X wcfs: Teach start to start successfully even after unclean wcfs shutdown

This patch does to start what b0ca031f did for join/serve.

Stop duplicating the code and factor logic to "attach to wcsrv, or
prepare to start" into _try_attach_wcsrv function.

wcfs/__init__.py

@@ -208,19 +208,10 @@ def join(zurl, autostart=_default_autostart()): # -> WCFS
             return wc
 
         # no. try opening .wcfs - if we succeed - wcfs is already running.
-        unclean = False
-        try:
-            f = open(mntpt + "/.wcfs/zurl")
-        except IOError as e:
-            if   e.errno == ENOENT:     # wcfs cleanly unmounted
-                pass
-            elif e.errno == ENOTCONN:   # wcfs crashed/killed
-                unclean = True
-            else:
-                raise
-        else:
+        fwcfs, trylockstartf = _try_attach_wcsrv(mntpt)
+        if fwcfs is not None:
             # already have it
-            wc = WCFS(mntpt, f, None)
+            wc = WCFS(mntpt, fwcfs, None)
             _wcregistry[mntpt] = wc
             return wc
 
@@ -228,10 +219,7 @@ def join(zurl, autostart=_default_autostart()): # -> WCFS
             raise RuntimeError("wcfs: join %s: server not running" % zurl)
 
         # start wcfs with telling it to automatically exit when there is no client activity.
-        # XXX race window if external process starts after ^^^ check
-        # TODO -> fs-level locking
-        if unclean:
-            _fuse_unmount(mntpt)
+        trylockstartf() # XXX retry access if another wcfs was started in the meantime
 
         wcsrv, fwcfs = _start(zurl, "-autoexit")
         wc = WCFS(mntpt, fwcfs, wcsrv)
@@ -242,11 +230,51 @@ def join(zurl, autostart=_default_autostart()): # -> WCFS
         return wc
 
 
+# _try_attach_wcsrv tries to attach to running wcfs server.
+#
+# if successful, it returns fwcfs - opened file handle for /.wcfs/zurl
+# if unsuccessful, it returns fwcfs=None, and trylockstartf function that can
+# be used to prepare to start new WCFS server.
+def _try_attach_wcsrv(mntpt): # -> (fwcfs, trylockstartf)
+    # try opening .wcfs - if we succeed - wcfs is already running.
+    unclean = False
+    try:
+        fwcfs = open(mntpt + "/.wcfs/zurl")
+    except IOError as e:
+        if   e.errno == ENOENT:     # wcfs cleanly unmounted
+            pass
+        elif e.errno == ENOTCONN:   # wcfs crashed/killed
+            unclean = True
+        else:
+            raise
+    else:
+        return (fwcfs, None)
+
+    # the server is not running.
+    # return func to prepare start of another wcfs server
+    def trylockstartf():
+        # XXX race window if external process starts after ^^^ check
+        # TODO -> fs-level locking
+        if unclean:
+            _fuse_unmount(mntpt)
+    return (None, trylockstartf)
+
+
 # start starts wcfs server for ZODB @ zurl.
 #
 # optv can be optionally given to pass flags to wcfs.
 def start(zurl, *optv): # -> Server
-    # XXX check already started?
+    # verify that wcfs is not already running
+    mntpt = _mntpt_4zurl(zurl)
+
+    fwcfs, trylockstartf = _try_attach_wcsrv(mntpt)
+    if fwcfs is not None:
+        fwcfs.close()
+        raise RuntimeError("wcfs: start %s: already running" % zurl)
+
+    # seems to be ok to start
+    trylockstartf() # XXX -> "already running" if lock fails
+
     wcsrv, fwcfs = _start(zurl, *optv)
     fwcfs.close()
     return wcsrv
@@ -618,26 +646,14 @@ def serve(zurl, optv, exec_=False, _tstartingq=None):
     # XXX take $WENDELIN_CORE_WCFS_OPTIONS into account?
 
     # try opening .wcfs - it is an error if we can do it.
-    # XXX -> option to wcfs itself to verify wcfs/something is already mounted?
-    unclean = False
-    try:
-        f = open(mntpt + "/.wcfs/zurl")
-    except IOError as e:
-        if   e.errno == ENOENT:     # wcfs cleanly unmounted
-            pass
-        elif e.errno == ENOTCONN:   # wcfs crashed/killed
-            unclean = True
-        else:
-            raise
-    else:
-        f.close()
-        raise RuntimeError("wcfs: start %s: already running" % zurl)
+    fwcfs, trylockstartf = _try_attach_wcsrv(mntpt)
+    if fwcfs is not None:
+        fwcfs.close()
+        raise RuntimeError("wcfs: serve %s: already running" % zurl)
 
     # seems to be ok to start
-    # XXX race window if external process starts after ^^^ check
-    # TODO -> fs-level locking
-    if unclean:
-        _fuse_unmount(mntpt)
+    trylockstartf() # XXX -> "already running" if lock fails
+
     if _tstartingq is not None:
         _tstartingq.close()
     argv = [_wcfs_exe()] + list(optv) + [zurl, mntpt]

wcfs/wcfs_test.py

@@ -198,6 +198,28 @@ def test_join_after_crash():
         procmounts_lookup_wcfs(zurl)
 
 
+# verify that start successfuly starts server if previous wcfs exited uncleanly.
+@func
+def test_start_after_crash():
+    zurl  = testzurl
+    mntpt = testmntpt
+
+    wc = start_and_crash_wcfs(zurl, mntpt)
+
+    wcsrv = wcfs.start(zurl)
+    defer(wcsrv.stop)
+    assert wcsrv.mountpoint == mntpt
+    assert readfile(mntpt + "/.wcfs/zurl") == zurl
+
+    # /proc/mounts should contain wcfs entry
+    assert procmounts_lookup_wcfs(zurl) == mntpt
+
+    # stop the server - /proc/mounts entry should be gone
+    wcsrv.stop()
+    with raises(KeyError):
+        procmounts_lookup_wcfs(zurl)
+
+
 # verify that serve successfully starts if previous wcfs exited uncleanly.
 @func
 def test_serve_after_crash():
@@ -226,8 +248,6 @@ def test_serve_after_crash():
     assert procmounts_lookup_wcfs(zurl) == mntpt
 
 
-# XXX test_start_after_crash?
-
 
 # start_and_crash_wcfs starts wcfs and then kills it.
 # it returns closed WCFS connection that was connected to the killed WCFS server.