{-# LANGUAGE CPP #-} module Main (main) where import qualified Data.Aeson as Aeson import qualified Data.ByteString as BS import qualified Data.ByteString.Lazy as LBS import qualified Hasql.Connection as C import qualified Hasql.Notifications as N import qualified Hasql.Pool as P import qualified Hasql.Session as S import qualified Hasql.Transaction.Sessions as HT import Control.AutoUpdate (defaultUpdateSettings, mkAutoUpdate, updateAction) import Control.Debounce (debounceAction, debounceEdge, debounceFreq, defaultDebounceSettings, mkDebounce, trailingEdge) import Control.Retry (RetryStatus, capDelay, exponentialBackoff, retrying, rsPreviousDelay) import Data.IORef (IORef, atomicWriteIORef, newIORef, readIORef) import Data.String (IsString (..)) import Data.Text.IO (hPutStrLn) import Data.Time.Clock (getCurrentTime) import Network.Wai.Handler.Warp (defaultSettings, runSettings, setHost, setPort, setServerName) import System.CPUTime (getCPUTime) import System.IO (BufferMode (..), hSetBuffering) import Text.Printf (hPrintf) import PostgREST.App (postgrest) import PostgREST.Config import PostgREST.DbStructure (getDbStructure, getPgVersion) import PostgREST.Error (PgError (PgError), checkIsFatal, errorPayload) import PostgREST.Types (ConnectionStatus (..), DbStructure, PgVersion (..), SCacheStatus (..), minimumPgVersion) import Protolude hiding (hPutStrLn, head, toS) import Protolude.Conv (toS) #ifndef mingw32_HOST_OS import System.Posix.Signals import UnixSocket #endif -- | This is where everything starts. main :: IO () main = do -- -- LineBuffering: the entire output buffer is flushed whenever a newline is -- output, the buffer overflows, a hFlush is issued or the handle is closed -- -- NoBuffering: output is written immediately and never stored in the buffer hSetBuffering stdout LineBuffering hSetBuffering stdin LineBuffering hSetBuffering stderr NoBuffering -- read PGRST_ env variables env <- readEnvironment -- read path from commad line opts <- readCLIShowHelp env -- build the 'AppConfig' from the config file path conf <- readValidateConfig env $ cliPath opts -- These are config values that can't be reloaded at runtime. Reloading some of them would imply restarting the web server. let host = configServerHost conf port = configServerPort conf maybeSocketAddr = configServerUnixSocket conf #ifndef mingw32_HOST_OS socketFileMode = configServerUnixSocketMode conf #endif dbUri = toS (configDbUri conf) (dbChannelEnabled, dbChannel) = (configDbChannelEnabled conf, toS $ configDbChannel conf) serverSettings = setHost ((fromString . toS) host) -- Warp settings . setPort port . setServerName (toS $ "postgrest/" <> prettyVersion) $ defaultSettings poolSize = configDbPoolSize conf poolTimeout = configDbPoolTimeout' conf logLevel = configLogLevel conf case cliCommand opts of CmdDumpConfig -> do putStr $ dumpAppConfig conf exitSuccess CmdDumpSchema -> do dumpedSchema <- dumpSchema conf putStrLn dumpedSchema exitSuccess CmdRun -> pass -- create connection pool with the provided settings, returns either a 'Connection' or a 'ConnectionError'. Does not throw. pool <- P.acquire (poolSize, poolTimeout, dbUri) -- Used to sync the listener(NOTIFY reload) with the connectionWorker. No connection for the listener at first. Only used if dbChannelEnabled=true. mvarConnectionStatus <- newEmptyMVar -- No schema cache at the start. Will be filled in by the connectionWorker refDbStructure <- newIORef Nothing -- Helper ref to make sure just one connectionWorker can run at a time refIsWorkerOn <- newIORef False -- Config that can change at runtime refConf <- newIORef conf -- This is passed to the connectionWorker method so it can kill the main thread if the PostgreSQL's version is not supported. mainTid <- myThreadId let connWorker = connectionWorker mainTid pool refConf refDbStructure refIsWorkerOn (dbChannelEnabled, mvarConnectionStatus) -- Sets the initial refDbStructure connWorker #ifndef mingw32_HOST_OS -- Only for systems with signals: -- -- releases the connection pool whenever the program is terminated, -- see https://github.com/PostgREST/postgrest/issues/268 forM_ [sigINT, sigTERM] $ \sig -> void $ installHandler sig (Catch $ do P.release pool throwTo mainTid UserInterrupt ) Nothing -- The SIGUSR1 signal updates the internal 'DbStructure' by running 'connectionWorker' exactly as before. void $ installHandler sigUSR1 ( Catch connWorker ) Nothing -- Re-read the config on SIGUSR2, but only if we have a config file when (isJust $ cliPath opts) $ void $ installHandler sigUSR2 ( Catch $ reReadConfig env (cliPath opts) refConf ) Nothing #endif -- reload schema cache on NOTIFY when dbChannelEnabled $ listener dbUri dbChannel pool refConf refDbStructure mvarConnectionStatus connWorker -- ask for the OS time at most once per second getTime <- mkAutoUpdate defaultUpdateSettings {updateAction = getCurrentTime} let postgrestApplication = postgrest logLevel refConf refDbStructure pool getTime connWorker #ifndef mingw32_HOST_OS -- run the postgrest application with user defined socket. Only for UNIX systems. whenJust maybeSocketAddr $ runAppInSocket serverSettings postgrestApplication socketFileMode #endif -- run the postgrest application whenNothing maybeSocketAddr $ do putStrLn $ ("Listening on port " :: Text) <> show port runSettings serverSettings postgrestApplication -- Time constants _32s :: Int _32s = 32000000 :: Int -- 32 seconds _1s :: Int _1s = 1000000 :: Int -- 1 second {-| The purpose of this worker is to obtain a healthy connection to pg and an up-to-date schema cache(DbStructure). This method is meant to be called by multiple times by the same thread, but does nothing if the previous invocation has not terminated. In all cases this method does not halt the calling thread, the work is preformed in a separate thread. Note: 'atomicWriteIORef' is essentially a lazy semaphore that prevents two threads from running 'connectionWorker' at the same time. Background thread that does the following : 1. Tries to connect to pg server and will keep trying until success. 2. Checks if the pg version is supported and if it's not it kills the main program. 3. Obtains the dbStructure. If this fails, it goes back to 1. -} connectionWorker :: ThreadId -- ^ Main thread id. Killed if pg version is unsupported -> P.Pool -- ^ The pg connection pool -> IORef AppConfig -- ^ mutable reference to AppConfig -> IORef (Maybe DbStructure) -- ^ mutable reference to 'DbStructure' -> IORef Bool -- ^ Used as a binary Semaphore -> (Bool, MVar ConnectionStatus) -- ^ For interacting with the LISTEN channel -> IO () connectionWorker mainTid pool refConf refDbStructure refIsWorkerOn (dbChannelEnabled, mvarConnectionStatus) = do isWorkerOn <- readIORef refIsWorkerOn unless isWorkerOn $ do -- Prevents multiple workers to be running at the same time. Could happen on too many SIGUSR1s. atomicWriteIORef refIsWorkerOn True void $ forkIO work where work = do putStrLn ("Attempting to connect to the database..." :: Text) connected <- connectionStatus pool when dbChannelEnabled $ void $ tryPutMVar mvarConnectionStatus connected -- tryPutMVar doesn't lock the thread. It should always succeed since the worker is the only mvar producer. case connected of FatalConnectionError reason -> hPutStrLn stderr reason >> killThread mainTid -- Fatal error when connecting NotConnected -> return () -- Unreachable because connectionStatus will keep trying to connect Connected actualPgVersion -> do -- Procede with initialization putStrLn ("Connection successful" :: Text) scStatus <- loadSchemaCache pool actualPgVersion refConf refDbStructure case scStatus of SCLoaded -> pure () -- do nothing and proceed if the load was successful SCOnRetry -> work -- retry SCFatalFail -> killThread mainTid -- die if our schema cache query has an error liftIO $ atomicWriteIORef refIsWorkerOn False {-| Check if a connection from the pool allows access to the PostgreSQL database. If not, the pool connections are released and a new connection is tried. Releasing the pool is key for rapid recovery. Otherwise, the pool timeout would have to be reached for new healthy connections to be acquired. Which might not happen if the server is busy with requests. No idle connection, no pool timeout. The connection tries are capped, but if the connection times out no error is thrown, just 'False' is returned. -} connectionStatus :: P.Pool -> IO ConnectionStatus connectionStatus pool = retrying (capDelay _32s $ exponentialBackoff _1s) shouldRetry (const $ P.release pool >> getConnectionStatus) where getConnectionStatus :: IO ConnectionStatus getConnectionStatus = do pgVersion <- P.use pool getPgVersion case pgVersion of Left e -> do let err = PgError False e hPutStrLn stderr . toS $ errorPayload err case checkIsFatal err of Just reason -> return $ FatalConnectionError reason Nothing -> return NotConnected Right version -> if version < minimumPgVersion then return . FatalConnectionError $ "Cannot run in this PostgreSQL version, PostgREST needs at least " <> pgvName minimumPgVersion else return . Connected $ version shouldRetry :: RetryStatus -> ConnectionStatus -> IO Bool shouldRetry rs isConnSucc = do let delay = fromMaybe 0 (rsPreviousDelay rs) `div` _1s itShould = NotConnected == isConnSucc when itShould $ putStrLn $ "Attempting to reconnect to the database in " <> (show delay::Text) <> " seconds..." return itShould -- | Load the DbStructure by using a connection from the pool. loadSchemaCache :: P.Pool -> PgVersion -> IORef AppConfig -> IORef (Maybe DbStructure) -> IO SCacheStatus loadSchemaCache pool actualPgVersion refConf refDbStructure = do conf <- readIORef refConf result <- P.use pool $ HT.transaction HT.ReadCommitted HT.Read $ getDbStructure (toList $ configDbSchemas conf) (configDbExtraSearchPath conf) actualPgVersion (configDbPreparedStatements conf) case result of Left e -> do let err = PgError False e putErr = hPutStrLn stderr . toS . errorPayload $ err case checkIsFatal err of Just _ -> do hPutStrLn stderr ("A fatal error ocurred when loading the schema cache" :: Text) putErr hPutStrLn stderr ("This is probably a bug in PostgREST, please report it at https://github.com/PostgREST/postgrest/issues" :: Text) return SCFatalFail Nothing -> do hPutStrLn stderr ("An error ocurred when loading the schema cache" :: Text) >> putErr return SCOnRetry Right dbStructure -> do atomicWriteIORef refDbStructure $ Just dbStructure putStrLn ("Schema cache loaded" :: Text) return SCLoaded {-| Starts a dedicated pg connection to LISTEN for notifications. When a NOTIFY - with an empty payload - is done, it refills the schema cache. It uses the connectionWorker in case the LISTEN connection dies. -} listener :: ByteString -> Text -> P.Pool -> IORef AppConfig -> IORef (Maybe DbStructure) -> MVar ConnectionStatus -> IO () -> IO () listener dbUri dbChannel pool refConf refDbStructure mvarConnectionStatus connWorker = start where start = do connStatus <- takeMVar mvarConnectionStatus -- takeMVar makes the thread wait if the MVar is empty(until there's a connection). case connStatus of Connected actualPgVersion -> void $ forkFinally (do -- forkFinally allows to detect if the thread dies dbOrError <- C.acquire dbUri -- Debounce in case too many NOTIFYs arrive. Could happen on a migration(assuming a pg EVENT TRIGGER is set up). -- This might not be needed according to pg docs https://www.postgresql.org/docs/12/sql-notify.html: -- "If the same channel name is signaled multiple times from the same transaction with identical payload strings, the database server can decide to deliver a single notification only." -- But we do it to be extra safe. scFiller <- mkDebounce (defaultDebounceSettings { -- It's not necessary to check the loadSchemaCache success here. If the connection drops, the thread will die and proceed to recover below. debounceAction = void $ loadSchemaCache pool actualPgVersion refConf refDbStructure, debounceEdge = trailingEdge, -- wait until the function hasn’t been called in _1s debounceFreq = _1s }) case dbOrError of Right db -> do putStrLn $ "Listening for notifications on the " <> dbChannel <> " channel" let channelToListen = N.toPgIdentifier dbChannel N.listen db channelToListen N.waitForNotifications (\_ msg -> if BS.null msg then scFiller -- reload the schema cache else pure ()) db -- Do nothing if anything else than an empty message is sent _ -> die errorMessage) (\_ -> do -- if the thread dies, we try to recover putStrLn retryMessage connWorker -- assume the pool connection was also lost, call the connection worker start) -- retry the listener _ -> putStrLn errorMessage -- Should be unreachable. connectionStatus will retry until there's a connection. errorMessage = "Could not listen for notifications on the " <> dbChannel <> " channel" :: Text retryMessage = "Retrying listening for notifications on the " <> dbChannel <> " channel.." :: Text #ifndef mingw32_HOST_OS -- | Re-reads the config at runtime. Invoked on SIGUSR2. -- | If it panics(config path was changed, invalid setting), it'll show an error but won't kill the main thread. reReadConfig :: Environment -> Maybe FilePath -> IORef AppConfig -> IO () reReadConfig env path refConf = do conf <- readValidateConfig env path atomicWriteIORef refConf conf putStrLn ("Config file reloaded" :: Text) #endif -- | Dump DbStructure schema to JSON dumpSchema :: AppConfig -> IO LBS.ByteString dumpSchema conf = do Right conn <- C.acquire . toS $ configDbUri conf Right pgVersion <- S.run getPgVersion conn let getDbStructureTransaction = HT.transaction HT.ReadCommitted HT.Read $ getDbStructure (toList $ configDbSchemas conf) (configDbExtraSearchPath conf) pgVersion (configDbPreparedStatements conf) result <- timeToStderr "Loaded schema in %.3f seconds" $ S.run getDbStructureTransaction conn C.release conn case result of Left e -> do hPutStrLn stderr $ "An error ocurred when loading the schema cache:\n" <> show e exitFailure Right dbStructure -> return $ Aeson.encode dbStructure -- | Print the time taken to run an IO action to stderr with the given printf string timeToStderr :: [Char] -> IO (Either a b) -> IO (Either a b) timeToStderr fmtString a = do start <- getCPUTime result <- a end <- getCPUTime let duration :: Double duration = fromIntegral (end - start) / picoseconds when (isRight result) $ hPrintf stderr (fmtString ++ "\n") duration return result -- | 10^12 picoseconds per second picoseconds :: Double picoseconds = 1000000000000 -- Utility functions. #ifndef mingw32_HOST_OS whenJust :: Applicative f => Maybe a -> (a -> f ()) -> f () whenJust (Just x) f = f x whenJust Nothing _ = pass #endif whenNothing :: Applicative f => Maybe a -> f () -> f () whenNothing Nothing f = f whenNothing _ _ = pass