README
¶
fsnotify is a Go library to provide cross-platform filesystem notifications on Windows, Linux, macOS, BSD, and illumos.
Go 1.17 or newer is required; the full documentation is at https://pkg.go.dev/github.com/fsnotify/fsnotify
Platform support:
| Backend | OS | Status |
|---|---|---|
| inotify | Linux | Supported |
| kqueue | BSD, macOS | Supported |
| ReadDirectoryChangesW | Windows | Supported |
| FEN | illumos | Supported |
| fanotify | Linux 5.9+ | Not yet |
| AHAFS | AIX | aix branch; experimental due to lack of maintainer and test environment |
| FSEvents | macOS | Needs support in x/sys/unix |
| USN Journals | Windows | Needs support in x/sys/windows |
| Polling | All | Not yet |
Linux and illumos should include Android and Solaris, but these are currently untested.
Usage
A basic example:
package main
import (
"log"
"github.com/fsnotify/fsnotify"
)
func main() {
// Create new watcher.
watcher, err := fsnotify.NewWatcher()
if err != nil {
log.Fatal(err)
}
defer watcher.Close()
// Start listening for events.
go func() {
for {
select {
case event, ok := <-watcher.Events:
if !ok {
return
}
log.Println("event:", event)
if event.Has(fsnotify.Write) {
log.Println("modified file:", event.Name)
}
case err, ok := <-watcher.Errors:
if !ok {
return
}
log.Println("error:", err)
}
}
}()
// Add a path.
err = watcher.Add("/tmp")
if err != nil {
log.Fatal(err)
}
// Block main goroutine forever.
<-make(chan struct{})
}
Some more examples can be found in cmd/fsnotify, which can be run with:
% go run ./cmd/fsnotify
Further detailed documentation can be found in godoc: https://pkg.go.dev/github.com/fsnotify/fsnotify
FAQ
Will a file still be watched when it's moved to another directory?
No, not unless you are watching the location it was moved to.
Are subdirectories watched?
No, you must add watches for any directory you want to watch (a recursive watcher is on the roadmap: #18).
Do I have to watch the Error and Event channels in a goroutine?
Yes. You can read both channels in the same goroutine using select (you don't
need a separate goroutine for both channels; see the example).
Why don't notifications work with NFS, SMB, FUSE, /proc, or /sys?
fsnotify requires support from underlying OS to work. The current NFS and SMB protocols does not provide network level support for file notifications, and neither do the /proc and /sys virtual filesystems.
This could be fixed with a polling watcher (#9), but it's not yet implemented.
Why do I get many Chmod events?
Some programs may generate a lot of attribute changes; for example Spotlight on macOS, anti-virus programs, backup applications, and some others are known to do this. As a rule, it's typically best to ignore Chmod events. They're often not useful, and tend to cause problems.
Spotlight indexing on macOS can result in multiple events (see #15). A temporary workaround is to add your folder(s) to the Spotlight Privacy settings until we have a native FSEvents implementation (see #11).
Watching a file doesn't work well
Watching individual files (rather than directories) is generally not recommended as many programs (especially editors) update files atomically: it will write to a temporary file which is then moved to to destination, overwriting the original (or some variant thereof). The watcher on the original file is now lost, as that no longer exists.
The upshot of this is that a power failure or crash won't leave a half-written file.
Watch the parent directory and use Event.Name to filter out files you're not
interested in. There is an example of this in cmd/fsnotify/file.go.
Platform-specific notes
Linux
When a file is removed a REMOVE event won't be emitted until all file descriptors are closed; it will emit a CHMOD instead:
fp := os.Open("file")
os.Remove("file") // CHMOD
fp.Close() // REMOVE
This is the event that inotify sends, so not much can be changed about this.
The fs.inotify.max_user_watches sysctl variable specifies the upper limit for
the number of watches per user, and fs.inotify.max_user_instances specifies
the maximum number of inotify instances per user. Every Watcher you create is an
"instance", and every path you add is a "watch".
These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
/proc/sys/fs/inotify/max_user_instances
To increase them you can use sysctl or write the value to proc file:
# The default values on Linux 5.18
sysctl fs.inotify.max_user_watches=124983
sysctl fs.inotify.max_user_instances=128
To make the changes persist on reboot edit /etc/sysctl.conf or
/usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check your
distro's documentation):
fs.inotify.max_user_watches=124983
fs.inotify.max_user_instances=128
Reaching the limit will result in a "no space left on device" or "too many open files" error.
kqueue (macOS, all BSD systems)
kqueue requires opening a file descriptor for every file that's being watched; so if you're watching a directory with five files then that's six file descriptors. You will run in to your system's "max open files" limit faster on these platforms.
The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
control the maximum number of open files.
Documentation
¶
Overview ¶
Package fsnotify provides a cross-platform interface for file system notifications.
Currently supported systems:
Linux 2.6.32+ via inotify BSD, macOS via kqueue Windows via ReadDirectoryChangesW illumos via FEN
Index ¶
Constants ¶
This section is empty.
Variables ¶
var ( // ErrNonExistentWatch is used when Remove() is called on a path that's not // added. ErrNonExistentWatch = errors.New("fsnotify: can't remove non-existent watch") // ErrClosed is used when trying to operate on a closed Watcher. ErrClosed = errors.New("fsnotify: watcher already closed") // ErrEventOverflow is reported from the Errors channel when there are too // many events: // // - inotify: There are too many queued events (fs.inotify.max_queued_events sysctl) // - windows: The buffer size is too small; WithBufferSize() can be used to increase it. // - kqueue, fen: Not used. ErrEventOverflow = errors.New("fsnotify: queue or buffer overflow") )
Functions ¶
func WithBufferSize ¶ added in v1.7.0
func WithBufferSize(bytes int) addOpt
WithBufferSize sets the ReadDirectoryChangesW buffer size.
This only has effect on Windows systems, and is a no-op for other backends.
The default value is 64K (65536 bytes) which is the highest value that works on all filesystems and should be enough for most applications, but if you have a large burst of events it may not be enough. You can increase it if you're hitting "queue or buffer overflow" errors (ErrEventOverflow).
Types ¶
type Event ¶ added in v1.0.0
type Event struct {
// Path to the file or directory.
//
// Paths are relative to the input; for example with Add("dir") the Name
// will be set to "dir/file" if you create that file, but if you use
// Add("/path/to/dir") it will be "/path/to/dir/file".
Name string
// File operation that triggered the event.
//
// This is a bitmask and some systems may send multiple operations at once.
// Use the Event.Has() method instead of comparing with ==.
Op Op
}
Event represents a file system notification.
type Op ¶ added in v1.0.0
type Op uint32
Op describes a set of file operations.
const ( // A new pathname was created. Create Op = 1 << iota // The pathname was written to; this does *not* mean the write has finished, // and a write can be followed by more writes. Write // The path was removed; any watches on it will be removed. Some "remove" // operations may trigger a Rename if the file is actually moved (for // example "remove to trash" is often a rename). Remove // The path was renamed to something else; any watched on it will be // removed. Rename // File attributes were changed. // // It's generally not recommended to take action on this event, as it may // get triggered very frequently by some software. For example, Spotlight // indexing on macOS, anti-virus software, backup software, etc. Chmod )
The operations fsnotify can trigger; see the documentation on Watcher for a full description, and check them with Event.Has.
type Watcher ¶
type Watcher struct {
// Events sends the filesystem change events.
//
// fsnotify can send the following events; a "path" here can refer to a
// file, directory, symbolic link, or special file like a FIFO.
//
// fsnotify.Create A new path was created; this may be followed by one
// or more Write events if data also gets written to a
// file.
//
// fsnotify.Remove A path was removed.
//
// fsnotify.Rename A path was renamed. A rename is always sent with the
// old path as Event.Name, and a Create event will be
// sent with the new name. Renames are only sent for
// paths that are currently watched; e.g. moving an
// unmonitored file into a monitored directory will
// show up as just a Create. Similarly, renaming a file
// to outside a monitored directory will show up as
// only a Rename.
//
// fsnotify.Write A file or named pipe was written to. A Truncate will
// also trigger a Write. A single "write action"
// initiated by the user may show up as one or multiple
// writes, depending on when the system syncs things to
// disk. For example when compiling a large Go program
// you may get hundreds of Write events, and you may
// want to wait until you've stopped receiving them
// (see the dedup example in cmd/fsnotify).
//
// Some systems may send Write event for directories
// when the directory content changes.
//
// fsnotify.Chmod Attributes were changed. On Linux this is also sent
// when a file is removed (or more accurately, when a
// link to an inode is removed). On kqueue it's sent
// when a file is truncated. On Windows it's never
// sent.
Events chan Event
// Errors sends any errors.
Errors chan error
// contains filtered or unexported fields
}
Watcher watches a set of paths, delivering events on a channel.
A watcher should not be copied (e.g. pass it by pointer, rather than by value).
Linux notes ¶
When a file is removed a Remove event won't be emitted until all file descriptors are closed, and deletes will always emit a Chmod. For example:
fp := os.Open("file")
os.Remove("file") // Triggers Chmod
fp.Close() // Triggers Remove
This is the event that inotify sends, so not much can be changed about this.
The fs.inotify.max_user_watches sysctl variable specifies the upper limit for the number of watches per user, and fs.inotify.max_user_instances specifies the maximum number of inotify instances per user. Every Watcher you create is an "instance", and every path you add is a "watch".
These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and /proc/sys/fs/inotify/max_user_instances
To increase them you can use sysctl or write the value to the /proc file:
# Default values on Linux 5.18 sysctl fs.inotify.max_user_watches=124983 sysctl fs.inotify.max_user_instances=128
To make the changes persist on reboot edit /etc/sysctl.conf or /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check your distro's documentation):
fs.inotify.max_user_watches=124983 fs.inotify.max_user_instances=128
Reaching the limit will result in a "no space left on device" or "too many open files" error.
kqueue notes (macOS, BSD) ¶
kqueue requires opening a file descriptor for every file that's being watched; so if you're watching a directory with five files then that's six file descriptors. You will run in to your system's "max open files" limit faster on these platforms.
The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to control the maximum number of open files, as well as /etc/login.conf on BSD systems.
Windows notes ¶
Paths can be added as "C:\path\to\dir", but forward slashes ("C:/path/to/dir") will also work.
When a watched directory is removed it will always send an event for the directory itself, but may not send events for all files in that directory. Sometimes it will send events for all times, sometimes it will send no events, and often only for some files.
The default ReadDirectoryChangesW() buffer size is 64K, which is the largest value that is guaranteed to work with SMB filesystems. If you have many events in quick succession this may not be enough, and you will have to use WithBufferSize to increase the value.
func NewBufferedWatcher ¶ added in v1.7.0
NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events channel.
The main use case for this is situations with a very large number of events where the kernel buffer size can't be increased (e.g. due to lack of permissions). An unbuffered Watcher will perform better for almost all use cases, and whenever possible you will be better off increasing the kernel buffers instead of adding a large userspace buffer.
func (*Watcher) Add ¶ added in v1.0.0
Add starts monitoring the path for changes.
A path can only be watched once; watching it more than once is a no-op and will not return an error. Paths that do not yet exist on the filesystem cannot be watched.
A watch will be automatically removed if the watched path is deleted or renamed. The exception is the Windows backend, which doesn't remove the watcher on renames.
Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special filesystems (/proc, /sys, etc.) generally don't work.
Returns ErrClosed if Watcher.Close was called.
See Watcher.AddWith for a version that allows adding options.
Watching directories ¶
All files in a directory are monitored, including new files that are created after the watcher is started. Subdirectories are not watched (i.e. it's non-recursive).
Watching files ¶
Watching individual files (rather than directories) is generally not recommended as many programs (especially editors) update files atomically: it will write to a temporary file which is then moved to to destination, overwriting the original (or some variant thereof). The watcher on the original file is now lost, as that no longer exists.
The upshot of this is that a power failure or crash won't leave a half-written file.
Watch the parent directory and use Event.Name to filter out files you're not interested in. There is an example of this in cmd/fsnotify/file.go.
func (*Watcher) AddWith ¶ added in v1.7.0
AddWith is like Watcher.Add, but allows adding options. When using Add() the defaults described below are used.
Possible options are:
- WithBufferSize sets the buffer size for the Windows backend; no-op on other platforms. The default is 64K (65536 bytes).
func (*Watcher) Remove ¶ added in v1.0.0
Remove stops monitoring the path for changes.
Directories are always removed non-recursively. For example, if you added /tmp/dir and /tmp/dir/subdir then you will need to remove both.
Removing a path that has not yet been added returns ErrNonExistentWatch.
Returns nil if Watcher.Close was called.
func (*Watcher) WatchList ¶ added in v1.7.0
WatchList returns all paths explicitly added with Watcher.Add (and are not yet removed).
Returns nil if Watcher.Close was called.
Source Files
Directories