poll_fd(fd, timeout_s::Real=-1; readable=false, writable=false)

Monitor a file descriptor fd for changes in the read or write availability, and with a timeout given by timeout_s seconds.

The keyword arguments determine which of read and/or write status should be monitored; at least one of them must be set to true.

The returned value is an object with boolean fields readable, writable, and timedout, giving the result of the polling.

This is a thin wrapper over calling wait on a FDWatcher, which implements the functionality but requires the user to call close manually when finished with it, or risk serious crashes.

poll_file(path::AbstractString, interval_s::Real=5.007, timeout_s::Real=-1) -> (previous::StatStruct, current)

Monitor a file for changes by polling every interval_s seconds until a change occurs or timeout_s seconds have elapsed. The interval_s should be a long period; the default is 5.007 seconds.

Returns a pair of status objects (previous, current) when a change is detected. The previous status is always a StatStruct, but it may have all of the fields zeroed (indicating the file didn't previously exist, or wasn't previously accessible).

The current status object may be a StatStruct, an EOFError (indicating the timeout elapsed), or some other Exception subtype (if the stat operation failed: for example, if the path does not exist).

To determine when a file was modified, compare !(current isa StatStruct && prev == current) to detect notification of changes to the mtime or inode. However, using watch_file for this operation is preferred, since it is more reliable and efficient, although in some situations it may not be available.

This is a thin wrapper over calling wait on a PollingFileWatcher, which implements the functionality, but this function has a small race window between consecutive calls to poll_file where the file might change without being detected.

watch_file(path::AbstractString, timeout_s::Real=-1)

Watch file or directory path for changes until a change occurs or timeout_s seconds have elapsed. This function does not poll the file system and instead uses platform-specific functionality to receive notifications from the operating system (e.g. via inotify on Linux). See the NodeJS documentation linked below for details.

The returned value is an object with boolean fields renamed, changed, and timedout, giving the result of watching the file.

This behavior of this function varies slightly across platforms. See https://nodejs.org/api/fs.html#fs_caveats for more detailed information.

This is a thin wrapper over calling wait on a FileMonitor. This function has a small race window between consecutive calls to watch_file where the file might change without being detected. To avoid this race, use

fm = FileMonitor(path)
wait(fm)

directly, re-using the same fm each time you wait.

watch_folder(path::AbstractString, timeout_s::Real=-1)

Watch a file or directory path for changes until a change has occurred or timeout_s seconds have elapsed. This function does not poll the file system and instead uses platform-specific functionality to receive notifications from the operating system (e.g. via inotify on Linux). See the NodeJS documentation linked below for details.

This will continuing tracking changes for path in the background until unwatch_folder is called on the same path.

The returned value is an pair where the first field is the name of the changed file (if available) and the second field is an object with boolean fields renamed, changed, and timedout, giving the event.

This behavior of this function varies slightly across platforms. See https://nodejs.org/api/fs.html#fs_caveats for more detailed information.

This function is a thin wrapper over calling wait on a FolderMonitor, with added timeout support.

unwatch_folder(path::AbstractString)

Stop background tracking of changes for path. It is not recommended to do this while another task is waiting for watch_folder to return on the same path, as the result may be unpredictable.

FileMonitor(path::AbstractString)

Watch file or directory path (which must exist) for changes until a change occurs. This function does not poll the file system and instead uses platform-specific functionality to receive notifications from the operating system (e.g. via inotify on Linux). See the NodeJS documentation linked below for details.

fm = FileMonitor(path) acts like an auto-reset Event, so wait(fm) blocks until there has been at least one event in the file originally at the given path and then returns an object with boolean fields renamed, changed, timedout summarizing all changes that have occurred since the last call to wait returned.

This behavior of this function varies slightly across platforms. See https://nodejs.org/api/fs.html#fs_caveats for more detailed information.

FolderMonitor(folder::AbstractString)

Watch a file or directory path for changes until a change has occurred. This function does not poll the file system and instead uses platform-specific functionality to receive notifications from the operating system (e.g. via inotify on Linux). See the NodeJS documentation linked below for details.

This acts similar to a Channel, so calling take! (or wait) blocks until some change has occurred. The wait function will return a pair where the first field is the name of the changed file (if available) and the second field is an object with boolean fields renamed and changed, giving the event that occurred on it.

This behavior of this function varies slightly across platforms. See https://nodejs.org/api/fs.html#fs_caveats for more detailed information.

PollingFileWatcher(path::AbstractString, interval_s::Real=5.007)

Monitor a file for changes by polling stat every interval_s seconds until a change occurs or timeout_s seconds have elapsed. The interval_s should be a long period; the default is 5.007 seconds. Call stat on it to get the most recent, but old, result.

This acts like an auto-reset Event, so calling wait blocks until the stat result has changed since the previous value captured upon entry to the wait call. The wait function will return a pair of status objects (previous, current) once any stat change is detected since the previous time that wait was called. The previous status is always a StatStruct, but it may have all of the fields zeroed (indicating the file didn't previously exist, or wasn't previously accessible).

The current status object may be a StatStruct, an EOFError (if the wait is canceled by closing this object), or some other Exception subtype (if the stat operation failed: for example, if the path is removed). Note that stat value may be outdated if the file has changed again multiple times.

Using FileMonitor for this operation is preferred, since it is more reliable and efficient, although in some situations it may not be available.

FDWatcher(fd::Union{RawFD,WindowsRawSocket}, readable::Bool, writable::Bool)

Monitor a file descriptor fd for changes in the read or write availability.

The keyword arguments determine which of read and/or write status should be monitored; at least one of them must be set to true.

The returned value is an object with boolean fields readable, writable, and timedout, giving the result of the polling.

This acts like a level-set event, so calling wait blocks until one of those conditions is met, but then continues to return without blocking until the condition is cleared (either there is no more to read, or no more space in the write buffer, or both).

mkpidlock([f::Function], at::String, [pid::Cint]; kwopts...)
mkpidlock(at::String, proc::Process; kwopts...)

Create a pidfile lock for the path "at" for the current process or the process identified by pid or proc. Can take a function to execute once locked, for usage in do blocks, after which the lock will be automatically closed. If the lock fails and wait is false, then an error is thrown.

The lock will be released by either close, a finalizer, or shortly after proc exits. Make sure the return value is live through the end of the critical section of your program, so the finalizer does not reclaim it early.

Optional keyword arguments:

• mode: file access mode (modified by the process umask). Defaults to world-readable.
• poll_interval: Specify the maximum time to between attempts (if watch_file doesn't work)
• stale_age: Delete an existing pidfile (ignoring the lock) if it is older than this many seconds, based on its mtime. The file won't be deleted until 5x longer than this if the pid in the file appears that it may be valid. Or 25x longer if refresh is overridden to 0 to disable lock refreshing. By default this is disabled (stale_age = 0), but a typical recommended value would be about 3-5x an estimated normal completion time.
• refresh: Keeps a lock from becoming stale by updating the mtime every interval of time that passes. By default, this is set to stale_age/2, which is the recommended value.
• wait: If true, block until we get the lock, if false, raise error if lock fails.

trymkpidlock([f::Function], at::String, [pid::Cint]; kwopts...)
trymkpidlock(at::String, proc::Process; kwopts...)

Like mkpidlock except returns false instead of waiting if the file is already locked.

close(lock::LockMonitor)

Release a pidfile lock.

open_exclusive(path::String; mode, poll_interval, wait, stale_age, refresh) :: File

Create a new a file for read-write advisory-exclusive access. If wait is false then error out if the lock files exist otherwise block until we get the lock.

For a description of the keyword arguments, see mkpidlock.

tryopen_exclusive(path::String, mode::Integer = 0o444) :: Union{Void, File}

Try to create a new file for read-write advisory-exclusive access, return nothing if it already exists.

write_pidfile(io, pid)

Write our pidfile format to an open IO descriptor.

parse_pidfile(file::Union{IO, String}) => (pid, hostname, age)

Attempt to parse our pidfile format, replaced an element with (0, "", 0.0), respectively, for any read that failed.

stale_pidfile(path::String, stale_age::Real, refresh::Real) :: Bool

Helper function for open_exclusive for deciding if a pidfile is stale.

isvalidpid(hostname::String, pid::Cuint) :: Bool

Attempt to conservatively estimate whether pid is a valid process id.

Base.touch(::Pidfile.LockMonitor)

Update the mtime on the lock, to indicate it is still fresh.

See also the refresh keyword in the mkpidlock constructor.