diff --git a/.gitignore b/.gitignore
index 0479bb49..48ee1944 100644
--- a/.gitignore
+++ b/.gitignore
@@ -11,3 +11,4 @@ cabal.project.local
 
 .backup/
 .hbs2-git/
+bin/
diff --git a/hbs2-peer/app/RefChan.hs b/hbs2-peer/app/RefChan.hs
index 05f4adb2..769f5340 100644
--- a/hbs2-peer/app/RefChan.hs
+++ b/hbs2-peer/app/RefChan.hs
@@ -656,8 +656,8 @@ refChanWorker env brains = do
 
         let byChan = HashMap.fromListWith (<>) [ (x, [y]) | (x,y) <- catMaybes trans ]
 
-        -- FIXME: process-in-parallel
-        forM_ (HashMap.toList byChan) $ \(c,new) -> do
+        -- FIXME: thread-num-hardcode-to-remove
+        pooledForConcurrentlyN_ 4 (HashMap.toList byChan) $ \(c,new) -> do
           mbLog <- liftIO $ getRef sto c
 
           hashes <- maybe1 mbLog (pure mempty) $ readLog (getBlock sto) . HashRef
diff --git a/hbs2-tests/hbs2-tests.cabal b/hbs2-tests/hbs2-tests.cabal
index 6c6c7722..e6ba7a3e 100644
--- a/hbs2-tests/hbs2-tests.cabal
+++ b/hbs2-tests/hbs2-tests.cabal
@@ -995,3 +995,41 @@ executable test-pipe-mess
 
 
 
+executable test-merge-limits
+    import:           shared-properties
+    default-language: Haskell2010
+
+    -- other-extensions:
+
+    hs-source-dirs:   test
+    main-is:          TestMergeLimits.hs
+    build-depends:
+          base, hbs2-core, hbs2-storage-simple
+        , async
+        , bytestring
+        , cache
+        , containers
+        , hashable
+        , microlens-platform
+        , mtl
+        , prettyprinter
+        , QuickCheck
+        , quickcheck-instances
+        , random
+        , safe
+        , serialise
+        , stm
+        , streaming
+        , tasty
+        , tasty-quickcheck
+        , tasty-hunit
+        , tasty-quickcheck
+        , transformers
+        , uniplate
+        , vector
+        , filepath
+        , temporary
+        , unliftio
+        , unordered-containers
+        , unix
+        , timeit
diff --git a/hbs2-tests/test/TestMergeLimits.hs b/hbs2-tests/test/TestMergeLimits.hs
new file mode 100644
index 00000000..4b8eae60
--- /dev/null
+++ b/hbs2-tests/test/TestMergeLimits.hs
@@ -0,0 +1,67 @@
+{-# Language NumericUnderscores #-}
+module Main where
+
+import HBS2.Prelude.Plated
+
+import HBS2.Hash
+import HBS2.Data.Types.Refs
+
+import Data.ByteString.Lazy qualified as LBS
+import Data.Function
+import Streaming.Prelude qualified as S
+import System.TimeIt
+import Data.HashSet qualified as HS
+import Data.HashSet (HashSet)
+import Data.List qualified as List
+import UnliftIO
+import System.Random
+import Data.Word
+import Control.Monad
+
+rndHash :: IO HashRef
+rndHash = do
+  w1 <- replicateM 4 $ randomIO @Word64
+  pure $ HashRef $ hashObject @HbSync (serialise w1)
+
+main :: IO ()
+main = do
+  rnd <- openFile "/dev/random" ReadMode
+
+  lbs <- LBS.hGetNonBlocking rnd $ 32_000_000 * 10
+
+  hashList <- S.toList_ do
+    flip fix lbs $ \next rest -> do
+      let (a,rest') = LBS.splitAt 32 rest
+      S.yield $ HashRef $! HbSyncHash (LBS.toStrict a)
+      unless (LBS.null rest') $ next rest'
+
+  chunks <- S.toList_ do
+    flip fix hashList $ \next rest -> do
+      let (c, rest') = List.splitAt 1_000_000 rest
+      S.yield c
+      unless (List.null rest') $ next rest'
+
+  pieces <- forConcurrently chunks (pure . HS.fromList)
+
+  hs <- timeItNamed "rebuild index" do
+    let hashSet = HS.unions pieces
+    print $ length hashSet
+    pure hashSet
+
+  void $ timeItNamed "calculate hash" do
+    let bs = serialise hs
+    let hx = hashObject @HbSync bs
+    print $ pretty hx
+
+  putStrLn "now we have partially sorted index"
+
+  hashes <- replicateM 100 rndHash
+
+  timeItNamed "add new items" do
+    let hs2 = HS.union hs (HS.fromList hashes)
+    -- let hx = hashObject @HbSync (serialise hs2)
+    print $ pretty (HS.size hs2) -- <+> pretty hx
+
+  pure ()
+
+