limit bootstrapping to min of 4. fix RingIterator logic.

salesforce · Jun 11, 2024 · 33cc83f · 33cc83f
1 parent a2c3a9f
commit 33cc83f
Show file tree

Hide file tree

Showing 4 changed files with 17 additions and 15 deletions.
diff --git a/ethereal/src/main/java/com/salesforce/apollo/ethereal/memberships/ChRbcGossip.java b/ethereal/src/main/java/com/salesforce/apollo/ethereal/memberships/ChRbcGossip.java
@@ -28,9 +28,7 @@
 import org.slf4j.LoggerFactory;
 
 import java.time.Duration;
-import java.util.ArrayList;
 import java.util.Collection;
-import java.util.List;
 import java.util.Optional;
 import java.util.concurrent.Executors;
 import java.util.concurrent.ScheduledExecutorService;
@@ -42,8 +40,8 @@
 import static com.salesforce.apollo.ethereal.memberships.comm.GossiperClient.getCreate;
 
 /**
- * Handles the gossip propigation of proposals, commits and preVotes from this node, as well as the notification of the
- * adder of such from other nodes.
+ * Handles the gossip propagation of proposals, the commits and preVotes from this node, and the notification of the
+ * Adder of such events from other nodes.
  *
  * @author hal.hildebrand
  */
@@ -59,7 +57,6 @@ public class ChRbcGossip {
     private final        SliceIterator<Gossiper>                         ring;
     private final        AtomicBoolean                                   started  = new AtomicBoolean();
     private final        Terminal                                        terminal = new Terminal();
-    private final        List<Member>                                    membership;
     private volatile     ScheduledFuture<?>                              scheduled;
 
     public ChRbcGossip(Digest id, SigningMember member, Collection<Member> membership, Processor processor,
@@ -68,12 +65,10 @@ public ChRbcGossip(Digest id, SigningMember member, Collection<Member> membershi
         this.member = member;
         this.metrics = m;
         this.id = id;
-        this.membership = new ArrayList<>(membership);
         comm = communications.create(member, id, terminal, getClass().getCanonicalName(),
                                      r -> new GossiperServer(communications.getClientIdentityProvider(), metrics, r),
                                      getCreate(metrics), Gossiper.getLocalLoopback(member));
-        ring = new SliceIterator<Gossiper>("ChRbcGossip[%s on: %s]".formatted(id, member.getId()), member, membership,
-                                           comm);
+        ring = new SliceIterator<>("ChRbcGossip[%s on: %s]".formatted(id, member.getId()), member, membership, comm);
     }
 
     /**
@@ -150,7 +145,7 @@ private Update gossipRound(Gossiper link) {
         if (!started.get()) {
             return null;
         }
-        log.trace("gossiping[{}] with {} on {}", id, link.getMember(), member);
+        log.trace("gossiping[{}] with {} on {}", id, link.getMember(), member.getId());
         try {
             return link.gossip(processor.gossip(id));
         } catch (StatusRuntimeException e) {

diff --git a/fireflies/src/main/java/com/salesforce/apollo/fireflies/View.java b/fireflies/src/main/java/com/salesforce/apollo/fireflies/View.java
@@ -85,7 +85,7 @@ public class View {
     private static final String FINALIZE_VIEW_CHANGE  = "FINALIZE VIEW CHANGE";
     private static final Logger log                   = LoggerFactory.getLogger(View.class);
     private static final String SCHEDULED_VIEW_CHANGE = "Scheduled View Change";
-    
+
     final            CommonCommunications<Fireflies, Service>    comm;
     final            AtomicBoolean                               started               = new AtomicBoolean();
     private final    CommonCommunications<Entrance, Service>     approaches;
@@ -343,8 +343,8 @@ void finalizeViewChange() {
             return;
         }
         viewChange(() -> {
-            final var superMajority =
-            context.size() == 1 ? 1 : context.getRingCount() - ((context.getRingCount() - 1) / 4);
+            var rings = context.getRingCount();
+            final var superMajority = context.size() == 1 ? 1 : rings - ((rings - 1) / 4);
             if (observations.size() < superMajority) {
                 log.trace("Do not have superMajority: {} required: {} observers: {} for: {} on: {}",
                           observations.size(), superMajority, viewManagement.observersList(), currentView(),

diff --git a/fireflies/src/main/java/com/salesforce/apollo/fireflies/ViewManagement.java b/fireflies/src/main/java/com/salesforce/apollo/fireflies/ViewManagement.java
@@ -374,6 +374,12 @@ boolean joined() {
      * start a view change if there are any offline members or joining members
      */
     void maybeViewChange() {
+        if (context.size() == 1 && joins.size() < 3) {
+            log.info("Do not have minimum cluster size: {} required: {} for: {} on: {}", joins.size() + context.size(),
+                     4, currentView(), node.getId());
+            view.scheduleViewChange();
+            return;
+        }
         if ((context.offlineCount() > 0 || !joins.isEmpty())) {
             if (isObserver()) {
                 log.trace("Initiating view change: {} (non observer) joins: {} leaves: {} on: {}", currentView(),

diff --git a/memberships/src/main/java/com/salesforce/apollo/ring/RingIterator.java b/memberships/src/main/java/com/salesforce/apollo/ring/RingIterator.java
@@ -171,16 +171,17 @@ private void proceed(Digest key, final boolean allow, Runnable onMajority, Runna
         if (!finalIteration) {
             log.trace(
             "Determining: {} continuation of: {} for digest: {} tally: {} majority: {} final itr: {} allow: {} on: {}",
-            current, key, context.getId(), tally.get(), context.majority(), finalIteration, allow, member.getId());
+            current, key, context.getId(), tally.get(), context.toleranceLevel() + 1, finalIteration, allow,
+            member.getId());
         }
         if (finalIteration && allow) {
             log.trace("Completing iteration: {} of: {} for digest: {} tally: {} on: {}", iteration(), key,
                       context.getId(), tally.get(), member.getId());
             if (failedMajority != null && !majorityFailed) {
-                if (tally.get() < context.majority()) {
+                if (tally.get() < context.toleranceLevel() + 1) {
                     majorityFailed = true;
                     log.debug("Failed to obtain majority of: {} for digest: {} tally: {} required: {} on: {}", key,
-                              context.getId(), tally.get(), context.majority(), member.getId());
+                              context.getId(), tally.get(), context.toleranceLevel() + 1, member.getId());
                     failedMajority.run();
                 }
             }