[flow] add chunk size to toml settings

Also, implement pure `mapConcurrently` in extract ngrams, as it's
closely related to chunk size.

bin/gargantext-cli/CLI/Ini.hs

@@ -88,7 +88,8 @@ convertConfigs ini@(Ini.GargConfig { .. }) iniMail nlpConfig connInfo =
                                                   , _wsLongJobTimeout = 3000
                                                   , _wsDefaultDelay = 0
                                                   , _wsAdditionalDelayAfterRead = 5
-                                                  , _wsDatabase = connInfo { PGS.connectDatabase = "pgmq"} }
+                                                  , _wsDatabase = connInfo { PGS.connectDatabase = "pgmq"}
+                                                  , _wsNlpConduitChunkSize = 10 }
                     , _gc_logging = Config.LogConfig {
                         _lc_log_level = INFO
                       , _lc_log_file = Nothing

gargantext-settings.toml_toModify

@@ -166,6 +166,10 @@ default_job_timeout = 60
 # default timeout for "long" jobs (in seconds)
 long_job_timeout = 3000
 
+# Batch size when sending data to NLP.
+# Preferably, set as much as the number of CPUs
+nlp_conduit_chunk_size = 10
+
 # if you leave the same credentials as in [database] section above,
 #  workers will try to set up the `gargantext_pgmq` database
 #  automatically

gargantext.cabal

@@ -510,8 +510,6 @@ library
     , ansi-terminal
     , array ^>= 0.5.4.0
     , async ^>= 2.2.4
-    , lifted-async
-    , lifted-base
     , attoparsec ^>= 0.14.4
     , base64-bytestring ^>= 1.2.1.0
     , bimap >= 0.5.0
@@ -572,6 +570,7 @@ library
     , json-stream ^>= 0.4.2.4
     , lens >= 5.2.2 && < 5.3
     , lens-aeson < 1.3
+    , lifted-async >= 0.10 && < 0.12
     , list-zipper
     , massiv < 1.1
     , matrix ^>= 0.3.6.1

src/Gargantext/Core/Config/Ini/NLP.hs

@@ -9,19 +9,12 @@ Portability : POSIX
 
 -}
 
-{-# LANGUAGE TemplateHaskell #-}
-
 module Gargantext.Core.Config.Ini.NLP (
   -- * Types
   NLPConfig(..)
 
   -- * Utility functions
   , readConfig
-
-  -- * Lenses
-  , nlp_default
-  , nlp_languages
-
   )
   where
 
@@ -59,4 +52,3 @@ readConfig fp = do
                                      , T.pack $ show m_nlp_other ]
     Just ret -> pure ret
 
-makeLenses ''NLPConfig

src/Gargantext/Core/Config/NLP.hs

@@ -19,7 +19,6 @@ module Gargantext.Core.Config.NLP (
   -- * Lenses
   , nlp_default
   , nlp_languages
-
   )
   where
 
@@ -48,9 +47,9 @@ data NLPConfig = NLPConfig { _nlp_default   :: URI
 instance FromValue NLPConfig where
   fromValue v = do
     _nlp_default <- parseTableFromValue (reqKey "EN") v
-    -- _nlp_languages <- fromValue <$> getTable
     MkTable t <- parseTableFromValue getTable v
     _nlp_languages <- mapM fromValue (snd <$> t)
+    
     return $ NLPConfig { .. }
 instance ToValue NLPConfig where
   toValue = defaultTableToValue
@@ -58,7 +57,7 @@ instance ToTable NLPConfig where
   toTable (NLPConfig { .. }) =
     table ([ k .= v | (k, v) <- Map.toList _nlp_languages ]
            -- output the default "EN" language as well
-           <> [ ("EN" :: Text) .= _nlp_default ])
+           <> [ ("EN" :: Text) .= _nlp_default ] )
 
 
 -- readConfig :: SettingsFile -> IO NLPConfig

src/Gargantext/Core/Config/Worker.hs

@@ -53,6 +53,8 @@ data WorkerSettings =
     , _wsDefaultDelay :: B.TimeoutS
     , _wsAdditionalDelayAfterRead :: B.TimeoutS
     , _wsDefinitions  :: ![WorkerDefinition]
+
+    , _wsNlpConduitChunkSize :: Int
   } deriving (Show, Eq)
 instance FromValue WorkerSettings where
   fromValue = parseTableFromValue $ do
@@ -61,6 +63,7 @@ instance FromValue WorkerSettings where
     _wsDefaultVisibilityTimeout <- reqKey "default_visibility_timeout"
     _wsDefaultJobTimeout <- reqKey "default_job_timeout"
     _wsLongJobTimeout <- reqKey "long_job_timeout"
+    _wsNlpConduitChunkSize <- reqKey "nlp_conduit_chunk_size"
     defaultDelay <- reqKey "default_delay"
     additionalDelayAfterRead <- reqKey "additional_delay_after_read"
     return $ WorkerSettings { _wsDatabase = unTOMLConnectInfo dbConfig
@@ -69,7 +72,8 @@ instance FromValue WorkerSettings where
                             , _wsDefinitions
                             , _wsDefaultVisibilityTimeout
                             , _wsDefaultDelay = B.TimeoutS defaultDelay
-                            , _wsAdditionalDelayAfterRead = B.TimeoutS additionalDelayAfterRead }
+                            , _wsAdditionalDelayAfterRead = B.TimeoutS additionalDelayAfterRead
+                            , _wsNlpConduitChunkSize }
 instance ToValue WorkerSettings where
   toValue = defaultTableToValue
 instance ToTable WorkerSettings where
@@ -80,7 +84,8 @@ instance ToTable WorkerSettings where
           , "default_visibility_timeout" .= _wsDefaultVisibilityTimeout
           , "default_delay" .= B._TimeoutS _wsDefaultDelay
           , "additional_delay_after_read" .= B._TimeoutS _wsAdditionalDelayAfterRead
-          , "definitions" .= _wsDefinitions ]
+          , "definitions" .= _wsDefinitions
+          , "nlp_conduit_chunk_size" .= _wsNlpConduitChunkSize ]
 
 data WorkerDefinition =
   WorkerDefinition {

src/Gargantext/Database/Action/Flow.hs

@@ -56,8 +56,6 @@ module Gargantext.Database.Action.Flow -- (flowDatabase, ngrams2list)
 
 import Conduit
 import Control.Concurrent.Async.Lifted qualified as AsyncL
-import Control.Concurrent.QSem.Lifted qualified as QSemL
-import Control.Exception.Lifted qualified as CEL
 import Control.Exception.Safe qualified as CES
 import Control.Lens ( to, view )
 import Control.Exception.Safe (catch, MonadCatch)
@@ -73,7 +71,8 @@ import Data.Text qualified as T
 import Gargantext.API.Ngrams.Tools (getTermsWith)
 import Gargantext.API.Ngrams.Types (NgramsTerm)
 import Gargantext.Core (Lang(..), withDefaultLanguage, NLPServerConfig)
-import Gargantext.Core.Config (GargConfig(..), hasConfig)
+import Gargantext.Core.Config (GargConfig(..), hasConfig, gc_worker)
+import Gargantext.Core.Config.Worker (wsNlpConduitChunkSize)
 import Gargantext.Core.Config.Types (APIsConfig(..))
 import Gargantext.Core.Ext.IMTUser (readFile_Annuaire)
 import Gargantext.Core.NLP (HasNLPServer, nlpServerGet)
@@ -101,7 +100,8 @@ import Gargantext.Database.Admin.Types.Hyperdata.Contact ( HyperdataContact )
 import Gargantext.Database.Admin.Types.Hyperdata.Corpus ( HyperdataAnnuaire, HyperdataCorpus(_hc_lang) )
 import Gargantext.Database.Admin.Types.Hyperdata.Document ( ToHyperdataDocument(toHyperdataDocument), HyperdataDocument )
 import Gargantext.Database.Admin.Types.Node hiding (ERROR, DEBUG) -- (HyperdataDocument(..), NodeType(..), NodeId, UserId, ListId, CorpusId, RootId, MasterCorpusId, MasterUserId)
-import Gargantext.Database.Prelude
+import Gargantext.Database.Class ( DBCmdWithEnv, IsDBCmd )
+import Gargantext.Database.Transactional ( DBUpdate, runDBTx )
 import Gargantext.Database.Query.Table.ContextNodeNgrams2 ( ContextNodeNgrams2Poly(..), insertContextNodeNgrams2 )
 import Gargantext.Database.Query.Table.Node ( MkCorpus, insertDefaultNodeIfNotExists, getOrMkList, getNodeWith )
 import Gargantext.Database.Query.Table.Node.Document.Add qualified as Doc (add)
@@ -111,8 +111,8 @@ import Gargantext.Database.Query.Table.NodeContext (selectDocNodesOnlyId)
 import Gargantext.Database.Query.Table.NodeNgrams (listInsertDb , getCgramsId)
 import Gargantext.Database.Query.Tree.Root (MkCorpusUser(..), getOrMkRoot, getOrMkRootWithCorpus, userFromMkCorpusUser)
 import Gargantext.Database.Schema.Ngrams ( indexNgrams, NgramsId )
-import Gargantext.Database.Schema.Node
-import Gargantext.Database.Types
+import Gargantext.Database.Schema.Node ( NodePoly(_node_id, _node_hash_id), node_hyperdata )
+import Gargantext.Database.Types ( Indexed(Indexed) )
 import Gargantext.Prelude hiding (catch, onException, to)
 import Gargantext.System.Logging ( logLocM, LogLevel(DEBUG, ERROR), MonadLogger )
 import Gargantext.Utils.Jobs.Monad ( JobHandle, MonadJobStatus(..), markFailureNoErr )
@@ -299,11 +299,12 @@ flow :: forall env err m a c.
         -> m CorpusId
 flow c mkCorpusUser la mfslw (count, docsC) jobHandle = do
   cfg <- view hasConfig
+  let chunkSize = cfg ^. gc_worker . wsNlpConduitChunkSize
   (_userId, userCorpusId, listId, msgs) <- runDBTx $ createNodes cfg mkCorpusUser c
   forM_ msgs ce_notify
   -- TODO if public insertMasterDocs else insertUserDocs
   runConduit (zipSources (yieldMany ([1..] :: [Int])) docsC
-    .| CList.chunksOf 10
+    .| CList.chunksOf chunkSize
     .| mapM_C (addDocumentsWithProgress userCorpusId)
     .| sinkNull) `CES.catches`
       [ CES.Handler $ \(e :: ClientError) -> do
@@ -548,11 +549,12 @@ extractNgramsFromDocuments :: forall doc env err m.
                            -> [doc]
                            -> m (UncommittedNgrams doc)
 extractNgramsFromDocuments nlpServer lang docs = do
-  sem <- QSemL.newQSem 10
-  let f = extractNgramsFromDocument nlpServer lang
-  ret <- AsyncL.mapConcurrently (\doc ->
-                                   CEL.bracket_ (QSemL.waitQSem sem) (QSemL.signalQSem sem) (f doc)
-                                ) docs
+  ret <- AsyncL.mapConcurrently (extractNgramsFromDocument nlpServer lang) docs
+  -- sem <- QSemL.newQSem 10
+  -- let f = extractNgramsFromDocument nlpServer lang
+  -- ret <- AsyncL.mapConcurrently (\doc ->
+  --                                  CEL.bracket_ (QSemL.waitQSem sem) (QSemL.signalQSem sem) (f doc)
+  --                               ) docs
   pure $ foldl (<>) mempty ret
   -- foldlM go mempty docs
   -- where

test-data/test_config.toml

@@ -99,6 +99,10 @@ default_job_timeout = 60
 # default timeout for "long" jobs (in seconds)
 long_job_timeout = 3000
 
+# Batch size when sending data to NLP.
+# Preferably, set as much as the number of CPUs
+nlp_conduit_chunk_size = 10
+
 # NOTE This is overridden by Test.Database.Setup
 [worker.database]
 host = "127.0.0.1"