Improve cache hits for tuple keys in key_split and intern results (d…

…ask#10547)
rjzamora · Oct 6, 2023 · 928a95a · 928a95a
1 parent 62dbbff
commit 928a95a
Showing 1 changed file with 4 additions and 3 deletions.
diff --git 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"