metaserver: Allow re-replication of chunks only after they are really…

… removed from evacuated server.
avodaniel · Jul 31, 2020 · 3528d7e · 3528d7e
1 parent f06e7ee
commit 3528d7e
Show file tree

Hide file tree

Showing 2 changed files with 27 additions and 0 deletions.
diff --git a/src/cc/meta/ChunkServer.h b/src/cc/meta/ChunkServer.h
@@ -987,6 +987,13 @@ class ChunkServer :
     static int GetMaxChunkServerCount()
         { return sMaxChunkServerCount; }
 
+    bool IsRemovePending(chunkId_t const id) const
+        { return mChunksRemovePending.Find(id); }
+    bool EraseRemovePending(chunkId_t const id)
+        { return mChunksRemovePending.Erase(id); }
+    void SetRemovePending(chunkId_t const id)
+        { mChunksRemovePending.Insert(id); }
+
 protected:
     ChunkServer(const NetConnectionPtr& conn, const string& peerName,
         bool replayFlag = false);
@@ -1071,6 +1078,8 @@ class ChunkServer :
 
     ChunkIdSet mChunksToEvacuate;
 
+    ChunkIdSet mChunksRemovePending;
+
     /// Location of the server at which clients can
     /// connect to
     ServerLocation mLocation;

diff --git a/src/cc/meta/LayoutManager.cc b/src/cc/meta/LayoutManager.cc
@@ -11046,10 +11046,12 @@ LayoutManager::ReplicateChunk(
             continue;
         }
         const char*    reason = "none";
+        bool isEvacuated = false;
         ChunkServerPtr dataServer;
         if (iter != servers.end()) {
             ChunkServer& ds = **iter;
             reason = "evacuation";
+            isEvacuated = true;
             if (recoveryInfo.HasRecovery()) {
                 reason     = "evacuation recovery";
                 dataServer = c;
@@ -11122,6 +11124,9 @@ LayoutManager::ReplicateChunk(
             SetReplicationState(clli,
                 CSMap::Entry::kStatePendingReplication);
         }
+        if (isEvacuated) {
+            dataServer->SetRemovePending(clli.GetChunkId());
+        }
         // Do not count synchronous failures.
         if (cs.ReplicateChunk(clli.GetFileId(), clli.GetChunkId(),
                 dataServer, recoveryInfo, tier, maxSTier, recovIt,
@@ -11513,6 +11518,19 @@ LayoutManager::CanReplicateChunkNow(
         " hibernated: " << hibernatedCount <<
         " needed: "     << extraReplicas <<
     KFS_LOG_EOM;
+
+    for (auto const& srv: servers) {
+        if (srv->IsRemovePending(chunkId)) {
+            if (mChunkToServerMap.HasServer(srv, c)) {
+                KFS_LOG_STREAM_DEBUG << "Skipping CanReplicate; srv:" << srv->GetServerLocation() << " and chunk:" << chunkId << KFS_LOG_EOM;
+                return false;
+            } else {
+                srv->EraseRemovePending(chunkId);
+                KFS_LOG_STREAM_DEBUG << "Re-allowing CanReplicate; srv:" << srv->GetServerLocation() << " and chunk:" << chunkId << KFS_LOG_EOM;
+            }
+        }
+    }
+
     if (readLeaseWaitFlag) {
         SetReplicationState(c, CSMap::Entry::kStatePendingReplication);
         return false;