CannyFS

CannyFS is a shim file system. That is, it mirrors an existing file system. This is implemented using FUSE, and the source code is a heavily modified fork of the https://github.com/libfuse/libfuse example https://github.com/libfuse/libfuse/blob/master/example/passthrough_fh.c (which used to be called fusexmp_fh.c).

What makes CannyFS different is that almost any write operation is completed asynchronously, while the code itself reports back to the caller that the task completed. This means that the calling process can proceed doing useful work. This is especially important if your I/O subsystem has high latency (due to lots of I/O or being hosted somewhere else within a network), and/or if your process makes a very high number of I/O requests, with a lot of flushing or writing to different files. Examples of this can be tasks that walk over complete directory trees, touching or writing to every file within them.

Preprint paper (not currently submitted anywhere) with some benchmarks: https://arxiv.org/abs/1612.06830

Intended usage mode

1. Mount a directory with CannyFS, in non-demon mode to easily see error messages written to stderr.
2. Do your work.
3. Kill the CannyFS process.
4. Check that the CannyFS process gracefully reports no errors.

Compiling CannyFS

The packages tbb, boost, and fuse are needed, beyond what's typically available in any Linux distro. In e.g. Ubuntu 16.04, this can be enough to get you going:

sudo apt-get install libtbb-dev libfuse-dev libboost-dev libboost-filesystem-dev libboost-system-dev

Then compile using at the very least a g++ compiler from the 5.x tree, 6.x highly recommended (C++ 14 support is needed).

g++ cannyfs.cpp -std=c++14 -O3 -lfuse -ltbb -lpthread -lboost_filesystem -lboost_system -D_FILE_OFFSET_BITS=64 -o cannyfs

Example script

The following script will create a mount that mirrors your local dir, with settings that are suitable for a Linux system (where default pipe buffers are typically 65536 bytes in length). The zip file archive.zip contains loads of small files and thus takes quite long to extract, especially if you are doing this over an NFS or CIFS mount. By mounting it in CannyFS, unzip can enqueue I/O operations to several target files, rather than performing a blocking wait for completion for each file.

CannyFS will need to be in your path, if it's found locally, adjust the command and kill jobspec to ./cannyfs:

#!/usr/bin/bash
mkdir mountpoint
cannyfs -f -o big_writes -o max_write=65536 -omodules=subdir,subdir=$HOME mountpoint &
# Linux-specific way of checking for mountpoint, arbitrary sleep generally works fine as well
until mountpoint -q $CANNY_PATH; do sleep 1; done
cd mountpoint
unzip $HOME/archive.zip
kill %cannyfs
rmdir mountpoint

You can also use the handy cannywrapper.sh script, assuming cannyfs is in your path. That way, you can wrap a single shell command in a CannyFS jail, where any file system I/O is tunneled through CannyFS. This is less than ideal since some operations to special file systems are not compatible, but it is fine for common I/O workhorse tools such as `rsync'.