[PATCH] utils: Add NFS EEXISTS/isdir failure workaround

We're seeing cases where we see tracebacks when creating directories in
SSTATE_DIR on an NFS server. The issue is that we see EEXISTS being returned
but isdir() is False, likely when multiple clients try and create the same
directory at the same time.

This is likely a relatively well known issue with NFS's attribute cache.
There is also a way to clear the attribute cache, which is to call
opendir() on the parent directory. That is what this workaround does. In
the rare case we're about to fail, try that using os.listdir() and
see if it helps. Testing showed that it would "fix" several cases
where we'd have had failures otherwise.

Signed-off-by: Richard Purdie <[email protected]>
---
 lib/bb/utils.py | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/lib/bb/utils.py b/lib/bb/utils.py
index 1347c29d020..8b8d381faa5 100644
--- a/lib/bb/utils.py
+++ b/lib/bb/utils.py
@@ -944,8 +944,20 @@ def mkdirhier(directory):
     try:
         os.makedirs(directory)
     except OSError as e:
-        if e.errno != errno.EEXIST or not os.path.isdir(directory):
+        if e.errno != errno.EEXIST:
             raise e
+        if os.path.isdir(directory):
+            return
+        # We can end up here if there is a race between two mkdirs on an NFS mount,
+        # which happens more often with sstate that you'd think. The server returns
+        # EEXIST but the local attribute cache is out of date. It can be refreshed with
+        # an opendir call, so try that (via listdir) and check the directory again
+        # before we really fail.
+        os.listdir(os.path.dirname(directory))
+        if os.path.isdir(directory):
+            return
+        bb.warn("mkdir: %s is not a directory?")
+        raise e
 
 def movefile(src, dest, newmtime = None, sstat = None):
     """Moves a file from ``src`` to ``dest``, preserving all permissions and