mirrors/git-annex
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages 1 Wiki Activity Actions

remove unncessary locking of ContentIdentifier db

Browse source 
Remote.Helper.ExportImport only reads from it, and locking is only
needed when writing.
This commit is contained in:
Joey Hess 2019-03-06 14:35:16 -04:00
commit b3d30e7d70
No known key found for this signature in database
GPG key ID: DB12DB0FF05F8F38
2 changed files with 5 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

13
Remote/Helper/ExportImport.hs
View file

@ -19,7 +19,6 @@ import qualified Database.Export as Export
import qualified Database.ContentIdentifier as ContentIdentifier
import Annex.Export
import Annex.Import
import Annex.LockFile
import Config
import Git.Types (fromRef)
import Logs.Export
@ -263,24 +262,22 @@ adjustExportImport r = case M.lookup "exporttree" (config r) of
exportupdated <- liftIO $ newTMVarIO ()
return (dbv, lcklckv, exportinconflict, exportupdated)
-- Only open the database once it's needed, and take an
-- exclusive write lock. The write lock will then remain
-- held while the process is running.
-- Only open the database once it's needed.
getciddb (dbtv, lcklckv) =
liftIO (atomically (tryReadTMVar dbtv)) >>= \case
Just (db, _lck) -> return db
Just db -> return db
-- let only one thread take the lock
Nothing -> ifM (liftIO $ atomically $ tryPutTMVar lcklckv ())
( do
lck <- takeExclusiveLock gitAnnexContentIdentifierLock
db <- ContentIdentifier.openDb
liftIO $ atomically $ putTMVar dbtv (db, lck)
liftIO $ atomically $ putTMVar dbtv db
return db
-- loser waits for winner to open the db and
-- can then also use its handle
, liftIO $ fst <$> atomically (readTMVar dbtv)
, liftIO $ atomically (readTMVar dbtv)
)
-- Only open the database once it's needed.
getexportdb (dbv, lcklckv, _, _) =
liftIO (atomically (tryReadTMVar dbv)) >>= \case
Just db -> return db

8
doc/todo/import_tree.mdwn
View file

@ -21,14 +21,6 @@ this.
and conflicting new files. Note need to check both sequences
import,export and export,import.
* Should the ContentIdentifier db be multiwriter? It would simplify
the situation with the long-lived lock of it in adjustExportImport
OTOH, if a single process, such as the assistant, exported one tree,
and then another, multiwriter would mean no guarantee that reading
from the contentidentifier database would see values that were earlier
written by the same process.
* Need to support annex-tracking-branch configuration, which documentation
says makes git-annex sync and assistant do imports.