Display a warning when concurrency is enabled but ssh connection caching is not enabled or won't work due to a crippled filesystem

A warning message is unsatisfying. But erroring out is too hard a failure,
especially since it may well work fine if the user has enabled passwordless
ssh.

I did think about falling back to one ssh connection at a time in this
case, but it would have needed a rework of every ssh call, which
seems far overboard for such a niche problem. There's no single place where
git-annex runs ssh, so no one place that it could block a concurrent call
on a semaphore. And, even if it did fall back to one ssh connection at a
time, it seems to me that doing so without warning the user about the
problem just invites bug reports like "git-annex is ignoring my -J2 and
only doing one download at a time". So a warning is needed, and I suppose
is good enough.

Annex/Ssh.hs

@@ -1,6 +1,6 @@
 {- git-annex ssh interface, with connection caching
  -
- - Copyright 2012-2017 Joey Hess <id@joeyh.name>
+ - Copyright 2012-2020 Joey Hess <id@joeyh.name>
  -
  - Licensed under the GNU AGPL version 3 or higher.
  -}
@@ -100,11 +100,44 @@ consumeStdinParams NoConsumeStdin = [Param "-n"]
 sshCachingInfo :: (SshHost, Maybe Integer) -> Annex (Maybe FilePath, [CommandParam])
 sshCachingInfo (host, port) = go =<< sshCacheDir
   where
-	go Nothing = return (Nothing, [])
 	go (Just dir) =
 		liftIO (bestSocketPath $ dir </> hostport2socket host port) >>= return . \case
 			Nothing -> (Nothing, [])
 			Just socketfile -> (Just socketfile, sshConnectionCachingParams socketfile)
+	-- No connection caching with concurrency is not a good
+	-- combination, so warn the user.
+	go Nothing = do
+		Annex.getState Annex.concurrency >>= \case
+                	NonConcurrent -> return ()
+			Concurrent {} -> warnnocaching
+			ConcurrentPerCpu -> warnnocaching
+		return (Nothing, [])
+	
+	warnnocaching = do
+		warning nocachingwarning
+		ifM (fromMaybe True . annexSshCaching <$> Annex.getGitConfig)
+			( whenM crippledFileSystem $
+				warning crippledfswarning
+			, warning enablecachingwarning
+			)
+	
+	nocachingwarning = unwords
+		[ "You have enabled concurrency, but ssh connection caching"
+		, "is not enabled. This may result in multiple ssh processes"
+		, "prompting for passwords at the same time."
+		]
+	
+	crippledfswarning = unwords
+		[ "This repository is on a crippled filesystem, so unix named"
+		, "pipes probably don't work, and ssh connection caching"
+		, "relies on those. One workaround is to set"
+		, sshSocketDirEnv
+		, "to point to a directory on a non-crippled filesystem."
+		, "(Or, disable concurrency.)"
+		]
+	
+	enablecachingwarning = 
+		"Enable annex.sshcaching (or disable concurrency) to avoid this problem."
 
 {- Given an absolute path to use for a socket file,
  - returns whichever is shorter of that or the relative path to the same
@@ -133,9 +166,11 @@ sshConnectionCachingParams socketfile =
 	, Param "-o", Param "ControlPersist=yes"
 	]
 
+sshSocketDirEnv :: String
+sshSocketDirEnv = "GIT_ANNEX_SSH_SOCKET_DIR"
+
 {- ssh connection caching creates sockets, so will not work on a
- - crippled filesystem. A GIT_ANNEX_TMP_DIR can be provided to use
+ - crippled filesystem. -}
- - a different filesystem. -}
 sshCacheDir :: Annex (Maybe FilePath)
 sshCacheDir
 	| BuildInfo.sshconnectioncaching = 
@@ -148,7 +183,7 @@ sshCacheDir
 			)
 	| otherwise = return Nothing
   where
-	gettmpdir = liftIO $ getEnv "GIT_ANNEX_TMP_DIR"
+	gettmpdir = liftIO $ getEnv sshSocketDirEnv
 	usetmpdir tmpdir = liftIO $ catchMaybeIO $ do
 		let socktmp = tmpdir </> "ssh"
 		createDirectoryIfMissing True socktmp

CHANGELOG

@@ -19,6 +19,8 @@ git-annex (7.20191231) UNRELEASED; urgency=medium
     uuid configured.
   * Support git remotes that need http basic auth to be accessed,
     using git credential to get the password.
+  * Display a warning when concurrency is enabled but ssh connection caching
+    is not enabled or won't work due to a crippled filesystem.
 
  -- Joey Hess <id@joeyh.name>  Wed, 01 Jan 2020 12:51:40 -0400
 

doc/bugs/2_ssh_connection_prompts_for_password.mdwn

@@ -63,3 +63,4 @@ get R042/R042-2013-08-16/R042-2013-08-16-CSC01a.ncs get R042/R042-2013-08-16/R04
 [[!meta author=yoh]]
 [[!tag projects/datalad]]
 
+> warning added; [[done]] --[[Joey]] 

doc/bugs/2_ssh_connection_prompts_for_password/comment_4_43e5a90d6215b414f4e3357ea91af663._comment

@@ -0,0 +1,21 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 4"""
+ date="2020-01-23T15:51:46Z"
+ content="""
+I notice that debug output has no BatchMode=true in any ssh call. But
+the version of git-annex you show always runs ssh with that when
+-J is used, unless sshcaching is disabled.
+
+More evidence that sshcaching is disabled in your transcript is that when
+it does run ssh, it does not pass -S.
+
+I think the repository must be on a crippled filesystem, on which
+git-annex can't do ssh connection caching, because the filesystem
+does not support unix sockets. (Or it potentially could be crippled in some
+other way.) So it ignores the annex.sshcaching setting.
+You could work around this by setting the (undocumented)
+GIT_ANNEX_TMP_DIR to some temporary directory on a non-crippled filesystem.
+
+I'm going to add a warning message in this situation.
+"""]]

standalone/linux/skel/runshell

@@ -212,8 +212,8 @@ case "$os" in
 		useproot=1
 		
 		# Store ssh connection caching sockets outside of sdcard.
-		GIT_ANNEX_TMP_DIR="$TMPDIR"
+		GIT_ANNEX_SSH_SOCKET_DIR="$TMPDIR"
-		export GIT_ANNEX_TMP_DIR
+		export GIT_ANNEX_SSH_SOCKET_DIR
 
 		GIT_ANNEX_STANDLONE_ENV="PATH GCONV_PATH MANPATH LOCPATH"
 		export GIT_ANNEX_STANDLONE_ENV