Fix rollback tests hanging and block hash identity issues

Author: Cmdv

cardano-chain-gen/src/Cardano/Mock/ChainSync/Server.hs

@@ -369,8 +369,10 @@ chainSyncServer state codec _blockVersion =
       atomically $ do
         cps <- readTVar state
         let chain = chainDB cps
+            chainTip = headTip chain
         case findFirstPoint (map castPoint points) chain of
-          Nothing -> pure (Nothing, castTip (headTip chain))
+          Nothing ->
+            pure (Nothing, castTip chainTip)
           Just ipoint -> do
             let !cps' = updateFollower rid ipoint cps
             writeTVar state cps'

cardano-chain-gen/test/Test/Cardano/Db/Mock/Config.hs

@@ -57,6 +57,7 @@ module Test.Cardano.Db.Mock.Config (
   withFullConfig,
   withFullConfigDropDB,
   withFullConfigDropDBLog,
+  withFullConfigDropDBNoFingerprint,
   withFullConfigLog,
   withCustomConfigDropDBLog,
   withCustomConfig,
@@ -470,6 +471,27 @@ withFullConfigDropDBLog =
     initCommandLineArgs
     Nothing
 
+-- For tests that rollback and restart, fingerprints create divergent chains
+withFullConfigDropDBNoFingerprint ::
+  -- | config filepath
+  FilePath ->
+  -- | test label
+  FilePath ->
+  (Interpreter -> ServerHandle IO CardanoBlock -> DBSyncEnv -> IO a) ->
+  IOManager ->
+  [(Text, Text)] ->
+  IO a
+withFullConfigDropDBNoFingerprint =
+  withFullConfig'
+    ( WithConfigArgs
+        { hasFingerprint = False
+        , shouldLog = False
+        , shouldDropDB = True
+        }
+    )
+    initCommandLineArgs
+    Nothing
+
 withFullConfigLog ::
   -- | config filepath
   FilePath ->

cardano-chain-gen/test/Test/Cardano/Db/Mock/Unit/Conway.hs

@@ -121,6 +121,7 @@ unitTests iom knownMigrations =
         , test "drepDistr rollback" Rollback.drepDistrRollback
         , test "sync bigger chain" Rollback.bigChain
         , test "rollback while db-sync is off" Rollback.restartAndRollback
+        , test "large rollback while db-sync is off" Rollback.restartAndRollbackLarge
         , test "big rollback executed lazily" Rollback.lazyRollback
         , test "lazy rollback on restart" Rollback.lazyRollbackRestart
         , test "rollback while rollbacking" Rollback.doubleRollback

cardano-chain-gen/test/Test/Cardano/Db/Mock/Unit/Conway/Rollback.hs

@@ -9,6 +9,7 @@ module Test.Cardano.Db.Mock.Unit.Conway.Rollback (
   simpleRollback,
   bigChain,
   restartAndRollback,
+  restartAndRollbackLarge,
   lazyRollback,
   lazyRollbackRestart,
   doubleRollback,
@@ -35,7 +36,7 @@ import Cardano.Mock.Forging.Types (PoolIndex (..), StakeIndex (..), UTxOIndex (.
 import Cardano.Prelude
 import Data.Maybe.Strict (StrictMaybe (..))
 import Ouroboros.Network.Block (blockPoint)
-import Test.Cardano.Db.Mock.Config (claFullMode, configPoolStats, conwayConfigDir, initCommandLineArgs, queryDBSync, startDBSync, stopDBSync, withCustomConfigDropDB, withFullConfigDropDB)
+import Test.Cardano.Db.Mock.Config (claFullMode, configPoolStats, conwayConfigDir, initCommandLineArgs, queryDBSync, startDBSync, stopDBSync, withCustomConfigDropDB, withFullConfigDropDB, withFullConfigDropDBNoFingerprint)
 import Test.Cardano.Db.Mock.Examples
 import Test.Cardano.Db.Mock.UnifiedApi
 import Test.Cardano.Db.Mock.Validate (assertBlockNoBackoff, assertEqQuery, assertTxCount)
@@ -94,7 +95,7 @@ bigChain =
 
 restartAndRollback :: IOManager -> [(Text, Text)] -> Assertion
 restartAndRollback =
-  withFullConfigDropDB conwayConfigDir testLabel $ \interpreter mockServer dbSync -> do
+  withFullConfigDropDBNoFingerprint conwayConfigDir testLabel $ \interpreter mockServer dbSync -> do
     -- Forge some blocks
     forM_ (replicate 101 mockBlock0) (forgeNextAndSubmit interpreter mockServer)
 
@@ -107,20 +108,49 @@ restartAndRollback =
     -- Wait for it to sync
     assertBlockNoBackoff dbSync 201
 
-    -- Forge some more blocks
-    forM_ (replicate 5 mockBlock2) (forgeNextAndSubmit interpreter mockServer)
+    -- Forge some more blocks (using mockBlock0 so the new block after rollback will have a different forger)
+    forM_ (replicate 5 mockBlock0) (forgeNextAndSubmit interpreter mockServer)
     assertBlockNoBackoff dbSync 206
 
     -- Rollback and restart
     stopDBSync dbSync
     void $ rollbackTo interpreter mockServer (blockPoint $ last blks)
     startDBSync dbSync
 
-    -- TODO: DBSync doesn't detect rollbacks while stopped
-    assertBlockNoBackoff dbSync 206
+    -- rollbackTo forges an empty block after rollback, so we expect 202 blocks
+    -- (rollback point at 201 + the new empty block)
+    assertBlockNoBackoff dbSync 202
   where
     testLabel = "conwayRestartAndRollback"
 
+restartAndRollbackLarge :: IOManager -> [(Text, Text)] -> Assertion
+restartAndRollbackLarge =
+  withFullConfigDropDBNoFingerprint conwayConfigDir testLabel $ \interpreter mockServer dbSync -> do
+    -- Forge initial blocks
+    forM_ (replicate 101 mockBlock0) (forgeNextAndSubmit interpreter mockServer)
+
+    startDBSync dbSync
+    assertBlockNoBackoff dbSync 101
+
+    -- First batch: blocks to rollback to
+    blks1 <- forM (replicate 100 mockBlock0) (forgeNextAndSubmit interpreter mockServer)
+    assertBlockNoBackoff dbSync 201
+
+    -- Second batch: blocks that will be rolled back
+    forM_ (replicate 100 mockBlock0) (forgeNextAndSubmit interpreter mockServer)
+    assertBlockNoBackoff dbSync 301
+
+    -- Rollback and restart - rolling back 100 blocks
+    stopDBSync dbSync
+    void $ rollbackTo interpreter mockServer (blockPoint $ last blks1)
+    startDBSync dbSync
+
+    -- rollbackTo forges an empty block after rollback, so we expect 202 blocks
+    -- (rollback point at 201 + the new empty block)
+    assertBlockNoBackoff dbSync 202
+  where
+    testLabel = "conwayRestartAndRollbackLarge"
+
 lazyRollback :: IOManager -> [(Text, Text)] -> Assertion
 lazyRollback =
   withFullConfigDropDB conwayConfigDir testLabel $ \interpreter mockServer dbSync -> do

cardano-db-sync/src/Cardano/DbSync/Api.hs

@@ -448,7 +448,14 @@ getLatestPoints env = do
   case envLedgerEnv env of
     HasLedger hasLedgerEnv -> do
       snapshotPoints <- listKnownSnapshots hasLedgerEnv
-      verifySnapshotPoint env snapshotPoints
+      if null snapshotPoints
+        then do
+          -- Fallback: When no snapshots available (e.g., after restart),
+          -- query database for recent blocks to use as intersection points
+          lastPoints <- DB.runDbDirectSilent (envDbEnv env) DB.queryLatestPoints
+          pure $ mapMaybe convert lastPoints
+        else
+          verifySnapshotPoint env snapshotPoints
     NoLedger _ -> do
       -- Brings the 5 latest.
       lastPoints <- DB.runDbDirectSilent (envDbEnv env) DB.queryLatestPoints

cardano-db-sync/src/Cardano/DbSync/Sync.hs

@@ -318,7 +318,11 @@ chainSyncClient metricsSetters trce latestPoints currentTip tc = do
       ClientPipelinedStIdle n CardanoBlock (Point CardanoBlock) (Tip CardanoBlock) IO ()
     go mkPipelineDecision n clientTip serverTip mPoint =
       case (mPoint, n, runPipelineDecision mkPipelineDecision n clientTip serverTip) of
-        (Just points, _, _) -> drainThePipe n $ clientPipelinedStIdle clientTip points
+        (Just points, _, _) ->
+          -- When re-intersecting after rollback failure, reset clientTip to Origin
+          -- if falling back to genesis, otherwise keep current clientTip
+          let newClientTip = if points == [genesisPoint] then Origin else clientTip
+           in drainThePipe n $ clientPipelinedStIdle newClientTip points
         (_, _Zero, (Request, mkPipelineDecision')) ->
           SendMsgRequestNext (pure ()) clientStNext
           where