Maintain history back to the earliest persisted ledger:

This makes behavior consistent with configurations both with and
without online delete.

src/ripple/app/ledger/LedgerMaster.h

@@ -39,6 +39,7 @@
 #include <ripple/protocol/Protocol.h>
 #include <ripple/protocol/RippleLedgerHash.h>
 #include <ripple/protocol/STValidation.h>
+#include <boost/optional.hpp>
 
 #include <mutex>
 
@@ -280,11 +281,6 @@ private:
     // Try to publish ledgers, acquire missing ledgers.  Always called with
     // m_mutex locked.  The passed lock is a reminder to callers.
     void doAdvance(std::unique_lock<std::recursive_mutex>&);
-    bool shouldAcquire(
-        std::uint32_t const currentLedger,
-        std::uint32_t const ledgerHistory,
-        std::uint32_t const ledgerHistoryIndex,
-        std::uint32_t const candidateLedger) const;
 
     std::vector<std::shared_ptr<Ledger const>>
     findNewLedgersToPublish(std::unique_lock<std::recursive_mutex>&);
@@ -295,7 +291,6 @@ private:
     // The passed lock is a reminder to callers.
     bool newPFWork(const char *name, std::unique_lock<std::recursive_mutex>&);
 
-private:
     Application& app_;
     beast::Journal m_journal;
 

src/ripple/app/ledger/impl/LedgerMaster.cpp

@@ -38,6 +38,7 @@
 #include <ripple/basics/Log.h>
 #include <ripple/basics/TaggedCache.h>
 #include <ripple/basics/UptimeClock.h>
+#include <ripple/core/DatabaseCon.h>
 #include <ripple/core/TimeKeeper.h>
 #include <ripple/nodestore/DatabaseShard.h>
 #include <ripple/overlay/Overlay.h>
@@ -47,6 +48,7 @@
 #include <ripple/resource/Fees.h>
 #include <algorithm>
 #include <cassert>
+#include <limits>
 #include <memory>
 #include <vector>
 
@@ -140,6 +142,57 @@ static constexpr std::chrono::minutes MAX_LEDGER_AGE_ACQUIRE {1};
 // Don't acquire history if write load is too high
 static constexpr int MAX_WRITE_LOAD_ACQUIRE {8192};
 
+// Helper function for LedgerMaster::doAdvance()
+// Returns the minimum ledger sequence in SQL database, if any.
+static boost::optional<LedgerIndex>
+minSqlSeq(Application& app)
+{
+    boost::optional<LedgerIndex> seq;
+    auto db = app.getLedgerDB().checkoutDb();
+    *db << "SELECT MIN(LedgerSeq) FROM Ledgers", soci::into(seq);
+    return seq;
+}
+
+// Helper function for LedgerMaster::doAdvance()
+// Return true if candidateLedger should be fetched from the network.
+static bool
+shouldAcquire (
+    std::uint32_t const currentLedger,
+    std::uint32_t const ledgerHistory,
+    boost::optional<LedgerIndex> minSeq,
+    std::uint32_t const lastRotated,
+    std::uint32_t const candidateLedger,
+    beast::Journal j)
+{
+    bool ret = [&]()
+    {
+        // Fetch ledger if it may be the current ledger
+        if (candidateLedger >= currentLedger)
+            return true;
+
+        // Or if it is within our configured history range:
+        if (currentLedger - candidateLedger <= ledgerHistory)
+            return true;
+
+        // Or it's greater than or equal to both:
+        // - the minimum persisted ledger or the maximum possible
+        //   sequence value, if no persisted ledger, and
+        // - minimum ledger that will be persisted as of the next online
+        //   deletion interval, or 1 if online deletion is disabled.
+        return
+            candidateLedger >= std::max(
+                minSeq.value_or(std::numeric_limits<LedgerIndex>::max()),
+                lastRotated + 1);
+    }();
+
+    JLOG (j.trace())
+        << "Missing ledger "
+        << candidateLedger
+        << (ret ? " should" : " should NOT")
+        << " be acquired";
+    return ret;
+}
+
 LedgerMaster::LedgerMaster (Application& app, Stopwatch& stopwatch,
     Stoppable& parent,
     beast::insight::Collector::ptr const& collector, beast::Journal journal)
@@ -1697,31 +1750,6 @@ LedgerMaster::releaseReplay ()
     return std::move (replayData);
 }
 
-bool
-LedgerMaster::shouldAcquire (
-    std::uint32_t const currentLedger,
-    std::uint32_t const ledgerHistory,
-    std::uint32_t const ledgerHistoryIndex,
-    std::uint32_t const candidateLedger) const
-{
-
-    // Fetch ledger if it might be the current ledger,
-    // is requested by the advisory delete setting, or
-    // is within our configured history range
-
-    bool ret (candidateLedger >= currentLedger ||
-        ((ledgerHistoryIndex > 0) &&
-            (candidateLedger > ledgerHistoryIndex)) ||
-        (currentLedger - candidateLedger) <= ledgerHistory);
-
-    JLOG (m_journal.trace())
-        << "Missing ledger "
-        << candidateLedger
-        << (ret ? " should" : " should NOT")
-        << " be acquired";
-    return ret;
-}
-
 void
 LedgerMaster::fetchForHistory(
     std::uint32_t missing,
@@ -1875,7 +1903,9 @@ void LedgerMaster::doAdvance (std::unique_lock<std::recursive_mutex>& sl)
                         "tryAdvance discovered missing " << *missing;
                     if ((mFillInProgress == 0 || *missing > mFillInProgress) &&
                         shouldAcquire(mValidLedgerSeq, ledger_history_,
-                            app_.getSHAMapStore().getCanDelete(), *missing))
+                            minSqlSeq(app_),
+                            app_.getSHAMapStore().getLastRotated(), *missing,
+                            m_journal))
                     {
                         JLOG(m_journal.trace()) <<
                             "advanceThread should acquire";

src/ripple/app/misc/SHAMapStore.h

@@ -56,7 +56,9 @@ public:
     /** Whether advisory delete is enabled. */
     virtual bool advisoryDelete() const = 0;
 
-    /** Last ledger which was copied during rotation of backends. */
+    /** Maximum ledger that has been deleted, or will be deleted if
+     *  currently in the act of online deletion.
+     */
     virtual LedgerIndex getLastRotated() = 0;
 
     /** Highest ledger that may be deleted. */

src/ripple/app/misc/SHAMapStoreImp.cpp

@@ -322,7 +322,7 @@ void
 SHAMapStoreImp::run()
 {
     beast::setCurrentThreadName ("SHAMapStore");
-    LedgerIndex lastRotated = state_db_.getState().lastRotated;
+    lastRotated_ = state_db_.getState().lastRotated;
     netOPs_ = &app_.getOPs();
     ledgerMaster_ = &app_.getLedgerMaster();
     fullBelowCache_ = &app_.family().fullbelow();
@@ -333,7 +333,7 @@ SHAMapStoreImp::run()
     if (advisoryDelete_)
         canDelete_ = state_db_.getCanDelete ();
 
-    while (1)
+    while (true)
     {
         healthy_ = true;
         std::shared_ptr<Ledger const> validatedLedger;
@@ -356,20 +356,20 @@ SHAMapStoreImp::run()
                 continue;
         }
 
-        LedgerIndex validatedSeq = validatedLedger->info().seq;
-        if (!lastRotated)
+        LedgerIndex const validatedSeq = validatedLedger->info().seq;
+        if (!lastRotated_)
         {
-            lastRotated = validatedSeq;
-            state_db_.setLastRotated (lastRotated);
+            lastRotated_ = validatedSeq;
+            state_db_.setLastRotated (lastRotated_);
         }
 
-        // will delete up to (not including) lastRotated)
-        if (validatedSeq >= lastRotated + deleteInterval_
-                && canDelete_ >= lastRotated - 1)
+        // will delete up to (not including) lastRotated_
+        if (validatedSeq >= lastRotated_ + deleteInterval_
+                && canDelete_ >= lastRotated_ - 1)
         {
             JLOG(journal_.warn()) << "rotating  validatedSeq " << validatedSeq
-                    << " lastRotated " << lastRotated << " deleteInterval "
-                    << deleteInterval_ << " canDelete_ " << canDelete_;
+                << " lastRotated_ " << lastRotated_ << " deleteInterval "
+                << deleteInterval_ << " canDelete_ " << canDelete_;
 
             switch (health())
             {
@@ -383,7 +383,7 @@ SHAMapStoreImp::run()
                     ;
             }
 
-            clearPrior (lastRotated);
+            clearPrior (lastRotated_);
             switch (health())
             {
                 case Health::stopping:
@@ -448,13 +448,13 @@ SHAMapStoreImp::run()
 
             std::string nextArchiveDir =
                 dbRotating_->getWritableBackend()->getName();
-            lastRotated = validatedSeq;
+            lastRotated_ = validatedSeq;
             std::shared_ptr<NodeStore::Backend> oldBackend;
             {
                 std::lock_guard lock (dbRotating_->peekMutex());
 
                 state_db_.setState (SavedState {newBackend->getName(),
-                        nextArchiveDir, lastRotated});
+                                                nextArchiveDir, lastRotated_});
                 clearCaches (validatedSeq);
                 oldBackend = dbRotating_->rotateBackends(
                     std::move(newBackend),

src/ripple/app/misc/SHAMapStoreImp.h

@@ -25,6 +25,7 @@
 #include <ripple/core/DatabaseCon.h>
 #include <ripple/nodestore/DatabaseRotating.h>
 
+#include <atomic>
 #include <condition_variable>
 #include <thread>
 
@@ -84,6 +85,13 @@ private:
     static std::uint32_t const minimumDeletionInterval_ = 256;
     // minimum # of ledgers required for standalone mode.
     static std::uint32_t const minimumDeletionIntervalSA_ = 8;
+    // Ledger sequence at which the last deletion interval was triggered,
+    // or the current validated sequence as of first use
+    // if there have been no prior deletions. Deletion occurs up to (but
+    // not including) this value. All ledgers past this value are accumulated
+    // until the next online deletion. This value is persisted to SQLite
+    // nearly immediately after modification.
+    std::atomic<LedgerIndex> lastRotated_ {};
 
     NodeStore::Scheduler& scheduler_;
     beast::Journal const journal_;
@@ -159,7 +167,7 @@ public:
     LedgerIndex
     getLastRotated() override
     {
-        return state_db_.getState().lastRotated;
+        return lastRotated_;
     }
 
     // All ledgers before and including this are unprotected