propellor-5.4.1: property-based host configuration management in haskell

Propellor.Property.Atomic

Synopsis

# Documentation

atomicDirUpdate :: (Cannot_ensureProperty_WithInfo t ~ True, (Targets t NotSuperset Targets t) ~ CanCombine) => SingI t => FilePath -> (FilePath -> Property (MetaTypes t)) -> Property (MetaTypes t) Source #

Applies a Property to a directory such that the directory is updated fully atomically; there is no point in time in which the directory will be in an inconsistent state.

For example, git repositories are not usually updated atomically, and so while the repository is being updated, the files in it can be a mixture of two different versions, which could cause unexpected behavior to consumers. To avoid such problems:

	& atomicDirUpdate "/srv/web/example.com"
(\d -> Git.pulled "joey" "http://.." d Nothing)

This operates by making a second copy of the directory, and passing it to the Property, which can make whatever changes it needs to that copy, non-atomically. After the Property successfully makes a change, the copy is swapped into place, fully atomically.

This necessarily uses double the disk space, since there are two copies of the directory. The parent directory will actually contain three children: a symlink with the name of the directory itself, and two copies of the directory, with names suffixed with ".1" and ".2"

This can optionally be used after atomicDirUpdate to rsync the changes that were made over to the other copy of the directory. It's not necessary to use this, but it can improve efficiency.

For example:

	& atomicDirUpdate "/srv/web/example.com"
(\d -> Git.pulled "joey" "http://.." d Nothing)
onChange atomicDirSync "/srv/web/example.com"

Using atomicDirSync in the above example lets git only download the changes once, rather than the same changes being downloaded a second time to update the other copy of the directory the next time propellor runs.

Suppose that a web server program is run from the git repository, and needs to be restarted after the pull. That restart should be done after the atomicDirUpdate, but before the atomicDirSync. That way, the old web server process will not have its files changed out from under it.

	& atomicDirUpdate "/srv/web/example.com"
(\d -> Git.pulled "joey" "http://.." d Nothing)
onChange (webServerRestart before atomicDirSync "/srv/web/example.com")

atomicUpdate :: (Cannot_ensureProperty_WithInfo t ~ True, (Targets t NotSuperset Targets t) ~ CanCombine) => SingI t => AtomicResourcePair a -> CheckAtomicResourcePair a -> SwapAtomicResourcePair a -> (a -> Property (MetaTypes t)) -> Property (MetaTypes t) Source #

Makes a non-atomic Property be atomic, by applying it to the inactiveAtomicResource, and if it was successful, atomically activating that resource.

A pair of resources, one active and one inactive, which can swap positions atomically.

Constructors

 AtomicResourcePair Fields

Action that activates the inactiveAtomicResource, and deactivates the activeAtomicResource. This action must be fully atomic.

Checks which of the pair of resources is currently active and which is inactive, and puts them in the correct poisition in the AtomicResourcePair.