Merge #10148: Use non-atomic flushing with block replay

176c021 [qa] Test non-atomic chainstate writes (Suhas Daftuar)
d6af06d Dont create pcoinsTip until after ReplayBlocks. (Matt Corallo)
eaca1b7 Random db flush crash simulator (Pieter Wuille)
0580ee0 Adapt memory usage estimation for flushing (Pieter Wuille)
013a56a Non-atomic flushing using the blockchain as replay journal (Pieter Wuille)
b3a279c [MOVEONLY] Move LastCommonAncestor to chain (Pieter Wuille)

Tree-SHA512: 47ccc62303f9075c44d2a914be75bd6969ff881a857a2ff1227f05ec7def6f4c71c46680c5a28cb150c814999526797dc05cf2701fde1369c06169f46eccddee

1 files changed, 49 insertions, 2 deletions

diff --git a/src/txdb.cpp b/src/txdb.cpp
index 97e916fd22..d24162ba2d 100644
--- a/src/txdb.cpp
+++ b/src/txdb.cpp
@@ -7,8 +7,10 @@
 
 #include "chainparams.h"
 #include "hash.h"
+#include "random.h"
 #include "pow.h"
 #include "uint256.h"
+#include "util.h"
 
 #include <stdint.h>
 
@@ -21,6 +23,7 @@ static const char DB_TXINDEX = 't';
 static const char DB_BLOCK_INDEX = 'b';
 
 static const char DB_BEST_BLOCK = 'B';
+static const char DB_HEAD_BLOCKS = 'H';
 static const char DB_FLAG = 'F';
 static const char DB_REINDEX_FLAG = 'R';
 static const char DB_LAST_BLOCK = 'l';
@@ -68,10 +71,39 @@ uint256 CCoinsViewDB::GetBestBlock() const {
     return hashBestChain;
 }
 
+std::vector<uint256> CCoinsViewDB::GetHeadBlocks() const {
+    std::vector<uint256> vhashHeadBlocks;
+    if (!db.Read(DB_HEAD_BLOCKS, vhashHeadBlocks)) {
+        return std::vector<uint256>();
+    }
+    return vhashHeadBlocks;
+}
+
 bool CCoinsViewDB::BatchWrite(CCoinsMap &mapCoins, const uint256 &hashBlock) {
     CDBBatch batch(db);
     size_t count = 0;
     size_t changed = 0;
+    size_t batch_size = (size_t)GetArg("-dbbatchsize", nDefaultDbBatchSize);
+    int crash_simulate = GetArg("-dbcrashratio", 0);
+    assert(!hashBlock.IsNull());
+
+    uint256 old_tip = GetBestBlock();
+    if (old_tip.IsNull()) {
+        // We may be in the middle of replaying.
+        std::vector<uint256> old_heads = GetHeadBlocks();
+        if (old_heads.size() == 2) {
+            assert(old_heads[0] == hashBlock);
+            old_tip = old_heads[1];
+        }
+    }
+
+    // In the first batch, mark the database as being in the middle of a
+    // transition from old_tip to hashBlock.
+    // A vector is used for future extensibility, as we may want to support
+    // interrupting after partial writes from multiple independent reorgs.
+    batch.Erase(DB_BEST_BLOCK);
+    batch.Write(DB_HEAD_BLOCKS, std::vector<uint256>{hashBlock, old_tip});
+
     for (CCoinsMap::iterator it = mapCoins.begin(); it != mapCoins.end();) {
         if (it->second.flags & CCoinsCacheEntry::DIRTY) {
             CoinEntry entry(&it->first);
@@ -84,10 +116,25 @@ bool CCoinsViewDB::BatchWrite(CCoinsMap &mapCoins, const uint256 &hashBlock) {
         count++;
         CCoinsMap::iterator itOld = it++;
         mapCoins.erase(itOld);
+        if (batch.SizeEstimate() > batch_size) {
+            LogPrint(BCLog::COINDB, "Writing partial batch of %.2f MiB\n", batch.SizeEstimate() * (1.0 / 1048576.0));
+            db.WriteBatch(batch);
+            batch.Clear();
+            if (crash_simulate) {
+                static FastRandomContext rng;
+                if (rng.randrange(crash_simulate) == 0) {
+                    LogPrintf("Simulating a crash. Goodbye.\n");
+                    _Exit(0);
+                }
+            }
+        }
     }
-    if (!hashBlock.IsNull())
-        batch.Write(DB_BEST_BLOCK, hashBlock);
 
+    // In the last batch, mark the database as consistent with hashBlock again.
+    batch.Erase(DB_HEAD_BLOCKS);
+    batch.Write(DB_BEST_BLOCK, hashBlock);
+
+    LogPrint(BCLog::COINDB, "Writing final batch of %.2f MiB\n", batch.SizeEstimate() * (1.0 / 1048576.0));
     bool ret = db.WriteBatch(batch);
     LogPrint(BCLog::COINDB, "Committed %u changed transaction outputs (out of %u) to coin database...\n", (unsigned int)changed, (unsigned int)count);
     return ret;