Fix formatting

- If any [validator_list_keys] are not available after all
  [validator_list_sites] have had a chance to be queried, then fall
  back to loading cache files. Currently, cache files are only used if
  no sites are defined, or the request to one of them has an error. It
  does not include cases where not enough sites are defined, or if a
  site returns an invalid VL (or something else entirely).
- Resolves #5320

cfg/xrpld-example.cfg

@@ -940,7 +940,23 @@
 #
 #       path                Location to store the database
 #
-#   Optional keys for NuDB and RocksDB:
+#   Optional keys
+#
+#       cache_size          Size of cache for database records. Default is 16384.
+#                           Setting this value to 0 will use the default value.
+#
+#       cache_age           Length of time in minutes to keep database records
+#                           cached. Default is 5 minutes. Setting this value to
+#                           0 will use the default value.
+#
+#                           Note: if neither cache_size nor cache_age is
+#                           specified, the cache for database records will not
+#                           be created. If only one of cache_size or cache_age
+#                           is specified, the cache will be created using the
+#                           default value for the unspecified parameter.
+#
+#                           Note: the cache will not be created if online_delete
+#                           is specified.
 #
 #       fast_load           Boolean. If set, load the last persisted ledger
 #                           from disk upon process start before syncing to
@@ -948,6 +964,8 @@
 #                           if sufficient IOPS capacity is available.
 #                           Default 0.
 #
+#   Optional keys for NuDB or RocksDB:
+#
 #       earliest_seq        The default is 32570 to match the XRP ledger
 #                           network's earliest allowed sequence. Alternate
 #                           networks may set this value. Minimum value of 1.

include/xrpl/nodestore/detail/DatabaseNodeImp.h

@@ -24,6 +24,32 @@ public:
         beast::Journal j)
         : Database(scheduler, readThreads, config, j), backend_(std::move(backend))
     {
+        std::optional<int> cacheSize, cacheAge;
+
+        if (config.exists("cache_size"))
+        {
+            cacheSize = get<int>(config, "cache_size");
+            if (cacheSize.value() < 0)
+            {
+                Throw<std::runtime_error>("Specified negative value for cache_size");
+            }
+        }
+
+        if (config.exists("cache_age"))
+        {
+            cacheAge = get<int>(config, "cache_age");
+            if (cacheAge.value() < 0)
+            {
+                Throw<std::runtime_error>("Specified negative value for cache_age");
+            }
+        }
+
+        if (cacheSize != 0 || cacheAge != 0)
+        {
+            cache_ = std::make_shared<TaggedCache<uint256, NodeObject>>(
+                "DatabaseNodeImp", cacheSize.value_or(0), std::chrono::minutes(cacheAge.value_or(0)), stopwatch(), j);
+        }
+
         XRPL_ASSERT(
             backend_,
             "xrpl::NodeStore::DatabaseNodeImp::DatabaseNodeImp : non-null "
@@ -82,6 +108,9 @@ public:
     sweep() override;
 
 private:
+    // Cache for database objects. This cache is not always initialized. Check
+    // for null before using.
+    std::shared_ptr<TaggedCache<uint256, NodeObject>> cache_;
     // Persistent key/value storage
     std::shared_ptr<Backend> backend_;
 

src/libxrpl/nodestore/DatabaseNodeImp.cpp

@@ -10,6 +10,11 @@ DatabaseNodeImp::store(NodeObjectType type, Blob&& data, uint256 const& hash, st
 
     auto obj = NodeObject::createObject(type, std::move(data), hash);
     backend_->store(obj);
+    if (cache_)
+    {
+        // After the store, replace a negative cache entry if there is one
+        cache_->canonicalize(hash, obj, [](std::shared_ptr<NodeObject> const& n) { return n->getType() == hotDUMMY; });
+    }
 }
 
 void
@@ -18,41 +23,77 @@ DatabaseNodeImp::asyncFetch(
     std::uint32_t ledgerSeq,
     std::function<void(std::shared_ptr<NodeObject> const&)>&& callback)
 {
+    if (cache_)
+    {
+        std::shared_ptr<NodeObject> obj = cache_->fetch(hash);
+        if (obj)
+        {
+            callback(obj->getType() == hotDUMMY ? nullptr : obj);
+            return;
+        }
+    }
     Database::asyncFetch(hash, ledgerSeq, std::move(callback));
 }
 
 void
 DatabaseNodeImp::sweep()
 {
+    if (cache_)
+        cache_->sweep();
 }
 
 std::shared_ptr<NodeObject>
 DatabaseNodeImp::fetchNodeObject(uint256 const& hash, std::uint32_t, FetchReport& fetchReport, bool duplicate)
 {
-    std::shared_ptr<NodeObject> nodeObject = nullptr;
-    Status status;
+    std::shared_ptr<NodeObject> nodeObject = cache_ ? cache_->fetch(hash) : nullptr;
 
-    try
+    if (!nodeObject)
     {
-        status = backend_->fetch(hash.data(), &nodeObject);
-    }
-    catch (std::exception const& e)
-    {
-        JLOG(j_.fatal()) << "fetchNodeObject " << hash << ": Exception fetching from backend: " << e.what();
-        Rethrow();
-    }
+        JLOG(j_.trace()) << "fetchNodeObject " << hash << ": record not " << (cache_ ? "cached" : "found");
 
-    switch (status)
+        Status status;
+
+        try
+        {
+            status = backend_->fetch(hash.data(), &nodeObject);
+        }
+        catch (std::exception const& e)
+        {
+            JLOG(j_.fatal()) << "fetchNodeObject " << hash << ": Exception fetching from backend: " << e.what();
+            Rethrow();
+        }
+
+        switch (status)
+        {
+            case ok:
+                if (cache_)
+                {
+                    if (nodeObject)
+                        cache_->canonicalize_replace_client(hash, nodeObject);
+                    else
+                    {
+                        auto notFound = NodeObject::createObject(hotDUMMY, {}, hash);
+                        cache_->canonicalize_replace_client(hash, notFound);
+                        if (notFound->getType() != hotDUMMY)
+                            nodeObject = notFound;
+                    }
+                }
+                break;
+            case notFound:
+                break;
+            case dataCorrupt:
+                JLOG(j_.fatal()) << "fetchNodeObject " << hash << ": nodestore data is corrupted";
+                break;
+            default:
+                JLOG(j_.warn()) << "fetchNodeObject " << hash << ": backend returns unknown result " << status;
+                break;
+        }
+    }
+    else
     {
-        case ok:
-        case notFound:
-            break;
-        case dataCorrupt:
-            JLOG(j_.fatal()) << "fetchNodeObject " << hash << ": nodestore data is corrupted";
-            break;
-        default:
-            JLOG(j_.warn()) << "fetchNodeObject " << hash << ": backend returns unknown result " << status;
-            break;
+        JLOG(j_.trace()) << "fetchNodeObject " << hash << ": record found in cache";
+        if (nodeObject->getType() == hotDUMMY)
+            nodeObject.reset();
     }
 
     if (nodeObject)
@@ -64,29 +105,66 @@ DatabaseNodeImp::fetchNodeObject(uint256 const& hash, std::uint32_t, FetchReport
 std::vector<std::shared_ptr<NodeObject>>
 DatabaseNodeImp::fetchBatch(std::vector<uint256> const& hashes)
 {
+    std::vector<std::shared_ptr<NodeObject>> results{hashes.size()};
     using namespace std::chrono;
     auto const before = steady_clock::now();
-
-    std::vector<uint256 const*> batch{hashes.size()};
+    std::unordered_map<uint256 const*, size_t> indexMap;
+    std::vector<uint256 const*> cacheMisses;
+    uint64_t hits = 0;
+    uint64_t fetches = 0;
     for (size_t i = 0; i < hashes.size(); ++i)
     {
         auto const& hash = hashes[i];
-        batch.push_back(&hash);
-    }
-
-    std::vector<std::shared_ptr<NodeObject>> results{hashes.size()};
-    results = backend_->fetchBatch(batch).first;
-    for (size_t i = 0; i < results.size(); ++i)
-    {
-        if (!results[i])
+        // See if the object already exists in the cache
+        auto nObj = cache_ ? cache_->fetch(hash) : nullptr;
+        ++fetches;
+        if (!nObj)
         {
-            JLOG(j_.error()) << "fetchBatch - "
-                             << "record not found in db. hash = " << strHex(hashes[i]);
+            // Try the database
+            indexMap[&hash] = i;
+            cacheMisses.push_back(&hash);
+        }
+        else
+        {
+            results[i] = nObj->getType() == hotDUMMY ? nullptr : nObj;
+            // It was in the cache.
+            ++hits;
         }
     }
 
+    JLOG(j_.debug()) << "fetchBatch - cache hits = " << (hashes.size() - cacheMisses.size())
+                     << " - cache misses = " << cacheMisses.size();
+    auto dbResults = backend_->fetchBatch(cacheMisses).first;
+
+    for (size_t i = 0; i < dbResults.size(); ++i)
+    {
+        auto nObj = std::move(dbResults[i]);
+        size_t index = indexMap[cacheMisses[i]];
+        auto const& hash = hashes[index];
+
+        if (nObj)
+        {
+            // Ensure all threads get the same object
+            if (cache_)
+                cache_->canonicalize_replace_client(hash, nObj);
+        }
+        else
+        {
+            JLOG(j_.error()) << "fetchBatch - "
+                             << "record not found in db or cache. hash = " << strHex(hash);
+            if (cache_)
+            {
+                auto notFound = NodeObject::createObject(hotDUMMY, {}, hash);
+                cache_->canonicalize_replace_client(hash, notFound);
+                if (notFound->getType() != hotDUMMY)
+                    nObj = std::move(notFound);
+            }
+        }
+        results[index] = std::move(nObj);
+    }
+
     auto fetchDurationUs = std::chrono::duration_cast<std::chrono::microseconds>(steady_clock::now() - before).count();
-    updateFetchMetrics(hashes.size(), 0, fetchDurationUs);
+    updateFetchMetrics(fetches, hits, fetchDurationUs);
     return results;
 }
 

src/test/app/SHAMapStore_test.cpp

@@ -490,8 +490,19 @@ public:
         Env env(*this, envconfig(onlineDelete));
 
         /////////////////////////////////////////////////////////////
-        // Create NodeStore with two backends to allow online deletion of data.
-        // Normally, SHAMapStoreImp handles all these details.
+        // Create the backend. Normally, SHAMapStoreImp handles all these
+        // details
+        auto nscfg = env.app().config().section(ConfigSection::nodeDatabase());
+
+        // Provide default values:
+        if (!nscfg.exists("cache_size"))
+            nscfg.set(
+                "cache_size", std::to_string(env.app().config().getValueFor(SizedItem::treeCacheSize, std::nullopt)));
+
+        if (!nscfg.exists("cache_age"))
+            nscfg.set(
+                "cache_age", std::to_string(env.app().config().getValueFor(SizedItem::treeCacheAge, std::nullopt)));
+
         NodeStoreScheduler scheduler(env.app().getJobQueue());
 
         std::string const writableDb = "write";
@@ -499,8 +510,9 @@ public:
         auto writableBackend = makeBackendRotating(env, scheduler, writableDb);
         auto archiveBackend = makeBackendRotating(env, scheduler, archiveDb);
 
+        // Create NodeStore with two backends to allow online deletion of
+        // data
         constexpr int readThreads = 4;
-        auto nscfg = env.app().config().section(ConfigSection::nodeDatabase());
         auto dbr = std::make_unique<NodeStore::DatabaseRotatingImp>(
             scheduler,
             readThreads,

src/xrpld/app/misc/SHAMapStoreImp.cpp

@@ -130,6 +130,14 @@ std::unique_ptr<NodeStore::Database>
 SHAMapStoreImp::makeNodeStore(int readThreads)
 {
     auto nscfg = app_.config().section(ConfigSection::nodeDatabase());
+
+    // Provide default values:
+    if (!nscfg.exists("cache_size"))
+        nscfg.set("cache_size", std::to_string(app_.config().getValueFor(SizedItem::treeCacheSize, std::nullopt)));
+
+    if (!nscfg.exists("cache_age"))
+        nscfg.set("cache_age", std::to_string(app_.config().getValueFor(SizedItem::treeCacheAge, std::nullopt)));
+
     std::unique_ptr<NodeStore::Database> db;
 
     if (deleteInterval_)
@@ -218,6 +226,8 @@ SHAMapStoreImp::run()
     LedgerIndex lastRotated = state_db_.getState().lastRotated;
     netOPs_ = &app_.getOPs();
     ledgerMaster_ = &app_.getLedgerMaster();
+    fullBelowCache_ = &(*app_.getNodeFamily().getFullBelowCache());
+    treeNodeCache_ = &(*app_.getNodeFamily().getTreeNodeCache());
 
     if (advisoryDelete_)
         canDelete_ = state_db_.getCanDelete();
@@ -480,13 +490,16 @@ void
 SHAMapStoreImp::clearCaches(LedgerIndex validatedSeq)
 {
     ledgerMaster_->clearLedgerCachePrior(validatedSeq);
+    fullBelowCache_->clear();
 }
 
 void
 SHAMapStoreImp::freshenCaches()
 {
-    freshenCache(*app_.getNodeFamily().getTreeNodeCache());
-    freshenCache(app_.getMasterTransaction().getCache());
+    if (freshenCache(*treeNodeCache_))
+        return;
+    if (freshenCache(app_.getMasterTransaction().getCache()))
+        return;
 }
 
 void

src/xrpld/app/misc/SHAMapStoreImp.h

@@ -94,6 +94,8 @@ private:
     // as of run() or before
     NetworkOPs* netOPs_ = nullptr;
     LedgerMaster* ledgerMaster_ = nullptr;
+    FullBelowCache* fullBelowCache_ = nullptr;
+    TreeNodeCache* treeNodeCache_ = nullptr;
 
     static constexpr auto nodeStoreName_ = "NodeStore";
 

src/xrpld/app/misc/detail/ValidatorSite.cpp

@@ -124,7 +124,11 @@ ValidatorSite::load(std::vector<std::string> const& siteURIs, std::lock_guard<st
     {
         try
         {
-            sites_.emplace_back(uri);
+            // This is not super efficient, but it doesn't happen often.
+            bool found =
+                std::ranges::any_of(sites_, [&uri](auto const& site) { return site.loadedResource->uri == uri; });
+            if (!found)
+                sites_.emplace_back(uri);
         }
         catch (std::exception const& e)
         {
@@ -183,6 +187,15 @@ ValidatorSite::stop()
 void
 ValidatorSite::setTimer(std::lock_guard<std::mutex> const& site_lock, std::lock_guard<std::mutex> const& state_lock)
 {
+    if (!sites_.empty() &&  //
+        std::ranges::all_of(sites_, [](auto const& site) { return site.lastRefreshStatus.has_value(); }))
+    {
+        // If all of the sites have been handled at least once (including
+        // errors and timeouts), call missingSite, which will load the cache
+        // files for any lists that are still unavailable.
+        missingSite(site_lock);
+    }
+
     auto next = std::min_element(
         sites_.begin(), sites_.end(), [](Site const& a, Site const& b) { return a.nextRefresh < b.nextRefresh; });
 
@@ -285,12 +298,14 @@ ValidatorSite::onRequestTimeout(std::size_t siteIdx, error_code const& ec)
         // processes a network error. Usually, this function runs first,
         // but on extremely rare occasions, the response handler can run
         // first, which will leave activeResource empty.
-        auto const& site = sites_[siteIdx];
+        auto& site = sites_[siteIdx];
         if (site.activeResource)
             JLOG(j_.warn()) << "Request for " << site.activeResource->uri << " took too long";
         else
             JLOG(j_.error()) << "Request took too long, but a response has "
                                 "already been processed";
+        if (!site.lastRefreshStatus)
+            site.lastRefreshStatus.emplace(Site::Status{clock_type::now(), ListDisposition::invalid, "timeout"});
     }
 
     std::lock_guard lock_state{state_mutex_};