msem

Provides a convenient interface for creating and controlling shared semaphores in the filesystem. There are only three primary operations, lock, unlock, and relax (unlock all).

Requires System V UNIX semaphores, which are not POSIX-compliant but which ship with the Linux kernel.

System V Semaphores

Need to provide synchronized interactive behavior to a number of distributed clients without wasting the resources of the server? You think epoll, or websockets -- what about System V semaphores?

System V semaphores can support a form of kernel queue, in which locked processes are put to sleep until they are popped from the queue. This allows subscribers to an event feed to remain at the server until a timeout occurs, or until a controller process (typically another web request which alters state) flushes the semaphore and unwinds the queue.

This provides a simple way to coordinate clients in a long-poll based solution, and has been validated in years of production use.

Advantages

• Naturally enforces queue behavior
• Processes sleep when they are waiting in the queue
• Provides a timeout after which a process will disappear from the queue
• Performs all operations in kernel space without a context switch

Disadvantages

• Unless shared memory is used, the file descriptor must be on a single server
• Latency for waking processes depending on configuration
• Not POSIX-compliant

Semaphore files

A semaphore file can locate multiple semaphores, each referenced by a unique identifier (tag), usually an ASCII character. If the file does not exist, it will be created. If the file does exist, and does not contain a semaphore with the specified tag, a new semaphore will be created inside the file.

Semaphore files are created and stored in /tmp/sem_*.

Usage:

    SYNOPSIS
    msem

    -c, --create <path> [uid] [initial_value]
            Create a new semaphore

    -d, --delete <path> [uid]
            Delete an existing semaphore

    -l, -p, --lock <path> [uid] [timeout]
            Lock a semaphore by decrementing its value by 1

    -u, -v, --unlock <path> [uid]
            Unlock a semaphore by incrementing its value by 1.
            The next process in the waiting queue will be able
            to proceed.

    -u+, -v+, --relax <path> [uid]
            Relax a semaphore. Any processes which have locked
            the semaphore will be released, as though the semaphore
            had been incremented for all of them,

    -f --follow <path> [uid]
            Continuously print the status of a semaphore to stdout
            (similar to tail -f)