diff --git a/src/ripple/nodestore/backend/NuDBFactory.cpp b/src/ripple/nodestore/backend/NuDBFactory.cpp
index 74afe2ed7f1..3bb4253dbff 100644
--- a/src/ripple/nodestore/backend/NuDBFactory.cpp
+++ b/src/ripple/nodestore/backend/NuDBFactory.cpp
@@ -266,7 +266,7 @@ class NuDBBackend : public Backend, public BatchWriter::Callback
     void
     store(std::shared_ptr<NodeObject> const& no) override
     {
-        batch_.store(no);
+        batch_.store(no, j_);
     }
 
     void
diff --git a/src/ripple/nodestore/backend/RocksDBFactory.cpp b/src/ripple/nodestore/backend/RocksDBFactory.cpp
index b34560dba89..9443e91f233 100644
--- a/src/ripple/nodestore/backend/RocksDBFactory.cpp
+++ b/src/ripple/nodestore/backend/RocksDBFactory.cpp
@@ -343,7 +343,7 @@ class RocksDBBackend : public Backend, public BatchWriter::Callback
     void
     store(std::shared_ptr<NodeObject> const& object) override
     {
-        m_batch.store(object);
+        m_batch.store(object, m_journal);
     }
 
     void
diff --git a/src/ripple/nodestore/impl/BatchWriter.cpp b/src/ripple/nodestore/impl/BatchWriter.cpp
index 692032016cc..4ce6f62fad4 100644
--- a/src/ripple/nodestore/impl/BatchWriter.cpp
+++ b/src/ripple/nodestore/impl/BatchWriter.cpp
@@ -37,21 +37,31 @@ BatchWriter::~BatchWriter()
 }
 
 void
-BatchWriter::store(std::shared_ptr<NodeObject> const& object)
+BatchWriter::store(
+    std::shared_ptr<NodeObject> const& object,
+    beast::Journal const& j)
 {
     std::unique_lock<decltype(mWriteMutex)> sl(mWriteMutex);
 
-    // If the batch has reached its limit, we wait
-    // until the batch writer is finished
-    while (mWriteSet.size() >= batchWriteLimitSize)
-        mWriteCondition.wait(sl);
-
     mWriteSet.push_back(object);
 
     if (!mWritePending)
     {
         mWritePending = true;
 
+        // Log if the write batch size is too big. The trade-off here
+        // is that new ledgers can't be persisted if a limit is enforced,
+        // which causes desync. Versus eventual memory exhaustion if
+        // there are no limits. But we at least stay in sync longer in the
+        // latter case.
+        if (mWriteSet.size() >= batchWriteLimitSize)
+        {
+            JLOG(j.warn())
+                << "Pending write batch size " << mWriteSet.size()
+                << " exceeds threshold of " << batchWriteLimitSize
+                << ". This may be caused by slow i/o. More memory is consumed "
+                   " as pending write count increases.";
+        }
         m_scheduler.scheduleTask(*this);
     }
 }
diff --git a/src/ripple/nodestore/impl/BatchWriter.h b/src/ripple/nodestore/impl/BatchWriter.h
index 9ce4f120329..2c664e1411b 100644
--- a/src/ripple/nodestore/impl/BatchWriter.h
+++ b/src/ripple/nodestore/impl/BatchWriter.h
@@ -20,6 +20,7 @@
 #ifndef RIPPLE_NODESTORE_BATCHWRITER_H_INCLUDED
 #define RIPPLE_NODESTORE_BATCHWRITER_H_INCLUDED
 
+#include <ripple/basics/Log.h>
 #include <ripple/nodestore/Scheduler.h>
 #include <ripple/nodestore/Task.h>
 #include <ripple/nodestore/Types.h>
@@ -68,7 +69,7 @@ class BatchWriter : private Task
         write the batch out.
     */
     void
-    store(std::shared_ptr<NodeObject> const& object);
+    store(std::shared_ptr<NodeObject> const& object, beast::Journal const& j);
 
     /** Get an estimate of the amount of writing I/O pending. */
     int