Documentation
¶
Overview ¶
Package fdleak provides file descriptor leak detection.
It takes a snapshot of open file descriptors before and after running the tested function.
Any file descriptors present in the "after" snapshot but not in the "before" snapshot — and not of a filtered Kind — are considered leaks.
Platform support ¶
- Linux: enumerates /proc/self/fd and classifies FDs from the readlink target (socket:/pipe:/anon_inode:/ path).
- darwin: enumerates /dev/fd and resolves each FD via fcntl(F_GETPATH), falling back to fstat to classify sockets, pipes and kqueues.
- other: Snapshot returns an error.
Filtering ¶
Sockets, pipes and other kernel-internal descriptors (Linux anon_inode, darwin kqueue) are excluded from leak reports by default, as these are typically managed by the Go runtime or external libraries.
Concurrency ¶
This approach is inherently process-wide: the FD table lists all file descriptors for the process.
Any concurrent I/O from other goroutines may cause false positives. A mutex serializes Leaked calls to prevent multiple leak checks from interfering with each other, but cannot protect against external concurrent file operations.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func FormatLeaked ¶
FormatLeaked formats leaked file descriptors into a human-readable message. Returns the empty string if the slice is empty.
func Leaked ¶
Leaked takes a before/after snapshot around the tested function and returns a formatted description of leaked file descriptors.
Returns the empty string if no leaks are found. On unsupported platforms, Leaked returns an error.
func Snapshot ¶
Snapshot returns a map of currently open file descriptors for the running process.
The set of supported platforms is determined at build time; see the per-platform implementations (fdleak_linux.go, fdleak_darwin.go).
On unsupported platforms, Snapshot returns an error.
FDs that close between enumeration and resolution are silently skipped.
Types ¶
type FDInfo ¶
FDInfo describes an open file descriptor.
It should remain a human-readable description: a file-system path for vnode-backed FDs, or a synthetic label such as "socket:[<inode>]" for non-vnode kinds.
Kind is the authoritative classification; use it rather than parsing Target when filtering.
type Kind ¶ added in v2.5.0
type Kind int
Kind classifies an open file descriptor independently of platform-specific target-string conventions.
const ( // KindUnknown is used when the descriptor kind could not be determined. KindUnknown Kind = iota // KindFile denotes anything backed by a path in the file system: // regular files, directories, symlinks, block devices. KindFile // KindSocket denotes a socket (AF_UNIX, AF_INET, …). KindSocket // KindPipe denotes a pipe or FIFO. KindPipe // KindChar denotes a character device without a resolvable path // (fallback for darwin when F_GETPATH fails on a char device). KindChar // KindOther covers kernel-level descriptors that are not directly // opened by user code: Linux anon_inode (epoll, timerfd, …), darwin // kqueue, and similar. KindOther )
Source Files