osxfs is a new shared file system solution, exclusive to Docker for Mac.
osxfs provides a close-to-native user experience for bind mounting OS X file
system trees into Docker containers. To this end,
osxfs features a number of
unique capabilities as well as differences from a classical Linux file system.
With Docker for Mac, file systems are shared from OS X into containers in the same way as they operate in OS X. As a result, if a file system on OS X is case-insensitive that behavior is shared by any bind mount from OS X into a container. The default OS X file system is HFS+ and, during installation, it is installed as case-insensitive by default. To get case-sensitive behavior from your bind mounts, you must either create and format a ramdisk or external volume as HFS+ with case-sensitivity or reformat your OS root partition with HFS+ with case-sensitivity. We do not recommend reformatting your root partition as some Mac software dubiously relies on case-insensitivity to function.
osxfs, and therefore Docker, can access only those file system resources that
the Docker for Mac user has access to.
osxfs does not run as
root. If the OS
X user is an administrator,
osxfs inherits those administrator privileges. We
are still evaluating which privileges to drop in the file system process to
balance security and ease-of-use.
osxfs performs no additional permissions
checks and enforces no extra access control on accesses made through it. All
processes in containers can access the same objects in the same way as the
Docker user who started the containers.
Much of the OS X file system that is accessible to the user is also available to
containers using the
-v bind mount syntax. By default, you can share files in
/tmp directly. To add or remove
directory trees that are exported to Docker, use the File sharing tab in
Docker preferences -> Preferences ->
File sharing. (See Preferences.) All other paths
-v bind mounts are sourced from the Moby Linux VM running the Docker
containers, so arguments such as
should work as expected. If an OS X path is not shared and does not exist in the
VM, an attempt to bind mount it will fail rather than create it in the VM. Paths
that already exist in the VM and contain files are reserved by Docker and cannot
be exported from OS X.
Initially, any containerized process that requests ownership metadata of
an object is told that its
gid own the object. When any
containerized process changes the ownership of a shared file system
object, e.g. with
chown, the new ownership information is persisted in
com.docker.owner extended attribute of the object. Subsequent
requests for ownership metadata will return the previously set
values. Ownership-based permissions are only enforced at the OS X file
system level with all accessing processes behaving as the user running
Docker. If the user does not have permission to read extended attributes
on an object (such as when that object’s permissions are
will attempt to add an access control list (ACL) entry that allows
the user to read and write extended attributes. If this attempt
fails, the object will appear to be owned by the process accessing
it until the extended attribute is readable again.
inotify events are supported in bind mounts, and likely
fanotify (though they have not been tested) are also supported. This means
that file system events from OS X are sent into containers and trigger any
listening processes there.
The following are supported file system events:
The following are partially supported file system events:
IN_DELETEon the source of the rename and
IN_MODIFYon the destination of the rename
The following are unsupported file system events:
Some events may be delivered multiple times. These limitations do not apply to events between containers, only to those events originating in OS X.
The OS X mount structure is not visible in the shared volume, but volume contents are visible. Volume contents appear in the same file system as the rest of the shared file system. Mounting/unmounting OS X volumes that are also bind mounted into containers may result in unexpected behavior in those containers. Unmount events are not supported. Mount export support is planned but is still under development.
Symlinks are shared unmodified. This may cause issues when symlinks contain paths that rely on the default case-insensitivity of the default OS X file system, HFS+.
Symlinks, hardlinks, socket files, named pipes, regular files, and directories are supported. Socket files and named pipes only transmit between containers and between OS X processes – no transmission across the hypervisor is supported, yet. Character and block device files are not supported.
Extended attributes are not yet supported.
osxfs does not use OSXFUSE.
osxfs does not run under, inside, or
between OS X userspace processes and the OS X kernel.
With regard to reported performance issues (GitHub issue 77: File access in mounted volumes extremely slow), and a similar thread on Docker for Mac forums on topic: File access in mounted volumes extremely slow, this topic provides an explanation of the issues, what we are doing to address them, how the community can help us, and what you can expect in the future. This explanation is a slightly re-worked version of an understanding performance post from David Sheets (@dsheets) on the Docker development team to the forum topic just mentioned. We want to surface it in the documentation for wider reach.
Perhaps the most important thing to understand is that shared file system
performance is multi-dimensional. This means that, depending on your workload,
you may experience exceptional, adequate, or poor performance with
file system server in Docker for Mac. File system APIs are very wide (20-40
message types) with many intricate semantics involving on-disk state, in-memory
cache state, and concurrent access by multiple processes. Additionally,
integrates a mapping between OS X’s FSEvents API and Linux’s inotify API
which is implemented inside of the file system itself complicating matters
further (cache behavior in particular).
At the highest level, there are two dimensions to file system performance:
throughput (read/write IO) and latency (roundtrip time). In a traditional file
system on a modern SSD, applications can generally expect throughput of a few
GB/s. With large sequential IO operations,
osxfs can achieve throughput of
around 250 MB/s which, while not native speed, will not be the bottleneck for
most applications which perform acceptably on HDDs.
Latency is the time it takes for a file system call to complete. For instance,
the time between a thread issuing write in a container and resuming with the
number of bytes written. With a classical block-based file system, this latency
is typically under 10μs (microseconds). With
osxfs, latency is presently
around 200μs for most operations or 20x slower. For workloads which demand many
sequential roundtrips, this results in significant observable slowdown. To
reduce the latency, we need to shorten the data path from a Linux system call to
OS X and back again. This requires tuning each component in the data path in
turn – some of which require significant engineering effort. Even if we achieve
a huge latency reduction of 100μs/roundtrip, we will still “only” see a doubling
of performance. This is typical of performance engineering, which requires
significant effort to analyze slowdowns and develop optimized components. We
know how we can likely halve the roundtrip time but we haven’t implemented those
improvements yet (more on this below in
What you can do).
There is hope for significant performance improvement in the near term despite these fundamental communication channel properties, which are difficult to overcome (latency in particular). This hope comes in the form of increased caching (storing “recent” values closer to their use to prevent roundtrips completely). The Linux kernel’s VFS layer contains a number of caches which can be used to greatly improve performance by reducing the required communication with the file system. Using this caching comes with a number of trade-offs:
It requires understanding the cache behavior in detail in order to write correct, stateful functionality on top of those caches.
It harms the coherence or consistency of the file system as observed from Linux containers and the OS X file system interfaces.
We are actively working on both increasing caching while mitigating the associated issues and on reducing the file system data path latency. This requires significant analysis of file system traces and speculative development of system improvements to try to address specific performance issues. Perhaps surprisingly, application workload can have a huge effect on performance. As an example, here are two different use cases contributed on the forum topic and how their performance differs and suffers due to latency, caching, and coherence:
A rake example (see below) appears to attempt to access 37000+ different files that don’t exist on the shared volume. We can work very hard to speed up all use cases by 2x via latency reduction but this use case will still seem “slow”. The ultimate solution for rake is to use a “negative dcache” that keeps track of, in the Linux kernel itself, the files that do not exist. Unfortunately, even this is not sufficient for the first time rake is run on a shared directory. To handle that case, we actually need to develop a Linux kernel patch which negatively caches all directory entries not in a specified set – and this cache must be kept up-to-date in real-time with the OS X file system state even in the presence of missing OS X FSEvents messages and so must be invalidated if OS X ever reports an event delivery failure.
Running ember build in a shared file system results in ember creating many different temporary directories and performing lots of intermediate activity within them. An empty ember project is over 300MB. This usage pattern does not require coherence between Linux and OS X but, because we cannot distinguish this fact at run-time, we maintain coherence during its hundreds of thousands of file system accesses to manipulate temporary state. There is no “correct” solution in this case. Either ember needs to change, the volume mount needs to have coherence properties specified on it somehow, some heuristic needs to be introduced to detect this access pattern and compensate, or the behavior needs to be indicated via, e.g., extended attributes in the OS X file system.
These two examples come from performance use cases contributed by users and they are incredibly helpful in prioritizing aspects of file system performance to improve. We are developing statistical file system trace analysis tools to characterize slow-performing workloads more easily in order to decide what to work on next.
Under development, we have:
A Linux kernel patch to reduce data path latency by 2/7 copies and 2/5 context switches
Increased OS X integration to reduce the latency between the hypervisor and the file system server
A server-side directory read cache to speed up traversal of large directories
User-facing file system tracing capabilities so that you can send us recordings of slow workloads for analysis
A growing performance test suite of real world use cases (more on this below in What you can do)
Experimental support for using Linux’s inode, writeback, and page caches
End-user controls to configure the coherence of subsets of cross-OS bind mounts without exposing all of the underlying complexity
When you report shared file system performance issues, it is most helpful to include a minimal Real World reproduction test case that demonstrates poor performance.
Without a reproduction, it is very difficult for us to analyze your use case and determine what improvements would speed it up. When you don’t provide a reproduction, one of us has to take the time to figure out the specific software you are using and guess and hope that we have configured it in a typical way or a way that has poor performance. That usually takes 1-4 hours depending on your use case and once it is done, we must then determine what regular performance is like and what kind of slow-down your use case is experiencing. In some cases, it is not obvious what operation is even slow in your specific development workflow. The additional set-up to reproduce the problem means we have less time to fix bugs, develop analysis tools, or improve performance. So, please include simple, immediate performance issue reproduction test cases. The rake reproduction case by @hirowatari shown in the forums thread is a great example.
This example originally provided:
A version-controlled repository so any changes/improvements to the test case can be easily tracked.
A Dockerfile which constructs the exact image to run
A command-line invocation of how to start the container
A straight-forward way to measure the performance of the use case
A clear explanation (README) of how to run the test case
We will continue to work toward an optimized shared file system implementation on the Beta channel of Docker for Mac.
You can expect some of the performance improvement work mentioned above to reach the Beta channel in the coming release cycles.
In due course, we will open source all of our shared file system components. At that time, we would be very happy to collaborate with you on improving the implementation of osxfs and related software.
We still have on the slate to write up and publish details of shared file system performance analysis and improvement on the Docker blog. Look for or nudge @dsheets about those articles, which should serve as a jumping off point for understanding the system, measuring it, or contributing to it.
We hope this gives you a rough idea of where
osxfs performance is and where
it’s going. We are treating good performance as a top priority feature of the
file system sharing component and we are actively working on improving it
through a number of different avenues. The osxfs project started in December
2015. Since the first integration into Docker for Mac in February 2016, we’ve
improved performance by 50x or more for many workloads while achieving nearly
complete POSIX compliance and without compromising coherence (it is shared and
not simply synced). Of course, in the beginning there was lots of low-hanging
fruit and now many of the remaining performance improvements require significant
engineering work on custom low-level components.
We appreciate your understanding as we continue development of the product and work on all dimensions of performance. We want to continue to work with the community on this, so please continue to report issues as you find them. We look forward to collaborating with you on ideas and on the source code itself.