Improve race recovery code when committing to git-annex branch.

Annex/Branch.hs

@@ -29,6 +29,7 @@ import qualified Data.ByteString.Lazy as L
 import qualified Data.Set as S
 import qualified Data.Map as M
 import Data.Bits.Utils
+import Control.Concurrent (threadDelay)
 
 import Common.Annex
 import Annex.BranchState
@@ -232,28 +233,32 @@ forceCommit message = lockJournal $ \jl -> do
 
 {- Commits the staged changes in the index to the branch.
  - 
- - Ensures that the branch's index file is first updated to the state
+ - Ensures that the branch's index file is first updated to merge the state
  - of the branch at branchref, before running the commit action. This
  - is needed because the branch may have had changes pushed to it, that
  - are not yet reflected in the index.
  - 
- - Also safely handles a race that can occur if a change is being pushed
- - into the branch at the same time. When the race happens, the commit will
- - be made on top of the newly pushed change, but without the index file
- - being updated to include it. The result is that the newly pushed
- - change is reverted. This race is detected and another commit made
- - to fix it.
- - 
  - The branchref value can have been obtained using getBranch at any
  - previous point, though getting it a long time ago makes the race
  - more likely to occur.
+ -
+ - Note that changes may be pushed to the branch at any point in time!
+ - So, there's a race. If the commit is made using the newly pushed tip of
+ - the branch as its parent, and that ref has not yet been merged into the
+ - index, then the result is that the commit will revert the pushed
+ - changes, since they have not been merged into the index. This race
+ - is detected and another commit made to fix it.
+ -
+ - (It's also possible for the branch to be overwritten,
+ - losing the commit made here. But that's ok; the data is still in the
+ - index and will get committed again later.)
  -}
 commitIndex :: JournalLocked -> Git.Ref -> String -> [Git.Ref] -> Annex ()
 commitIndex jl branchref message parents = do
 	showStoringStateAction
-	commitIndex' jl branchref message parents
+	commitIndex' jl branchref message message 0 parents
-commitIndex' :: JournalLocked -> Git.Ref -> String -> [Git.Ref] -> Annex ()
+commitIndex' :: JournalLocked -> Git.Ref -> String -> String -> Integer -> [Git.Ref] -> Annex ()
-commitIndex' jl branchref message parents = do
+commitIndex' jl branchref message basemessage retrynum parents = do
 	updateIndex jl branchref
 	committedref <- inRepo $ Git.Branch.commitAlways Git.Branch.AutomaticCommit message fullname parents
 	setIndexSha committedref
@@ -276,12 +281,16 @@ commitIndex' jl branchref message parents = do
 		| otherwise = True -- race!
 		
 	{- To recover from the race, union merge the lost refs
-	 - into the index, and recommit on top of the bad commit. -}
+	 - into the index. -}
 	fixrace committedref lostrefs = do
+		showSideAction "recovering from race"
+		let retrynum' = retrynum+1
+		-- small sleep to let any activity that caused
+		-- the race settle down
+		liftIO $ threadDelay (100000 + fromInteger retrynum')
 		mergeIndex jl lostrefs
-		commitIndex jl committedref racemessage [committedref]
+		let racemessage = basemessage ++ " (recovery from race #" ++ show retrynum' ++ "; expected commit parent " ++ show branchref ++ " but found " ++ show lostrefs ++ " )"
-		
+		commitIndex' jl committedref racemessage basemessage retrynum' [committedref]
-	racemessage = message ++ " (recovery from race)"
 
 {- Lists all files on the branch. There may be duplicates in the list. -}
 files :: Annex [FilePath]

debian/changelog

@@ -17,6 +17,7 @@ git-annex (5.20150206) UNRELEASED; urgency=medium
   * webapp: Fix reversion in opening webapp when starting it manually
     inside a repository.
   * assistant: Improve sanity check for control characters when pairing.
+  * Improve race recovery code when committing to git-annex branch.
 
  -- Joey Hess <id@joeyh.name>  Fri, 06 Feb 2015 13:57:08 -0400
 

doc/bugs/git-annex_branch_shows_commit_with_looong_commitlog/comment_5_c97926b15ba320f57a6441f9844cb139._comment

@@ -0,0 +1,7 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 5"""
+ date="2015-02-09T21:10:58Z"
+ content="""
+I guess the thing to do in this case is to run `git annex forget`
+"""]]

doc/bugs/git-annex_branch_shows_commit_with_looong_commitlog/comment_6_3b70a60ef1c47871a3933176eac38174._comment

@@ -0,0 +1,40 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 6"""
+ date="2015-02-09T21:59:31Z"
+ content="""
+[[forum/repair_stuck_on_ls-tree_command]] is another case of that, and I got ahold of
+that repository for analysis.
+
+In that case, there was indeed an inverse pyramid effect where each commit
+added one more " (recovery from race)" to its parent commit.
+
+The code can clearly loop 
+if it keeps detecting a race and somehow fails to recover from it. Leading
+to a whole stack of commits with progressively longer messages.
+I don't see any way to get just one commit with a long message, which
+comment #1 seems to say happened.
+
+Apparently loops for a while and then succeeds in recovering from
+the race, since it then stops looping.
+
+I have added additional debug info to the commit message, in hopes of detecting
+what might be going wrong that causes it to loop.
+
+Seems to me there are two possibilities.
+
+One is that something else is continually changing the git-annex
+branch in a way that keeps triggering the race. If so, it might make
+sense for git-annex to do a brief random sleep (a few hundredths of a
+second) before recovering, to let whatever it is quiet down. I've done so.
+
+The other is some kind of bug where it detects a race when none
+occurred. Perhaps it's misparsing the commit or git cat-file is failing
+to output it, and so it's not finding the expected parent refs, for example.
+But in that case, why would it detect a race for many commits
+in a row, and then eventually not detect a race anymore?
+
+Also, I've made these messages no longer stack up even if it does go into a
+loop, which will at least help with the object size bloat, though not with the
+number of commits bloat.
+"""]]

doc/forum/repair_stuck_on_ls-tree_command/comment_11_1d951126a9633b206dffbc77bfc65f6a._comment

@@ -0,0 +1,12 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 11"""
+ date="2015-02-09T21:13:28Z"
+ content="""
+Finally got back to this. I downloaded the file.
+
+You may be able to fix your repository by running `git annex forget`
+
+I guess this is the same problem described in
+[[bugs/git-annex_branch_shows_commit_with_looong_commitlog]]
+"""]]