git-annex/Git/Queue.hs

268 lines
8.9 KiB
Haskell
Raw Normal View History

2010-10-27 17:12:02 +00:00
{- git repository command queue
2010-10-27 20:53:54 +00:00
-
- Copyright 2010-2022 Joey Hess <id@joeyh.name>
2010-10-27 20:53:54 +00:00
-
- Licensed under the GNU AGPL version 3 or higher.
-}
{-# LANGUAGE CPP, BangPatterns #-}
2011-06-30 17:25:37 +00:00
module Git.Queue (
Queue,
2011-12-20 18:37:53 +00:00
new,
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
defaultTimelimit,
addCommand,
addUpdateIndex,
addFlushAction,
FlushActionRunner(..),
size,
full,
flush,
merge,
) where
import Utility.SafeCommand
2011-12-20 18:37:53 +00:00
import Common
2011-06-30 17:25:37 +00:00
import Git
2011-12-14 19:56:11 +00:00
import Git.Command
import qualified Git.UpdateIndex
import qualified Data.Map.Strict as M
import Control.Monad.IO.Class
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
import Data.Time.Clock
import Data.Time.Clock.POSIX
{- Queable actions that can be performed in a git repository. -}
data Action m
{- Updating the index file, using a list of streamers that can
- be added to as the queue grows. -}
= UpdateIndexAction [Git.UpdateIndex.Streamer] -- in reverse order
{- A git command to run, on a list of files that can be added to
- as the queue grows. -}
| CommandAction
{ getCommonParams :: [CommandParam]
-- ^ parameters that come before the git subcommand
-- (in addition to the Repo's gitGlobalOpts.
, getSubcommand :: String
, getParams :: [CommandParam]
-- ^ parameters that come after the git subcommand
, getFiles :: [CommandParam]
}
{- A FlushAction can be added along with CommandActions or
- UpdateIndexActions, and when the queue later gets flushed,
- those will be run before the FlushAction is. -}
| FlushAction
{ getFlushActionRunner :: FlushActionRunner m
, getFlushActionFiles :: [(RawFilePath, IO Bool, FileSize)]
}
{- The String must be unique for each flush action. -}
data FlushActionRunner m = FlushActionRunner String (Repo -> [(RawFilePath, IO Bool, FileSize)] -> m ())
instance Eq (FlushActionRunner m) where
FlushActionRunner s1 _ == FlushActionRunner s2 _ = s1 == s2
{- A key that can uniquely represent an action in a Map.
-
- The ordering controls what order the actions are run in when flushing
- the queue. -}
data ActionKey
= UpdateIndexActionKey
| CommandActionKey [CommandParam] String [CommandParam]
| FlushActionKey String
deriving (Eq, Ord)
actionKey :: Action m -> ActionKey
actionKey (UpdateIndexAction _) = UpdateIndexActionKey
actionKey CommandAction { getCommonParams = c, getSubcommand = s, getParams = p } = CommandActionKey c s p
actionKey FlushAction { getFlushActionRunner = FlushActionRunner s _ } = FlushActionKey s
{- A queue of actions to perform (in any order) on a git repository,
- with lists of files to perform them on. This allows coalescing
- similar git commands. -}
data Queue m = Queue
{ size :: Int
, _limit :: Int
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
, _timelimit :: NominalDiffTime
, _lastchanged :: POSIXTime
, items :: M.Map ActionKey (Action m)
}
{- A recommended maximum size for the queue, after which it should be
- run.
-
- 10240 is semi-arbitrary. If we assume git filenames are between 10 and
- 255 characters long, then the queue will build up between 100kb and
- 2550kb long commands. The max command line length on linux is somewhere
- above 20k, so this is a fairly good balance -- the queue will buffer
- only a few megabytes of stuff and a minimal number of commands will be
- run by xargs. -}
defaultLimit :: Int
defaultLimit = 10240
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
{- How close together in seconds changes to the queue have to be happening
2021-12-14 17:53:36 +00:00
- in order for it to keep accumulating actions, rather than running actions
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
- immediately. -}
defaultTimelimit :: NominalDiffTime
defaultTimelimit = 60 * 5
{- Constructor for empty queue. -}
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
new :: Maybe Int -> Maybe NominalDiffTime -> IO (Queue m)
new lim tlim = do
now <- getPOSIXTime
return $ Queue 0
(fromMaybe defaultLimit lim)
(fromMaybe defaultTimelimit tlim)
now
M.empty
{- Adds an git command to the queue.
-
- Git commands with the same subcommand but different parameters are
- assumed to be equivilant enough to perform in any order with the same
- end result.
-}
addCommand :: MonadIO m => [CommandParam] -> String -> [CommandParam] -> [FilePath] -> Queue m -> Repo -> m (Queue m)
addCommand commonparams subcommand params files q repo =
updateQueue action conflicting (length files) q repo
2012-12-13 04:24:19 +00:00
where
action = CommandAction
{ getCommonParams = commonparams
, getSubcommand = subcommand
2012-12-13 04:24:19 +00:00
, getParams = params
, getFiles = map File files
2012-12-13 04:24:19 +00:00
}
conflicting (CommandAction { getSubcommand = s }) = s /= subcommand
conflicting (FlushAction {}) = False
conflicting _ = True
{- Adds an flush action to the queue. This can co-exist with anything else
- that gets added to the queue, and when the queue is eventually flushed,
- it will be run after the other things in the queue. -}
addFlushAction :: MonadIO m => FlushActionRunner m -> [(RawFilePath, IO Bool, FileSize)] -> Queue m -> Repo -> m (Queue m)
addFlushAction runner files q repo =
updateQueue action (const False) (length files) q repo
where
action = FlushAction
{ getFlushActionRunner = runner
, getFlushActionFiles = files
}
{- Adds an update-index streamer to the queue. -}
addUpdateIndex :: MonadIO m => Git.UpdateIndex.Streamer -> Queue m -> Repo -> m (Queue m)
addUpdateIndex streamer q repo =
updateQueue action conflicting 1 q repo
2012-12-13 04:24:19 +00:00
where
-- the list is built in reverse order
action = UpdateIndexAction [streamer]
conflicting (UpdateIndexAction _) = False
conflicting (FlushAction {}) = False
conflicting _ = True
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
{- Updates or adds an action in the queue.
-
- If the queue already contains a conflicting action, it will be flushed
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
- before adding the action; this is to ensure that conflicting actions,
- like add and rm, are run in the right order.
-
- If the queue's time limit has been exceeded, it will also be flushed,
- and the action will be run right away.
-}
updateQueue :: MonadIO m => Action m -> (Action m -> Bool) -> Int -> Queue m -> Repo -> m (Queue m)
updateQueue !action conflicting sizeincrease q repo = do
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
now <- liftIO getPOSIXTime
if now - (_lastchanged q) > _timelimit q
then if isconflicting
then do
q' <- flush q repo
flush (mk q') repo
else flush (mk q) repo
else if isconflicting
then mk <$> flush q repo
else return $ mk (q { _lastchanged = now })
2012-12-13 04:24:19 +00:00
where
isconflicting = not (null (filter conflicting (M.elems (items q))))
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
mk q' = newq
2012-12-13 04:24:19 +00:00
where
!newq = q'
{ size = newsize
, items = newitems
}
!newsize = size q' + sizeincrease
!newitems = M.insertWith combineNewOld (actionKey action) action (items q')
2018-08-17 17:19:37 +00:00
{- The new value comes first. It probably has a smaller list of files than
- the old value. So, the list append of the new value first is more
- efficient. -}
combineNewOld :: Action m -> Action m -> Action m
combineNewOld (CommandAction _cps1 _sc1 _ps1 fs1) (CommandAction cps2 sc2 ps2 fs2) =
CommandAction cps2 sc2 ps2 (fs1++fs2)
combineNewOld (UpdateIndexAction s1) (UpdateIndexAction s2) =
UpdateIndexAction (s1++s2)
combineNewOld (FlushAction _r1 fs1) (FlushAction r2 fs2) =
FlushAction r2 (fs1++fs2)
combineNewOld anew _aold = anew
{- Merges the contents of the second queue into the first.
- This should only be used when the two queues are known to contain
- non-conflicting actions. -}
merge :: Queue m -> Queue m -> Queue m
merge origq newq = origq
{ size = size origq + size newq
, items = M.unionWith combineNewOld (items newq) (items origq)
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
, _lastchanged = max (_lastchanged origq) (_lastchanged newq)
}
{- Is a queue large enough that it should be flushed? -}
full :: Queue m -> Bool
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
full (Queue cur lim _ _ _) = cur >= lim
{- Runs a queue on a git repository. -}
flush :: MonadIO m => Queue m -> Repo -> m (Queue m)
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
flush (Queue _ lim tlim _ m) repo = do
forM_ (M.elems m) $ runAction repo
improve git command queue flushing with time limit So that eg, addurl of several large files that take time to download will update the index for each file, rather than deferring the index updates to the end. In cases like an add of many smallish files, where a new file is being added every few seconds. In that case, the queue will still build up a lot of changes which are flushed at once, for best performance. Since the default queue size is 10240, often it only gets flushed once at the end, same as before. (Notice that updateQueue updated _lastchanged when adding a new item to the queue without flushing it; that is necessary to avoid it flushing the queue every 5 minutes in this case.) But, when it takes more than a 5 minutes to add a file, the overhead of updating the index immediately is probably small, so do it after each file. This avoids git-annex potentially taking a very very long time indeed to stage newly added files, which can be annoying to the user who would like to get on with doing something with the files it's already added, eg using git mv to rename them to a better name. This is only likely to cause a problem if it takes say, 30 seconds to update the index; doing an extra 30 seconds of work after every 5 minute file add would be less optimal. Normally, updating the index takes significantly less time than that. On a SSD with 100k files it takes less than 1 second, and the index write time is bound by disk read and write so is not too much worse on a hard drive. So I hope this will not impact users, although if it does turn out to, the time limit could be made configurable. A perhaps better way to do it would be to have a background worker thread that wakes up every 60 seconds or so and flushes the queue. That is made somewhat difficult because the queue can contain Annex actions and so this would add a new source of concurrency issues. So I'm trying to avoid that approach if possible. Sponsored-by: Erik Bjäreholt on Patreon
2021-12-14 15:48:07 +00:00
now <- liftIO getPOSIXTime
return $ Queue 0 lim tlim now M.empty
{- Runs an Action on a list of files in a git repository.
-
- Complicated by commandline length limits.
-
- Intentionally runs the command even if the list of files is empty;
- this allows queueing commands that do not need a list of files. -}
runAction :: MonadIO m => Repo -> Action m -> m ()
runAction repo (UpdateIndexAction streamers) =
2012-06-13 01:13:15 +00:00
-- list is stored in reverse order
liftIO $ Git.UpdateIndex.streamUpdateIndex repo $ reverse streamers
runAction repo action@(CommandAction {}) = liftIO $ do
#ifndef mingw32_HOST_OS
let p = (proc "xargs" $ "-0":"git":toCommand gitparams)
{ env = gitEnv repo
, std_in = CreatePipe
}
withCreateProcess p (go p)
#else
-- Using xargs on Windows is problematic, so just run the command
-- once per file (not as efficient.)
if null (getFiles action)
then void $ boolSystemEnv "git" gitparams (gitEnv repo)
else forM_ (getFiles action) $ \f ->
void $ boolSystemEnv "git" (gitparams ++ [f]) (gitEnv repo)
#endif
2012-12-13 04:24:19 +00:00
where
gitparams = gitCommandLine
(getCommonParams action++Param (getSubcommand action):getParams action)
repo
2020-11-23 18:00:17 +00:00
#ifndef mingw32_HOST_OS
2020-06-05 20:38:11 +00:00
go p (Just h) _ _ pid = do
hPutStr h $ intercalate "\0" $ toCommand $ getFiles action
hClose h
forceSuccessProcess p pid
go _ _ _ _ _ = error "internal"
2020-11-23 18:00:17 +00:00
#endif
runAction repo action@(FlushAction {}) =
let FlushActionRunner _ runner = getFlushActionRunner action
in runner repo (getFlushActionFiles action)