Merge branch 'develop' into mvadari/sponsor/ai-tests

include/xrpl/basics/TaggedCache.ipp

@@ -3,6 +3,9 @@
 #include <xrpl/basics/IntrusivePointer.ipp>
 #include <xrpl/basics/Log.h>  // IWYU pragma: keep
 #include <xrpl/basics/TaggedCache.h>
+#include <xrpl/basics/scope.h>
+
+#include <algorithm>
 
 namespace xrpl {
 
@@ -601,8 +604,42 @@ TaggedCache<Key, T, IsKeyCache, SharedWeakUnionPointer, SharedPointerType, Hash,
     std::vector<key_type> v;
 
     {
-        std::scoped_lock const lock(mutex_);
-        v.reserve(cache_.size());
+        // Keep track of how many iterations are needed. Exit the loop if the number of retries gets
+        // absurd. (Note that if this somehow ever happens, one more allocation will be done under
+        // lock, which is undesirable, but really should be almost impossible.)
+        std::size_t allocationIterations = 0;
+        std::unique_lock lock(mutex_);
+        for (auto size = cache_.size(); v.capacity() < size && allocationIterations < 20;
+             size = cache_.size())
+        {
+            ScopeUnlock const unlock(lock);
+            if (allocationIterations > 0)
+            {
+                JLOG(journal_.info())
+                    << "getKeys(): Cache grew beyond allocated capacity after "
+                    << allocationIterations << " prior attempt(s). Have " << v.capacity()
+                    << ", need " << size << ". Retrying allocation";
+            }
+            // Allocate the current size plus a little extra, in case the cache grows while
+            // allocating. Each time another allocation is needed, the extra also gets bigger until
+            // it ultimately doubles the size + 1.
+            constexpr std::size_t baseShift = 5;
+            auto const bufferOffset = std::min(allocationIterations, std::size_t{baseShift});
+            auto const bufferShift = baseShift - bufferOffset;
+            size += (size >> bufferShift) + 1;
+            v.reserve(size);
+            ++allocationIterations;
+        }
+        if (v.capacity() < cache_.size())
+        {
+            // LCOV_EXCL_START
+            UNREACHABLE("xrpl::TaggedCache::getKeys(): failed to allocate sufficient capacity");
+            v.reserve(cache_.size());
+            // LCOV_EXCL_STOP
+        }
+        XRPL_ASSERT(lock.owns_lock(), "xrpl::TaggedCache::getKeys(): owns lock");
+        XRPL_ASSERT(
+            v.capacity() >= cache_.size(), "xrpl::TaggedCache::getKeys(): sufficient capacity");
         for (auto const& _ : cache_)
             v.push_back(_.first);
     }

include/xrpl/nodestore/DatabaseRotating.h

@@ -41,6 +41,19 @@ public:
         std::unique_ptr<NodeStore::Backend>&& newBackend,
         std::function<void(std::string const& writableName, std::string const& archiveName)> const&
             f) = 0;
+
+    /**
+     * Marks an online-delete rotation as in progress (or completed).
+     *
+     * While in flight, a read served by the archive backend is copied
+     * forward into the writable backend even for ordinary
+     * (duplicate == false) fetches: the archive is about to be deleted,
+     * and a node body canonicalized into caches during the rotation
+     * window would otherwise survive only in RAM once the archive is
+     * dropped.
+     */
+    virtual void
+    setRotationInFlight(bool inFlight) = 0;
 };
 
 }  // namespace xrpl::NodeStore

include/xrpl/nodestore/detail/DatabaseRotatingImp.h

@@ -9,6 +9,7 @@
 #include <xrpl/nodestore/NodeObject.h>
 #include <xrpl/nodestore/Scheduler.h>
 
+#include <atomic>
 #include <cstdint>
 #include <functional>
 #include <memory>
@@ -69,11 +70,22 @@ public:
     void
     sweep() override;
 
+    void
+    setRotationInFlight(bool inFlight) override;
+
 private:
     std::shared_ptr<Backend> writableBackend_;
     std::shared_ptr<Backend> archiveBackend_;
     mutable std::mutex mutex_;
 
+    // True between SHAMapStore starting the cache-freshen phase and the
+    // completion of rotate(). While true, archive hits on ordinary
+    // (duplicate == false) fetches are copied forward into the writable
+    // backend; copyForwardCount_ tallies them per rotation for the
+    // summary line logged at swap.
+    std::atomic<bool> rotationInFlight_{false};
+    std::atomic<std::uint64_t> copyForwardCount_{0};
+
     std::shared_ptr<NodeObject>
     fetchNodeObject(uint256 const& hash, std::uint32_t, FetchReport& fetchReport, bool duplicate)
         override;

src/libxrpl/nodestore/DatabaseRotatingImp.cpp

@@ -13,6 +13,7 @@
 #include <xrpl/nodestore/Scheduler.h>
 #include <xrpl/nodestore/Types.h>
 
+#include <atomic>
 #include <cstdint>
 #include <exception>
 #include <functional>
@@ -52,6 +53,7 @@ DatabaseRotatingImp::rotate(
     // callback finishes. Only then will the archive directory be
     // deleted.
     std::shared_ptr<NodeStore::Backend> oldArchiveBackend;
+    std::uint64_t copyForwards = 0;
     {
         std::scoped_lock const lock(mutex_);
 
@@ -62,11 +64,28 @@ DatabaseRotatingImp::rotate(
         newArchiveBackendName = archiveBackend_->getName();
 
         writableBackend_ = std::move(newBackend);
+
+        copyForwards = copyForwardCount_.exchange(0, std::memory_order_relaxed);
+    }
+
+    if (copyForwards > 0)
+    {
+        JLOG(j_.warn()) << "Rotating: copied forward " << copyForwards
+                        << " archive-served reads into the writable backend "
+                           "during the rotation window";
     }
 
     f(newWritableBackendName, newArchiveBackendName);
 }
 
+void
+DatabaseRotatingImp::setRotationInFlight(bool inFlight)
+{
+    rotationInFlight_.store(inFlight, std::memory_order_release);
+    JLOG(j_.debug()) << "Rotating: copy-forward on archive reads "
+                     << (inFlight ? "enabled" : "disabled");
+}
+
 std::string
 DatabaseRotatingImp::getName() const
 {
@@ -177,9 +196,18 @@ DatabaseRotatingImp::fetchNodeObject(
                 writable = writableBackend_;
             }
 
-            // Update writable backend with data from the archive backend
-            if (duplicate)
+            // Update writable backend with data from the archive backend.
+            // While a rotation is in flight, ordinary (duplicate == false)
+            // reads served by the archive are copied forward too: the
+            // archive is about to be deleted, and a body canonicalized
+            // into the cache after the freshen getKeys() snapshot would
+            // otherwise survive only in RAM once the archive is dropped.
+            if (duplicate || rotationInFlight_.load(std::memory_order_acquire))
+            {
+                if (!duplicate)
+                    copyForwardCount_.fetch_add(1, std::memory_order_relaxed);
                 writable->store(nodeObject);
+            }
         }
     }
 

src/libxrpl/tx/invariants/InvariantCheck.cpp

@@ -489,7 +489,7 @@ AccountRootsDeletedClean::finalize(
     // feature is enabled. Enabled, or not, though, a fatal-level message will
     // be logged
     [[maybe_unused]] bool const enforce = view.rules().enabled(fixCleanup3_2_0) ||
-        view.rules().enabled(featureSingleAssetVault) ||
+        view.rules().enabled(featureSponsor) || view.rules().enabled(featureSingleAssetVault) ||
         view.rules().enabled(featureLendingProtocol);
 
     auto const objectExists = [&view, enforce, &j](auto const& keylet) {

src/test/app/Invariants_test.cpp

@@ -414,6 +414,23 @@ class Invariants_test : public beast::unit_test::Suite
             XRPAmount{},
             STTx{ttACCOUNT_DELETE, [](STObject& tx) {}});
 
+        doInvariantCheck(
+            Env{*this, FeatureBitset{featureSponsor}},
+            {{"account deletion left behind a sponsorship field"}},
+            [&](Account const& a1, Account const& a2, ApplyContext& ac) {
+                auto const sleA1 = ac.view().peek(keylet::account(a1.id()));
+                if (!sleA1)
+                    return false;
+                sleA1->at(sfBalance) = beast::kZero;
+                sleA1->setAccountID(sfSponsor, a2.id());
+
+                ac.view().erase(sleA1);
+
+                return true;
+            },
+            XRPAmount{},
+            STTx{ttACCOUNT_DELETE, [](STObject& tx) {}});
+
         for (auto const& keyletInfo : kDirectAccountKeylets)
         {
             // TODO: Use structured binding once LLVM 16 is the minimum

src/xrpld/app/misc/SHAMapStoreImp.cpp

@@ -16,9 +16,11 @@
 #include <xrpl/ledger/Ledger.h>
 #include <xrpl/nodestore/Database.h>
 #include <xrpl/nodestore/Manager.h>
+#include <xrpl/nodestore/NodeObject.h>
 #include <xrpl/nodestore/Scheduler.h>
 #include <xrpl/nodestore/detail/DatabaseRotatingImp.h>
 #include <xrpl/protocol/Protocol.h>
+#include <xrpl/protocol/Serializer.h>
 #include <xrpl/server/NetworkOPs.h>
 #include <xrpl/server/State.h>
 #include <xrpl/shamap/SHAMapMissingNode.h>
@@ -254,8 +256,24 @@ bool
 SHAMapStoreImp::copyNode(std::uint64_t& nodeCount, SHAMapTreeNode const& node)
 {
     // Copy a single record from node to dbRotating_
-    dbRotating_->fetchNodeObject(
+    auto obj = dbRotating_->fetchNodeObject(
         node.getHash().asUInt256(), 0, NodeStore::FetchType::Synchronous, true);
+    if (!obj)
+    {
+        XRPL_ASSERT(node.cowid() == 0, "SHAMapStoreImp::copyNode : rescued node must be clean");
+        // Reachable from the validated state map in memory, but present in
+        // neither backend: its only on-disk copy lived in a backend removed by
+        // an earlier rotation, and it was never rewritten because it is clean
+        // (cowid == 0, so flushDirty skips it). Persist the in-memory body
+        // directly into the writable backend so it survives this rotation
+        // instead of later surfacing as an unresolvable SHAMapMissingNode.
+        auto const hash = node.getHash().asUInt256();
+        Serializer s;
+        node.serializeWithPrefix(s);
+        dbRotating_->store(NodeObjectType::AccountNode, std::move(s.modData()), hash, 0);
+        JLOG(journal_.warn()) << "copyNode: re-stored node missing from both backends, hash="
+                              << hash << " type=" << static_cast<int>(node.getType());
+    }
     if ((++nodeCount % checkHealthInterval_) == 0u)
     {
         if (healthWait() == HealthResult::Stopping)
@@ -348,6 +366,23 @@ SHAMapStoreImp::run()
             JLOG(journal_.debug())
                 << "copied ledger " << validatedSeq << " nodecount " << nodeCount;
 
+            // Close the getKeys()->swap exposure window: from here until
+            // rotate() completes, an ordinary read served by the archive is
+            // copied forward into the writable backend, so a node fetched
+            // from the doomed archive cannot be left RAM-only when the
+            // archive is deleted. RAII so the early returns below (and any
+            // exception) also clear the flag.
+            struct RotationExposureGuard
+            {
+                NodeStore::DatabaseRotating& db;
+                ~RotationExposureGuard()
+                {
+                    db.setRotationInFlight(false);
+                }
+            };
+            RotationExposureGuard const rotationExposureGuard{*dbRotating_};
+            dbRotating_->setRotationInFlight(true);
+
             JLOG(journal_.debug()) << "freshening caches";
             freshenCaches();
             if (healthWait() == HealthResult::Stopping)