateGiteaTransport(*giteaUrl)
 	configs, err := common.ResolveWorkflowConfigs(giteaTransport, configData)
 	if err != nil {
-		common.LogError("Failed to resolve some configuration repositories:", err)
-		if *exitOnConfigError {
-			return
-		}
+		common.LogError("Cannot parse workflow configs:", err)
+		return
 	}
 
 	reviewer, err := giteaTransport.GetCurrentUser()

integration/Makefile

@@ -70,7 +70,7 @@ wait_healthy:
 	@echo "All services are healthy!"
 
 pytest:
-	podman-compose exec tester pytest -v tests/*
+	podman-compose exec tester pytest -v tests
 
 build:
 	podman pull docker.io/library/rabbitmq:3.13.7-management

integration/podman-compose.yml

@@ -81,6 +81,11 @@ services:
     init: true
     networks:
       - gitea-network
+    healthcheck:
+      test: ["CMD-SHELL", "pgrep workflow-pr && [ $$(awk '{print $22}' /proc/$$(pgrep workflow-pr)/stat) -lt $$(($$(awk '{print $1}' /proc/uptime | cut -d. -f1)*100 - 1000)) ]"]
+      interval: 5s
+      timeout: 5s
+      retries: 5
     depends_on:
       gitea:
         condition: service_started
@@ -96,7 +101,6 @@ services:
       - ./workflow-pr-repos:/var/lib/workflow-pr/repos:Z
     command: [ 
       "-check-on-start",
-      "-exit-on-config-error",
       "-debug", 
       "-gitea-url", "http://gitea-test:3000", 
       "-url", "amqps://rabbitmq-test:5671", 
@@ -130,9 +134,7 @@ services:
     networks:
       - gitea-network
     depends_on:
-      gitea:
-        condition: service_started
-      tester:
+      workflow-pr:
         condition: service_started
     environment:
       - OBS_USER=mock

integration/tests/conftest.py

@@ -162,6 +162,10 @@ BRANCH_CONFIG_CUSTOM = {
         "workflow.config": {
             "MergeMode": "devel"
         }
+    },
+    "zz-ready-to-start": {
+        "workflow.config": {
+        }
     }
 }
 

integration/tests/lib/common_test_utils.py

@@ -129,7 +129,8 @@ class GiteaAPIClient:
                     "gitignores": "Go",
                     "license": "MIT",
                     "private": False,
-                    "readme": "Default"
+                    "readme": "Default",
+                    "object_format_name": "sha256"
                 }
                 response, duration = self._request("POST", f"orgs/{org_name}/repos", json=data)
                 vprint(f"[{duration:.3f}s] Repository '{org_name}/{repo_name}' created with a README.")
@@ -179,7 +180,7 @@ class GiteaAPIClient:
 
         diff_content = f"""diff --git a/.gitmodules b/.gitmodules
 new file mode 100644
-index 0000000..f1838bd
+index 00000000..f1838bd9
 --- /dev/null
 +++ b/.gitmodules
 @@ -0,0 +1,6 @@
@@ -191,14 +192,14 @@ index 0000000..f1838bd
 +	url = ../../mypool/pkgB.git
 diff --git a/pkgA b/pkgA
 new file mode 160000
-index 0000000..{pkg_a_sha}
+index 00000000..{pkg_a_sha}
 --- /dev/null
 +++ b/pkgA
 @@ -0,0 +1 @@
 +Subproject commit {pkg_a_sha}
 diff --git a/pkgB b/pkgB
 new file mode 160000
-index 0000000..{pkg_b_sha}
+index 00000000..{pkg_b_sha}
 --- /dev/null
 +++ b/pkgB
 @@ -0,0 +1 @@

integration/tests/test_pr_workflow.py

@@ -14,7 +14,7 @@ from tests.lib.common_test_utils import (
 def test_pr_workflow_succeeded(staging_main_env, mock_build_result):
     """End-to-end test for a successful PR workflow."""
     gitea_env, test_full_repo_name, merge_branch_name = staging_main_env
-    diff = "diff --git a/test.txt b/test.txt\nnew file mode 100644\nindex 0000000..e69de29\n"
+    diff = "diff --git a/test.txt b/test.txt\nnew file mode 100644\nindex 00000000..473a0f4c\n"
     pr = gitea_env.create_gitea_pr("mypool/pkgA", diff, "Test PR - should succeed", False, base_branch=merge_branch_name)
     initial_pr_number = pr["number"]
 
@@ -59,7 +59,7 @@ def test_pr_workflow_succeeded(staging_main_env, mock_build_result):
 def test_pr_workflow_failed(staging_main_env, mock_build_result):
     """End-to-end test for a failed PR workflow."""
     gitea_env, test_full_repo_name, merge_branch_name = staging_main_env
-    diff = "diff --git a/another_test.txt b/another_test.txt\nnew file mode 100644\nindex 0000000..e69de29\n"
+    diff = "diff --git a/another_test.txt b/another_test.txt\nnew file mode 100644\nindex 00000000..473a0f4c\n"
     pr = gitea_env.create_gitea_pr("mypool/pkgA", diff, "Test PR - should fail", False, base_branch=merge_branch_name)
     initial_pr_number = pr["number"]
 

integration/tests/workflow_pr_label_test.py

@@ -27,7 +27,7 @@ def test_001_project_pr_labels(label_env, staging_bot_client):
     # 1. Create a package PR
     diff = """diff --git a/label_test_fixture.txt b/label_test_fixture.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgA on branch {branch_name} ---")
     package_pr = gitea_env.create_gitea_pr("mypool/pkgA", diff, "Test Labels Fixture", False, base_branch=branch_name)

integration/tests/workflow_pr_merge_test.py

@@ -17,7 +17,7 @@ def test_001_automerge(automerge_env, test_user_client):
     # 1. Create a package PR
     diff = """diff --git a/automerge_test.txt b/automerge_test.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgA on branch {merge_branch_name} ---")
     package_pr = test_user_client.create_gitea_pr("mypool/pkgA", diff, "Test Automerge Fixture", False, base_branch=merge_branch_name)
@@ -49,7 +49,7 @@ def test_002_manual_merge(manual_merge_env, test_user_client, usera_client, stag
     # 1. Create a package PR
     diff = """diff --git a/manual_merge_test.txt b/manual_merge_test.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgA on branch {merge_branch_name} ---")
     package_pr = gitea_env.create_gitea_pr("mypool/pkgA", diff, "Test Manual Merge Fixture", False, base_branch=merge_branch_name)
@@ -149,7 +149,7 @@ def test_003_refuse_manual_merge(manual_merge_env, test_user_client, ownerB_clie
     # 1. Create a package PR
     diff = """diff --git a/manual_merge_test.txt b/manual_merge_test.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgA on branch {merge_branch_name} ---")
     package_pr = test_user_client.create_gitea_pr("mypool/pkgA", diff, "Test Manual Merge Fixture", False, base_branch=merge_branch_name)
@@ -245,7 +245,7 @@ def test_008_merge_mode_ff_only_success(merge_ff_env, test_user_client):
     # 1. Create a package PR (this will be FF-mergeable by default)
     diff = """diff --git a/ff_test.txt b/ff_test.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     package_pr = test_user_client.create_gitea_pr("mypool/pkgA", diff, "Test FF Merge", False, base_branch=merge_branch_name)
     package_pr_number = package_pr["number"]
@@ -269,7 +269,7 @@ def test_009_merge_mode_ff_only_failure(merge_ff_env, ownerA_client):
     # 1. Create a package PR that adds a file
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/{filename}
 @@ -0,0 +1 @@
@@ -291,7 +291,7 @@ index 0000000..e69de29
 
     print("Pushing another change to PR branch to trigger sync...")
     gitea_env.modify_gitea_pr("mypool/pkgA", package_pr_number, 
-                             "diff --git a/sync_test.txt b/sync_test.txt\nnew file mode 100644\nindex 0000000..e69de29\n",
+                             "diff --git a/sync_test.txt b/sync_test.txt\nnew file mode 100644\nindex 00000000..473a0f4c\n",
                              "Trigger Sync")
 
     # The bot should detect it's not FF and NOT merge, and re-request reviews because of the new commit
@@ -323,7 +323,7 @@ def test_010_merge_mode_devel_success(merge_devel_env, ownerA_client):
     # 1. Create a package PR that adds a file
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/{filename}
 @@ -0,0 +1 @@
@@ -365,7 +365,7 @@ def test_011_merge_mode_replace_success(merge_replace_env, ownerA_client):
     # 1. Create a package PR that adds a file
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/{filename}
 @@ -0,0 +1 @@
@@ -419,7 +419,7 @@ def test_012_merge_mode_devel_ff_success(merge_devel_env, ownerA_client):
     # 1. Create a package PR (this will be FF-mergeable by default)
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/{filename}
 @@ -0,0 +1 @@
@@ -461,7 +461,7 @@ def test_013_merge_mode_replace_ff_success(merge_replace_env, ownerA_client):
     # 1. Create a package PR (this will be FF-mergeable by default)
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/{filename}
 @@ -0,0 +1 @@

integration/tests/workflow_pr_review_test.py

@@ -20,7 +20,7 @@ def test_001_review_requests_matching_config(automerge_env, ownerA_client):
     filename = f"pkgB_test_{ts}.txt"
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgB on branch {branch_name} as ownerA ---")
     package_pr = ownerA_client.create_gitea_pr("mypool/pkgB", diff, "Test Review Requests Config", True, base_branch=branch_name)
@@ -86,7 +86,7 @@ def test_004_maintainer(maintainer_env, ownerA_client):
     filename = f"maintainer_test_{ts}.txt"
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgA on branch {branch_name} as ownerA ---")
     package_pr = ownerA_client.create_gitea_pr("mypool/pkgA", diff, "Test Maintainer Merge", True, base_branch=branch_name)
@@ -155,7 +155,7 @@ def test_005_any_maintainer_approval_sufficient(maintainer_env, ownerA_client, o
     filename = f"pkgB_test_{ts}.txt"
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgB on branch {branch_name} as ownerA ---")
     package_pr = ownerA_client.create_gitea_pr("mypool/pkgB", diff, "Test Single Maintainer Merge", True, base_branch=branch_name)
@@ -217,7 +217,7 @@ def test_006_maintainer_rejection_removes_other_requests(maintainer_env, ownerA_
     filename = f"pkgB_rejection_test_{ts}.txt"
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgB on branch {branch_name} as ownerA ---")
     package_pr = ownerA_client.create_gitea_pr("mypool/pkgB", diff, "Test Maintainer Rejection", True, base_branch=branch_name)
@@ -277,7 +277,7 @@ def test_007_review_required_needs_all_approvals(review_required_env, ownerA_cli
     filename = f"pkgB_review_required_test_{ts}.txt"
     diff = f"""diff --git a/{filename} b/{filename}
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 """
     print(f"--- Creating package PR in mypool/pkgB on branch {branch_name} as ownerA ---")
     package_pr = ownerA_client.create_gitea_pr("mypool/pkgB", diff, "Test Review Required", True, base_branch=branch_name)

integration/tests/workflow_pr_sync_test.py

@@ -22,7 +22,7 @@ pytest.forwarded_pr_number = None
 @pytest.mark.dependency()
 def test_001_project_pr(gitea_env):
     """Forwarded PR correct title"""
-    diff = "diff --git a/another_test.txt b/another_test.txt\nnew file mode 100644\nindex 0000000..e69de29\n"
+    diff = "diff --git a/another_test.txt b/another_test.txt\nnew file mode 100644\nindex 00000000..473a0f4c\n"
     pytest.pr = gitea_env.create_gitea_pr("mypool/pkgA", diff, "Test PR", False)
     pytest.initial_pr_number = pytest.pr["number"]
     time.sleep(5) # Give Gitea some time to process the PR and make the timeline available
@@ -114,7 +114,7 @@ def test_005_NoProjectGitPR_edits_disabled(no_project_git_pr_env, test_user_clie
     # 1. Create a Package PR (without "Allow edits from maintainers" enabled)
     initial_diff = """diff --git a/first_file.txt b/first_file.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/first_file.txt
 @@ -0,0 +1 @@
@@ -160,7 +160,7 @@ index {pkgA_main_sha[:7]}..{pkgA_pr_head_sha[:7]} 160000
     # 4. Trigger an update on the Package PR to prompt the bot to react to the manual Project PR
     new_diff_content = """diff --git a/trigger_bot.txt b/trigger_bot.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/trigger_bot.txt
 @@ -0,0 +1 @@
@@ -200,7 +200,7 @@ def test_006_NoProjectGitPR_edits_enabled(no_project_git_pr_env, test_user_clien
     # 2. Create a Package PR with "Allow edits from maintainers" enabled
     diff = """diff --git a/new_feature.txt b/new_feature.txt
 new file mode 100644
-index 0000000..e69de29
+index 00000000..473a0f4c
 --- /dev/null
 +++ b/new_feature.txt
 @@ -0,0 +1 @@

integration/workflow-pr/entrypoint.sh

@@ -19,22 +19,18 @@ export GITEA_TOKEN
 echo "GITEA_TOKEN exported (length: ${#GITEA_TOKEN})"
 
 # Wait for the dummy data to be created by the gitea setup script
-echo "Waiting for workflow.config in myproducts/mySLFO..."
-API_URL="http://gitea-test:3000/api/v1/repos/myproducts/mySLFO/contents/workflow.config"
+echo "Waiting for workflow.config in myproducts/mySLFO (branch zz-ready-to-start)..."
+API_URL="http://gitea-test:3000/api/v1/repos/myproducts/mySLFO/contents/workflow.config?ref=zz-ready-to-start"
 HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" -H "Authorization: token $GITEA_TOKEN" "$API_URL")
 
 WAITED=false
 while [ "$HTTP_STATUS" != "200" ]; do
-  WAITED=true
-  echo "workflow.config not found yet (HTTP Status: $HTTP_STATUS). Retrying in 5s..."
-  sleep 5
+  echo "workflow.config on zz-ready-to-start not found yet (HTTP Status: $HTTP_STATUS). Retrying in 1s..."
+  sleep 1
   HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" -H "Authorization: token $GITEA_TOKEN" "$API_URL")
 done
 
-if [ "$WAITED" = true ]; then
-  echo "workflow.config found. Sleeping 15s to let other configurations settle..."
-  sleep 15
-fi
+echo "workflow.config found on zz-ready-to-start."
 
 # Wait for the shared SSH key to be generated by the gitea setup script
 echo "Waiting for /var/lib/gitea/ssh-keys/id_ed25519..."

workflow-direct/main.go

@@ -503,10 +503,7 @@ func updateConfiguration(configFilename string, orgs *[]string) {
 		os.Exit(4)
 	}
 
-	configs, err := common.ResolveWorkflowConfigs(gitea, configFile)
-	if err != nil {
-		common.LogError("Failed to resolve some configuration repositories:", err)
-	}
+	configs, _ := common.ResolveWorkflowConfigs(gitea, configFile)
 	configuredRepos = make(map[string][]*common.AutogitConfig)
 	*orgs = make([]string, 0, 1)
 	for _, c := range configs {

workflow-pr/main.go

@@ -58,7 +58,6 @@ func main() {
 	checkOnStart := flag.Bool("check-on-start", common.GetEnvOverrideBool(os.Getenv("AUTOGITS_CHECK_ON_START"), false), "Check all repositories for consistency on start, without delays")
 	checkIntervalHours := flag.Float64("check-interval", 5, "Check interval (+-random delay) for repositories for consitency, in hours")
 	flag.BoolVar(&ListPROnly, "list-prs-only", false, "Only lists PRs without acting on them")
-	exitOnConfigError := flag.Bool("exit-on-config-error", false, "Exit if any repository in configuration cannot be resolved")
 	flag.Int64Var(&PRID, "id", -1, "Process only the specific ID and ignore the rest. Use for debugging")
 	basePath := flag.String("repo-path", common.GetEnvOverrideString(os.Getenv("AUTOGITS_REPO_PATH"), ""), "Repository path. Default is temporary directory")
 	pr := flag.String("only-pr", "", "Only specific PR to process. For debugging")
@@ -98,10 +97,8 @@ func main() {
 
 	configs, err := common.ResolveWorkflowConfigs(Gitea, config)
 	if err != nil {
-		common.LogError("Failed to resolve some configuration repositories:", err)
-		if *exitOnConfigError {
-			return
-		}
+		common.LogError("Cannot resolve config files:", err)
+		return
 	}
 
 	for _, c := range configs {

workflow-pr/pr_processor.go

@@ -8,6 +8,7 @@ import (
 	"runtime/debug"
 	"slices"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/opentracing/opentracing-go/log"
@@ -628,9 +629,29 @@ func (pr *PRProcessor) Process(req *models.PullRequest) error {
 	return err
 }
 
+// prLocks serialises concurrent processing of the same PR.
+// Both the RabbitMQ event loop and the consistency-checker goroutine call
+// ProcesPullRequest; without this lock they can race on reviewer add/remove.
+// Key format: "org/repo#num"
+var prLocks sync.Map // map[string]chan struct{}
+
+func prLockKey(pr *models.PullRequest) string {
+	return fmt.Sprintf("%s/%s#%d", pr.Base.Repo.Owner.UserName, pr.Base.Repo.Name, pr.Index)
+}
+
+func acquirePRLock(key string) chan struct{} {
+	v, _ := prLocks.LoadOrStore(key, make(chan struct{}, 1))
+	ch := v.(chan struct{})
+	ch <- struct{}{}
+	return ch
+}
+
+func releasePRLock(ch chan struct{}) {
+	<-ch
+}
+
 type RequestProcessor struct {
 	configuredRepos map[string][]*common.AutogitConfig
-	recursive       int
 }
 
 func (w *RequestProcessor) Process(pr *models.PullRequest) error {
@@ -647,6 +668,9 @@ func ProcesPullRequest(pr *models.PullRequest, configs []*common.AutogitConfig)
 		return nil
 	}
 
+	lock := acquirePRLock(prLockKey(pr))
+	defer releasePRLock(lock)
+
 	PRProcessor, err := AllocatePRProcessor(pr, configs)
 	if err != nil {
 		log.Error(err)
@@ -663,17 +687,23 @@ func (w *RequestProcessor) ProcessFunc(request *common.Request) (err error) {
 			common.LogInfo("panic cought --- recovered")
 			common.LogError(string(debug.Stack()))
 		}
-		w.recursive--
 	}()
 
-	w.recursive++
-	if w.recursive > 3 {
-		common.LogError("Recursion limit reached... something is wrong with this PR?")
-		return nil
-	}
-
 	var pr *models.PullRequest
 	if req, ok := request.Data.(*common.PullRequestWebhookEvent); ok {
+		// Skip pull_request_sync events triggered by the bot's own pushes to
+		// prjgit branches.  Those would re-run AssignReviewers immediately
+		// after the bot itself just set them, producing spurious add/remove
+		// cycles.  Human-triggered syncs have a different sender and are still
+		// processed normally.
+		if request.Type == common.RequestType_PRSync && CurrentUser != nil &&
+			req.Sender.Username == CurrentUser.UserName {
+			common.LogDebug("Skipping self-triggered pull_request_sync from", req.Sender.Username,
+				"on", req.Pull_Request.Base.Repo.Owner.Username+"/"+req.Pull_Request.Base.Repo.Name,
+				"#", req.Pull_Request.Number)
+			return nil
+		}
+
 		pr, err = Gitea.GetPullRequest(req.Pull_Request.Base.Repo.Owner.Username, req.Pull_Request.Base.Repo.Name, req.Pull_Request.Number)
 		if err != nil {
 			common.LogError("Cannot find PR for issue:", req.Pull_Request.Base.Repo.Owner.Username, req.Pull_Request.Base.Repo.Name, req.Pull_Request.Number)
@@ -710,8 +740,16 @@ func (w *RequestProcessor) ProcessFunc(request *common.Request) (err error) {
 		common.LogError("*** Cannot find config for org:", pr.Base.Repo.Owner.UserName)
 	}
 	if err = ProcesPullRequest(pr, configs); err == updatePrjGitError_requeue {
-		time.Sleep(time.Second * 5)
-		return w.ProcessFunc(request)
+		// Retry after a delay in a background goroutine so the event loop is
+		// not blocked while we wait.  The per-PR lock inside ProcesPullRequest
+		// ensures no other processing races with the retry.
+		go func() {
+			time.Sleep(time.Second * 5)
+			if err := ProcesPullRequest(pr, configs); err != nil {
+				common.LogError("requeue retry failed:", err)
+			}
+		}()
+		return nil
 	}
 	return err
 }

workflow-pr/pr_processor_test.go

@@ -989,18 +989,6 @@ func TestProcessFunc(t *testing.T) {
 		}
 	})
 
-	t.Run("Recursion limit", func(t *testing.T) {
-		reqProc.recursive = 3
-		err := reqProc.ProcessFunc(&common.Request{})
-		if err != nil {
-			t.Errorf("Expected nil error on recursion limit, got %v", err)
-		}
-		if reqProc.recursive != 3 {
-			t.Errorf("Expected recursive to be 3, got %d", reqProc.recursive)
-		}
-		reqProc.recursive = 0 // Reset
-	})
-
 	t.Run("Invalid data format", func(t *testing.T) {
 		err := reqProc.ProcessFunc(&common.Request{Data: nil})
 		if err == nil || !strings.Contains(err.Error(), "Invalid data format") {