wait in lock

.github/mergify.yml

@@ -16,3 +16,12 @@ pull_request_rules:
       backport:
         branches:
           - v0.34.x
+  - name: backport patches to v0.35.x branch
+    conditions:
+      - base=master
+      - label=S:backport-to-v0.35.x
+    actions:
+      backport:
+        branches:
+          - v0.35.x
+

.github/workflows/e2e-nightly-35x.yml

@@ -0,0 +1,74 @@
+# Runs randomly generated E2E testnets nightly on v0.35.x.
+
+# !! If you change something in this file, you probably want
+# to update the e2e-nightly-master workflow as well!
+
+name: e2e-nightly-35x
+on:
+  workflow_dispatch: # allow running workflow manually
+  schedule:
+    - cron: '0 2 * * *'
+
+jobs:
+  e2e-nightly-test-2:
+    # Run parallel jobs for the listed testnet groups (must match the
+    # ./build/generator -g flag)
+    strategy:
+      fail-fast: false
+      matrix:
+        p2p: ['legacy', 'new', 'hybrid']
+        group: ['00', '01', '02', '03']
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    steps:
+      - uses: actions/setup-go@v2
+        with:
+          go-version: '1.17'
+
+      - uses: actions/checkout@v2.3.4
+
+      - name: Build
+        working-directory: test/e2e
+        # Run make jobs in parallel, since we can't run steps in parallel.
+        run: make -j2 docker generator runner tests
+
+      - name: Generate testnets
+        working-directory: test/e2e
+        # When changing -g, also change the matrix groups above
+        run: ./build/generator -g 4 -d networks/nightly/${{ matrix.p2p }} -p ${{ matrix.p2p }}
+
+      - name: Run ${{ matrix.p2p }} p2p testnets in group ${{ matrix.group }}
+        working-directory: test/e2e
+        run: ./run-multiple.sh networks/nightly/${{ matrix.p2p }}/*-group${{ matrix.group }}-*.toml
+
+  e2e-nightly-fail-2:
+    needs: e2e-nightly-test-2
+    if: ${{ failure() }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Notify Slack on failure
+        uses: rtCamp/action-slack-notify@12e36fc18b0689399306c2e0b3e0f2978b7f1ee7
+        env:
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+          SLACK_CHANNEL: tendermint-internal
+          SLACK_USERNAME: Nightly E2E Tests
+          SLACK_ICON_EMOJI: ':skull:'
+          SLACK_COLOR: danger
+          SLACK_MESSAGE: Nightly E2E tests failed on v0.35.x
+          SLACK_FOOTER: ''
+
+  e2e-nightly-success: # may turn this off once they seem to pass consistently
+    needs: e2e-nightly-test-2
+    if: ${{ success() }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Notify Slack on success
+        uses: rtCamp/action-slack-notify@12e36fc18b0689399306c2e0b3e0f2978b7f1ee7
+        env:
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+          SLACK_CHANNEL: tendermint-internal
+          SLACK_USERNAME: Nightly E2E Tests
+          SLACK_ICON_EMOJI: ':white_check_mark:'
+          SLACK_COLOR: good
+          SLACK_MESSAGE: Nightly E2E tests passed on v0.35.x
+          SLACK_FOOTER: ''

internal/consensus/reactor.go

@@ -230,17 +230,15 @@ func (r *Reactor) OnStop() {
 	}
 
 	r.mtx.Lock()
-	peers := r.peers
+	// Close and wait for each of the peers to shutdown.
+	// This is safe to perform with the lock since none of the peers require the
+	// lock to complete any of the methods that the waitgroup is waiting on.
+	for _, state := range r.peers {
+		state.closer.Close()
+		state.broadcastWG.Wait()
+	}
 	r.mtx.Unlock()
 
-	// wait for all spawned peer goroutines to gracefully exit
-	for _, ps := range peers {
-		ps.closer.Close()
-	}
-	for _, ps := range peers {
-		ps.broadcastWG.Wait()
-	}
-
 	// Close the StateChannel goroutine separately since it uses its own channel
 	// to signal closure.
 	close(r.stateCloseCh)