git-annex/Database/Handle.hs

218 lines
6.3 KiB
Haskell
Raw Normal View History

{- Persistent sqlite database handles.
-
- Copyright 2015 Joey Hess <id@joeyh.name>
-
- Licensed under the GNU GPL version 3 or higher.
-}
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
{-# LANGUAGE BangPatterns #-}
module Database.Handle (
DbHandle,
initDb,
openDb,
queryDb,
closeDb,
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
Size,
queueDb,
flushQueueDb,
commitDb,
) where
import Utility.Exception
import Utility.Monad
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
import Database.Persist.Sqlite
import qualified Database.Sqlite as Sqlite
import Control.Monad
import Control.Monad.IO.Class (liftIO)
import Control.Concurrent
import Control.Concurrent.Async
import Control.Exception (throwIO)
import qualified Data.Text as T
import Control.Monad.Trans.Resource (runResourceT)
import Control.Monad.Logger (runNoLoggingT)
import Data.List
import Data.Time.Clock
2015-12-09 18:55:47 +00:00
import System.IO
{- A DbHandle is a reference to a worker thread that communicates with
- the database. It has a MVar which Jobs are submitted to. -}
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
data DbHandle = DbHandle (Async ()) (MVar Job) (MVar DbQueue)
{- Ensures that the database is initialized. Pass the migration action for
- the database.
-
- The database is put into WAL mode, to prevent readers from blocking
- writers, and prevent a writer from blocking readers.
-}
initDb :: FilePath -> SqlPersistM () -> IO ()
initDb f migration = do
let db = T.pack f
enableWAL db
runSqlite db migration
enableWAL :: T.Text -> IO ()
enableWAL db = do
conn <- Sqlite.open db
stmt <- Sqlite.prepare conn (T.pack "PRAGMA journal_mode=WAL;")
void $ Sqlite.step stmt
void $ Sqlite.finalize stmt
Sqlite.close conn
{- Opens the database, but does not perform any migrations. Only use
- if the database is known to exist and have the right tables. -}
openDb :: FilePath -> TableName -> IO DbHandle
openDb db tablename = do
jobs <- newEmptyMVar
worker <- async (workerThread (T.pack db) tablename jobs)
q <- newMVar =<< emptyDbQueue
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
return $ DbHandle worker jobs q
data Job
= QueryJob (SqlPersistM ())
| ChangeJob ((SqlPersistM () -> IO ()) -> IO ())
| CloseJob
type TableName = String
workerThread :: T.Text -> TableName -> MVar Job -> IO ()
workerThread db tablename jobs = catchNonAsync (run loop) showerr
where
2015-12-09 18:55:47 +00:00
showerr e = liftIO $ hPutStrLn stderr $
"sqlite worker thread crashed: " ++ show e
loop = do
job <- liftIO $ takeMVar jobs
case job of
QueryJob a -> a >> loop
-- change is run in a separate database connection
-- since sqlite only supports a single writer at a
-- time, and it may crash the database connection
ChangeJob a -> liftIO (a run) >> loop
CloseJob -> return ()
-- like runSqlite, but calls settle on the raw sql Connection.
run a = do
conn <- Sqlite.open db
settle conn
runResourceT $ runNoLoggingT $
withSqlConn (wrapConnection conn) $
runSqlConn a
-- Work around a bug in sqlite: New database connections can
-- sometimes take a while to become usable; select statements will
-- fail with ErrorBusy for some time. So, loop until a select
-- succeeds; once one succeeds the connection will stay usable.
-- <http://thread.gmane.org/gmane.comp.db.sqlite.general/93116>
settle conn = do
r <- tryNonAsync $ do
stmt <- Sqlite.prepare conn nullselect
void $ Sqlite.step stmt
void $ Sqlite.finalize stmt
case r of
Right _ -> return ()
Left e -> do
if "ErrorBusy" `isInfixOf` show e
then do
threadDelay 1000 -- 1/1000th second
settle conn
else throwIO e
-- This should succeed for any table.
nullselect = T.pack $ "SELECT null from " ++ tablename ++ " limit 1"
{- Makes a query using the DbHandle. This should not be used to make
- changes to the database!
-
- Note that the action is not run by the calling thread, but by a
- worker thread. Exceptions are propigated to the calling thread.
-
- Only one action can be run at a time against a given DbHandle.
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
- If called concurrently in the same process, this will block until
- it is able to run.
-}
queryDb :: DbHandle -> SqlPersistM a -> IO a
queryDb (DbHandle _ jobs _) a = do
res <- newEmptyMVar
putMVar jobs $ QueryJob $
liftIO . putMVar res =<< tryNonAsync a
(either throwIO return =<< takeMVar res)
`catchNonAsync` (const $ error "sqlite query crashed")
closeDb :: DbHandle -> IO ()
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
closeDb h@(DbHandle worker jobs _) = do
putMVar jobs CloseJob
wait worker
flushQueueDb h
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
type Size = Int
type LastCommitTime = UTCTime
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
{- A queue of actions to perform, with a count of the number of actions
- queued, and a last commit time. -}
data DbQueue = DbQueue Size LastCommitTime (SqlPersistM ())
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
emptyDbQueue :: IO DbQueue
emptyDbQueue = do
now <- getCurrentTime
return $ DbQueue 0 now (return ())
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
{- Queues a change to be made to the database. It will be buffered
- to be committed later, unless the commitchecker action returns true.
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
-
- (Be sure to call closeDb or flushQueueDb to ensure the change
- gets committed.)
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
-
- Transactions built up by queueDb are sent to sqlite all at once.
- If sqlite fails due to another change being made concurrently by another
- process, the transaction is put back in the queue. This solves
- the sqlite multiple writer problem.
-}
queueDb
:: DbHandle
-> (Size -> LastCommitTime -> IO Bool)
-> SqlPersistM ()
-> IO ()
queueDb h@(DbHandle _ _ qvar) commitchecker a = do
DbQueue sz lastcommittime qa <- takeMVar qvar
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
let !sz' = sz + 1
let qa' = qa >> a
let enqueue = putMVar qvar
ifM (commitchecker sz' lastcommittime)
( do
r <- commitDb h qa'
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
case r of
Left _ -> enqueue $ DbQueue sz' lastcommittime qa'
Right _ -> do
now <- getCurrentTime
enqueue $ DbQueue 0 now (return ())
, enqueue $ DbQueue sz' lastcommittime qa'
)
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
{- If flushing the queue fails, this could be because there is another
- writer to the database. Retry repeatedly for up to 10 seconds. -}
flushQueueDb :: DbHandle -> IO ()
flushQueueDb h@(DbHandle _ _ qvar) = do
DbQueue sz _ qa <- takeMVar qvar
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
when (sz > 0) $
robustly Nothing 100 (commitDb h qa)
allow for concurrent incremental fsck processes again (sorta) Sqlite doesn't support multiple concurrent writers at all. One of them will fail to write. It's not even possible to have two processes building up separate transactions at the same time. Before using sqlite, incremental fsck could work perfectly well with multiple fsck processes running concurrently. I'd like to keep that working. My partial solution, so far, is to make git-annex buffer writes, and every so often send them all to sqlite at once, in a transaction. So most of the time, nothing is writing to the database. (And if it gets unlucky and a write fails due to a collision with another writer, it can just wait and retry the write later.) This lets multiple processes write to the database successfully. But, for the purposes of concurrent, incremental fsck, it's not ideal. Each process doesn't immediately learn of files that another process has checked. So they'll tend to do redundant work. Only way I can see to improve this is to use some other mechanism for short-term IPC between the fsck processes. Not yet done. ---- Also, make addDb check if an item is in the database already, and not try to re-add it. That fixes an intermittent crash with "SQLite3 returned ErrorConstraint while attempting to perform step." I am not 100% sure why; it only started happening when I moved write buffering into the queue. It seemed to generally happen on the same file each time, so could just be due to multiple files having the same key. However, I doubt my sound repo has many duplicate keys, and I suspect something else is going on. ---- Updated benchmark, with the 1000 item queue: 6m33.808s
2015-02-17 20:39:35 +00:00
where
robustly :: Maybe SomeException -> Int -> IO (Either SomeException ()) -> IO ()
robustly e 0 _ = error $ "failed to commit changes to sqlite database: " ++ show e
robustly _ n a = do
r <- a
case r of
Right _ -> return ()
Left e -> do
threadDelay 100000 -- 1/10th second
robustly (Just e) (n-1) a
commitDb :: DbHandle -> SqlPersistM () -> IO (Either SomeException ())
commitDb (DbHandle _ jobs _) a = do
res <- newEmptyMVar
putMVar jobs $ ChangeJob $ \runner ->
liftIO $ putMVar res =<< tryNonAsync (runner a)
takeMVar res