Merge pull request #2006 from IntersectMBO/1992-custom-snapshots-freq

1992-custom-snapshot-intervals

Author: kderme

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

@@ -295,8 +295,6 @@ mkSyncNodeParams staticDir mutableDir CommandLineArgs {..} = do
       , enpHasCache = claHasCache
       , enpForceIndexes = claForceIndexes
       , enpHasInOut = True
-      , enpSnEveryFollowing = 35
-      , enpSnEveryLagging = 35
       , enpMaybeRollback = Nothing
       }
 

cardano-db-sync/app/cardano-db-sync.hs

@@ -89,8 +89,6 @@ pRunDbSyncNode = do
     <*> pHasCache
     <*> pForceIndexes
     <*> pHasInOut
-    <*> pure 500
-    <*> pure 10000
     <*> optional pRollbackSlotNo
 
 pConfigFile :: Parser ConfigFile

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

@@ -295,8 +295,7 @@ extractSyncOptions snp aop snc =
           isTxOutConsumedBootstrap'
           forceTxIn'
     , soptInsertOptions = iopts
-    , snapshotEveryFollowing = enpSnEveryFollowing snp
-    , snapshotEveryLagging = enpSnEveryLagging snp
+    , soptSnapshotInterval = dncSnapshotInterval snc
     }
   where
     maybeKeepMNames =

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

@@ -28,7 +28,7 @@ import Ouroboros.Network.Magic (NetworkMagic (..))
 
 import qualified Cardano.Db as DB
 import Cardano.DbSync.Cache.Types (CacheStatistics, CacheStatus)
-import Cardano.DbSync.Config.Types (SyncNodeConfig)
+import Cardano.DbSync.Config.Types (SnapshotIntervalConfig, SyncNodeConfig)
 import Cardano.DbSync.Ledger.Types (HasLedgerEnv)
 import Cardano.DbSync.LocalStateQuery (NoLedgerEnv)
 import Cardano.DbSync.Types (
@@ -70,8 +70,7 @@ data SyncOptions = SyncOptions
   , soptCache :: !Bool
   , soptPruneConsumeMigration :: !DB.PruneConsumeMigration
   , soptInsertOptions :: !InsertOptions
-  , snapshotEveryFollowing :: !Word64
-  , snapshotEveryLagging :: !Word64
+  , soptSnapshotInterval :: !SnapshotIntervalConfig
   }
   deriving (Show)
 

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

@@ -92,6 +92,7 @@ coalesceConfig pcfg ncfg adjustGenesisPath = do
       , dncConwayHardFork = ncConwayHardFork ncfg
       , dncInsertOptions = extractInsertOptions pcfg
       , dncIpfsGateway = endsInSlash <$> pcIpfsGateway pcfg
+      , dncSnapshotInterval = pcSnapshotInterval pcfg
       }
 
 mkAdjustPath :: SyncPreConfig -> (FilePath -> FilePath)

cardano-db-sync/src/Cardano/DbSync/Config/Types.hs

@@ -39,6 +39,7 @@ module Cardano.DbSync.Config.Types (
   GovernanceConfig (..),
   OffchainPoolDataConfig (..),
   JsonTypeConfig (..),
+  SnapshotIntervalConfig (..),
   LedgerStateDir (..),
   LogFileDir (..),
   NetworkName (..),
@@ -85,6 +86,8 @@ import Ouroboros.Consensus.Protocol.Praos (Praos)
 import Ouroboros.Consensus.Protocol.TPraos (TPraos)
 import Ouroboros.Consensus.Shelley.Ledger.Block (ShelleyBlock)
 
+{- HLINT ignore "Reduce duplication" -}
+
 newtype LogFileDir = LogFileDir
   { unLogFileDir :: FilePath
   }
@@ -109,8 +112,6 @@ data SyncNodeParams = SyncNodeParams
   , enpHasCache :: !Bool
   , enpForceIndexes :: !Bool
   , enpHasInOut :: !Bool
-  , enpSnEveryFollowing :: !Word64
-  , enpSnEveryLagging :: !Word64
   , enpMaybeRollback :: !(Maybe SlotNo)
   }
   deriving (Show)
@@ -147,6 +148,7 @@ data SyncNodeConfig = SyncNodeConfig
   , dncConwayHardFork :: !(CardanoHardForkTrigger (ShelleyBlock (Praos StandardCrypto) ConwayEra))
   , dncInsertOptions :: !SyncInsertOptions
   , dncIpfsGateway :: [Text]
+  , dncSnapshotInterval :: !SnapshotIntervalConfig
   }
 
 data SyncPreConfig = SyncPreConfig
@@ -159,6 +161,7 @@ data SyncPreConfig = SyncPreConfig
   , pcPrometheusPort :: !Int
   , pcInsertConfig :: !SyncInsertConfig
   , pcIpfsGateway :: ![Text]
+  , pcSnapshotInterval :: !SnapshotIntervalConfig
   }
   deriving (Show)
 
@@ -279,6 +282,11 @@ data JsonTypeConfig
   | JsonTypeDisable
   deriving (Eq, Show)
 
+newtype SnapshotIntervalConfig = SnapshotIntervalConfig
+  { sicNearTipEpoch :: Word64
+  }
+  deriving (Eq, Show)
+
 newtype GenesisFile = GenesisFile
   { unGenesisFile :: FilePath
   }
@@ -407,6 +415,7 @@ parseGenSyncNodeConfig o =
     <*> fmap (fromMaybe 8080) (o .:? "PrometheusPort")
     <*> o .:? "insert_options" .!= def
     <*> o .:? "ipfs_gateway" .!= ["https://ipfs.io/ipfs"]
+    <*> o .:? "snapshot_interval" .!= def
 
 instance FromJSON SyncProtocol where
   parseJSON o =
@@ -728,6 +737,23 @@ instance FromJSON JsonTypeConfig where
     "disable" -> pure JsonTypeDisable
     other -> fail $ "unexpected json_type: " <> show other
 
+instance ToJSON SnapshotIntervalConfig where
+  toJSON cfg =
+    Aeson.object
+      [ "near_tip_epoch" .= sicNearTipEpoch cfg
+      ]
+
+instance FromJSON SnapshotIntervalConfig where
+  parseJSON = Aeson.withObject "snapshot_interval" $ \obj ->
+    SnapshotIntervalConfig
+      <$> obj .:? "near_tip_epoch" .!= sicNearTipEpoch def
+
+instance Default SnapshotIntervalConfig where
+  def =
+    SnapshotIntervalConfig
+      { sicNearTipEpoch = 580 -- Epoch threshold to consider being near tip
+      }
+
 instance Default SyncInsertConfig where
   def = SyncInsertConfig Nothing def
 

cardano-db-sync/src/Cardano/DbSync/Ledger/State.hs

@@ -54,7 +54,6 @@ import qualified Cardano.Ledger.BaseTypes as Ledger
 import Cardano.Ledger.Shelley.AdaPots (AdaPots)
 import qualified Cardano.Ledger.Shelley.LedgerState as Shelley
 import Cardano.Prelude hiding (atomically)
-import Cardano.Slotting.Block (BlockNo (..))
 import Cardano.Slotting.EpochInfo (EpochInfo, epochInfoEpoch)
 import Cardano.Slotting.Slot (
   EpochNo (..),
@@ -128,7 +127,7 @@ import qualified Ouroboros.Consensus.Shelley.Ledger.Ledger as Consensus
 import Ouroboros.Consensus.Storage.Serialisation (DecodeDisk (..), EncodeDisk (..))
 import Ouroboros.Network.AnchoredSeq (AnchoredSeq (..))
 import qualified Ouroboros.Network.AnchoredSeq as AS
-import Ouroboros.Network.Block (HeaderHash, Point (..), blockNo)
+import Ouroboros.Network.Block (HeaderHash, Point (..))
 import qualified Ouroboros.Network.Point as Point
 import System.Directory (doesFileExist, listDirectory, removeFile)
 import System.FilePath (dropExtension, takeExtension, (</>))
@@ -186,8 +185,7 @@ mkHasLedgerEnv trce protoInfo dir nw systemStart syncOptions = do
       , leNetwork = nw
       , leSystemStart = systemStart
       , leAbortOnPanic = soptAbortOnInvalid syncOptions
-      , leSnapshotEveryFollowing = snapshotEveryFollowing syncOptions
-      , leSnapshotEveryLagging = snapshotEveryLagging syncOptions
+      , leSnapshotNearTipEpoch = sicNearTipEpoch $ soptSnapshotInterval syncOptions
       , leInterpreter = intervar
       , leStateVar = svar
       , leStateWriteQueue = swQueue
@@ -218,7 +216,8 @@ readStateUnsafe env = do
 applyBlockAndSnapshot :: HasLedgerEnv -> CardanoBlock -> Bool -> IO (ApplyResult, Bool)
 applyBlockAndSnapshot ledgerEnv blk isCons = do
   (oldState, appResult) <- applyBlock ledgerEnv blk
-  tookSnapshot <- storeSnapshotAndCleanupMaybe ledgerEnv oldState appResult (blockNo blk) isCons (isSyncedWithinSeconds (apSlotDetails appResult) 600)
+  -- 864000 seconds = 10 days; consider synced "near tip" if within 10 days of current time
+  tookSnapshot <- storeSnapshotAndCleanupMaybe ledgerEnv oldState appResult isCons (isSyncedWithinSeconds (apSlotDetails appResult) 864000)
   pure (appResult, tookSnapshot)
 
 -- The function 'tickThenReapply' does zero validation, so add minimal validation ('blockPrevHash'
@@ -321,32 +320,21 @@ storeSnapshotAndCleanupMaybe ::
   HasLedgerEnv ->
   CardanoLedgerState ->
   ApplyResult ->
-  BlockNo ->
   Bool ->
   SyncState ->
   IO Bool
-storeSnapshotAndCleanupMaybe env oldState appResult blkNo isCons syncState =
+storeSnapshotAndCleanupMaybe env oldState appResult isCons syncState =
   case maybeFromStrict (apNewEpoch appResult) of
     Just newEpoch
       | newEpochNo <- unEpochNo (Generic.neEpoch newEpoch)
       , newEpochNo > 0
-      , isCons || (newEpochNo `mod` 10 == 0) || newEpochNo >= 530 ->
+      , -- Snapshot every epoch when near tip, every 10 epochs when lagging, or always for epoch >= threshold
+        (isCons && syncState == SyncFollowing) || (newEpochNo `mod` 10 == 0) || newEpochNo >= leSnapshotNearTipEpoch env ->
           do
             -- TODO: Instead of newEpochNo - 1, is there any way to get the epochNo from 'lssOldState'?
             liftIO $ saveCleanupState env oldState (Just $ EpochNo $ newEpochNo - 1)
             pure True
-    _ ->
-      if timeToSnapshot syncState blkNo && isCons
-        then do
-          liftIO $ saveCleanupState env oldState Nothing
-          pure True
-        else pure False
-  where
-    timeToSnapshot :: SyncState -> BlockNo -> Bool
-    timeToSnapshot syncSt bNo =
-      case (syncSt, unBlockNo bNo) of
-        (SyncFollowing, bno) -> bno `mod` leSnapshotEveryFollowing env == 0
-        (SyncLagging, _) -> False
+    _ -> pure False
 
 saveCurrentLedgerState :: HasLedgerEnv -> CardanoLedgerState -> Maybe EpochNo -> IO ()
 saveCurrentLedgerState env lState mEpochNo = do
@@ -403,7 +391,8 @@ ledgerStateWriteLoop tracer swQueue codecConfig =
 
 mkLedgerStateFilename :: LedgerStateDir -> ExtLedgerState CardanoBlock -> Maybe EpochNo -> WithOrigin FilePath
 mkLedgerStateFilename dir ledger mEpochNo =
-  lsfFilePath . dbPointToFileName dir mEpochNo
+  lsfFilePath
+    . dbPointToFileName dir mEpochNo
     <$> getPoint (ledgerTipPoint @CardanoBlock (ledgerState ledger))
 
 saveCleanupState :: HasLedgerEnv -> CardanoLedgerState -> Maybe EpochNo -> IO ()

cardano-db-sync/src/Cardano/DbSync/Ledger/Types.hs

@@ -67,8 +67,7 @@ data HasLedgerEnv = HasLedgerEnv
   , leNetwork :: !Ledger.Network
   , leSystemStart :: !SystemStart
   , leAbortOnPanic :: !Bool
-  , leSnapshotEveryFollowing :: !Word64
-  , leSnapshotEveryLagging :: !Word64
+  , leSnapshotNearTipEpoch :: !Word64
   , leInterpreter :: !(StrictTVar IO (Strict.Maybe CardanoInterpreter))
   , leStateVar :: !(StrictTVar IO (Strict.Maybe LedgerDB))
   , leStateWriteQueue :: !(TBQueue (FilePath, CardanoLedgerState))

cardano-db-sync/test/Cardano/DbSync/Gen.hs

@@ -58,6 +58,12 @@ syncPreConfig =
     <*> Gen.int (Range.linear 0 10000)
     <*> syncInsertConfig
     <*> Gen.list (Range.linear 0 10) (Gen.text (Range.linear 0 100) Gen.unicode)
+    <*> snapshotIntervalConfig
+
+snapshotIntervalConfig :: Gen SnapshotIntervalConfig
+snapshotIntervalConfig =
+  SnapshotIntervalConfig
+    <$> Gen.word64 (Range.linear 100 1000)
 
 syncNodeParams :: MonadGen m => m SyncNodeParams
 syncNodeParams =
@@ -71,8 +77,6 @@ syncNodeParams =
     <*> Gen.bool
     <*> Gen.bool
     <*> Gen.bool
-    <*> Gen.word64 (Range.linear 0 1000)
-    <*> Gen.word64 (Range.linear 0 1000)
     <*> pure Nothing
 
 syncNodeConfig :: Logging.Configuration -> Gen SyncNodeConfig
@@ -104,6 +108,7 @@ syncNodeConfig loggingCfg =
     <*> triggerHardFork
     <*> syncInsertOptions
     <*> pure []
+    <*> snapshotIntervalConfig
 
 syncInsertConfig :: Gen SyncInsertConfig
 syncInsertConfig =

doc/configuration.md

@@ -66,6 +66,7 @@ Below is a sample `insert_options` section that shows all the defaults:
 | [pool\_stat](#pool-stat)                     | `enum`     | Optional |
 | [remove\_jsonb_from_schema](#remove-jsonb-from-schema) | `enum`     | Optional |
 | [stop\_at\_block](#stop-at-block)            | `integer`  | Optional |
+| [snapshot\_interval](#snapshot-interval)     | `object`   | Optional |
 
 ## Preset
 
@@ -602,3 +603,43 @@ Stops db-sync after processing the specified block number. Useful for testing an
 }
 ```
 
+## Snapshot Interval
+
+`snapshot_interval`
+
+* Type: `object`
+* Optional: When not specified, uses default values
+
+Controls how frequently ledger state snapshots are taken during sync. Taking snapshots less frequently during initial sync can significantly improve sync performance by reducing IOPS and disk throughput consumption.
+
+Snapshot Interval Properties:
+
+| Property                         | Type      | Required | Default |
+| :------------------------------- | :-------- | :------- | :------ |
+| [near\_tip\_epoch](#near-tip-epoch) | `integer` | Optional | 580     |
+
+### Near Tip Epoch
+
+`snapshot_interval.near_tip_epoch`
+
+* Type: `integer`
+* Default: `580`
+
+Epoch threshold used to determine snapshot behavior. When syncing reaches this epoch or later, db-sync is considered to be approaching or at the current tip of the chain. Combined with time-based detection (within 10 days of current time), this ensures snapshots are taken every epoch when near the tip for fast rollback recovery. During earlier epochs or when syncing behind, snapshots are taken every 10 epochs.
+
+### Example
+
+```json
+{
+  "snapshot_interval": {
+    "near_tip_epoch": 580
+  }
+}
+```
+
+### Performance Considerations
+
+- **Lower `near_tip_epoch` value**: Start taking frequent epoch-based snapshots earlier in the chain history
+- **Higher `near_tip_epoch` value**: Delay frequent snapshots until later in the chain, improving sync speed for longer. During initial sync (before reaching `near_tip_epoch`), snapshots are taken every 10 epochs
+- **Near tip detection**: Automatically switches to epoch-based snapshots when within 10 days of current time, regardless of epoch number
+