Remove telemetry batching (#4644)

This was initially done as an optimization but now that we’ve used
telemetry for a while this hasn’t really been necessary and we’ve
spend way more time on people being confused by the fact that
telemetry messages are only flushed on shutdown.

Furthermore, I’m also slighlty worried that flushing the telemetry
messages on shutdown might go wrong on Windows given that everything
there seems to insist on hard-killing proesses.

changelog_begin
changelog_end

libs-haskell/da-hs-base/src/DA/Service/Logger/Impl/GCP.hs

@@ -112,12 +112,6 @@ data SentData = SentData
 
 -- Parameters
 
--- | Number of messages that need to end up in the log before we send
--- them as a batch. This is done to avoid sending a ton of small
--- messages.
-batchSize :: Int
-batchSize = 10
-
 -- | Timeout on log requests
 requestTimeout :: Int
 requestTimeout = 5_000_000
@@ -174,9 +168,6 @@ initialiseGcpState GCPConfig{..} gcpFallbackLogger = do
     let gcpTag = gcpConfigTag
     pure GCPState {..}
 
-readBatch :: TChan a -> Int -> IO [a]
-readBatch chan n = atomically $ replicateM n (readTChan chan)
-
 -- | Read everything from the chan that you can within a single transaction.
 drainChan :: TChan a -> IO [a]
 drainChan chan = atomically $ unfoldM (tryReadTChan chan)
@@ -200,8 +191,8 @@ withGcpLogger config p hnd f = do
             }
     let worker = forever $ mask_ $ do
             -- We mask to avoid messages getting lost.
-            entries <- readBatch (gcpLogChan gcpState) batchSize
-            sendLogs gcpState entries
+            entry <- atomically (readTChan (gcpLogChan gcpState))
+            sendLogs gcpState [entry]
     (withAsync worker $ \_ -> do
         f gcpState logger) `finally` do
         logs <- drainChan (gcpLogChan gcpState)