From 928a95aa56f60da33a4e724ea2ca97797c612968 Mon Sep 17 00:00:00 2001
From: Florian Jetter <fjetter@users.noreply.github.com>
Date: Fri, 6 Oct 2023 12:48:10 +0200
Subject: [PATCH] Improve cache hits for tuple keys in `key_split` and intern
 results (#10547)

---
 dask/utils.py | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/dask/utils.py b/dask/utils.py
index 6713fb3b86f..e2ebb6a588e 100644
--- a/dask/utils.py
+++ b/dask/utils.py
@@ -1851,10 +1851,11 @@ def key_split(s):
     >>> key_split('_(x)')  # strips unpleasant characters
     'x'
     """
+    # If we convert the key, recurse to utilize LRU cache better
     if type(s) is bytes:
-        s = s.decode()
+        return key_split(s.decode())
     if type(s) is tuple:
-        s = s[0]
+        return key_split(s[0])
     try:
         words = s.split("-")
         if not words[0][0].isalpha():
@@ -1873,7 +1874,7 @@ def key_split(s):
         else:
             if result[0] == "<":
                 result = result.strip("<>").split()[0].split(".")[-1]
-            return result
+            return sys.intern(result)
     except Exception:
         return "Other"