Browse Source

Expand the preload docs (#1299)

nfsdebug
trapexit 11 months ago
committed by GitHub
parent
commit
d68ad9ac01
No known key found for this signature in database GPG Key ID: B5690EEEBB952194
  1. 82
      README.md
  2. 77
      man/mergerfs.1

82
README.md

@ -1292,17 +1292,38 @@ typedef char IOCTL_BUF[4096];
EXPERIMENTAL
This preloadable library overrides the creation and opening of files
in order to simulate passthrough file IO. It catches the
open/creat/fopen calls, lets mergerfs do the call, queries mergerfs
for the branch the file exists on, and reopens the file on the underlying
filesystem. Meaning that you will get native read/write performance.
This will only work on dynamically linked software. Anything
statically compiled will not work. Many GoLang and Rust apps are
statically compiled.
The library will not interfere with non-mergerfs filesystems.
For some time there has been work to enable passthrough IO in
FUSE. Passthrough IO would allow for near native performance with
regards to reads and writes (at the expense of certain mergerfs
features.) However, there have been several complications which have
kept the feature from making it into the mainline Linux kernel. Until
that feature is available there are two methods to provide similar
functionality. One method is using the LD_PRELOAD feature of the
dynamic linker. The other leveraging ptrace to intercept
syscalls. Each has their disadvantages. At the moment only a preload
based tool is available. A ptrace based tool may be developed later if
there is a need.
`/usr/lib/mergerfs/preload.so`
This [preloadable
library](https://man7.org/linux/man-pages/man8/ld.so.8.html#ENVIRONMENT)
overrides the creation and opening of files in order to simulate
passthrough file IO. It catches the open/creat/fopen calls, has
mergerfs do the call, queries mergerfs for the branch the file exists
on, reopens the file on the underlying filesystem and returns that
instead. Meaning that you will get native read/write performance
because mergerfs is no longer part of the workflow. Keep in mind that
this also means certain mergerfs features that work by interrupting
the read/write workflow, such as `moveonenospc`, will no longer work.
Also understand that this will only work on dynamically linked
software. Anything statically compiled will not work. Many GoLang and
Rust apps are statically compiled.
The library will not interfere with non-mergerfs filesystems. The
library is written to always fallback to returning the mergerfs opened
file on error.
While the library was written to account for a number of edgecases
there could be some yet accounted for so please report any oddities.
@ -1314,28 +1335,53 @@ prototyping the idea.
### general usage
```
```sh
LD_PRELOAD=/usr/lib/mergerfs/preload.so touch /mnt/mergerfs/filename
```
### Docker usage
Assume `/mnt/fs0` and `/mnt/fs1` are pooled with mergerfs at
`/mnt/mergerfs`.
Assume `/mnt/fs0` and `/mnt/fs1` are pooled with mergerfs at `/media`.
Remember that you must bind into the container the original host paths
to the same locations otherwise the preload module will not be able to
find the files.
All mergerfs branch paths *must* be bind mounted into the container at
the same path as found on the host so the preload library can see them.
```
```sh
docker run \
-e LD_PRELOAD=/usr/lib/mergerfs/preload.so \
-v /usr/lib/mergerfs/preload.so:/usr/lib/mergerfs/preload.so:ro \
-v /media:/data \
-v /mnt:/mnt \
ubuntu:latest \
bash
```
or more explicitly
```sh
docker run \
-e LD_PRELOAD=/usr/lib/mergerfs/preload.so \
-v /usr/lib/mergerfs/preload.so:/usr/lib/mergerfs/preload.so:ro \
-v /media:/data \
-v /mnt/fs0:/mnt/fs0 \
-v /mnt/fs1:/mnt/fs1 \
ubuntu:latest \
bash
```
### systemd unit
Use the `Environment` option to set the LD_PRELOAD variable.
* https://www.freedesktop.org/software/systemd/man/latest/systemd.service.html#Command%20lines
* https://serverfault.com/questions/413397/how-to-set-environment-variable-in-systemd-service
```
[Service]
Environment=LD_PRELOAD=/usr/lib/mergerfs/preload.so
```
## Misc
* https://github.com/trapexit/mergerfs-tools

77
man/mergerfs.1

@ -1726,18 +1726,41 @@ unused files to be released from memory.
.PP
EXPERIMENTAL
.PP
This preloadable library overrides the creation and opening of files in
order to simulate passthrough file IO.
It catches the open/creat/fopen calls, lets mergerfs do the call,
queries mergerfs for the branch the file exists on, and reopens the file
on the underlying filesystem.
Meaning that you will get native read/write performance.
.PP
This will only work on dynamically linked software.
For some time there has been work to enable passthrough IO in FUSE.
Passthrough IO would allow for near native performance with regards to
reads and writes (at the expense of certain mergerfs features.) However,
there have been several complications which have kept the feature from
making it into the mainline Linux kernel.
Until that feature is available there are two methods to provide similar
functionality.
One method is using the LD_PRELOAD feature of the dynamic linker.
The other leveraging ptrace to intercept syscalls.
Each has their disadvantages.
At the moment only a preload based tool is available.
A ptrace based tool may be developed later if there is a need.
.PP
\f[C]/usr/lib/mergerfs/preload.so\f[R]
.PP
This preloadable
library (https://man7.org/linux/man-pages/man8/ld.so.8.html#ENVIRONMENT)
overrides the creation and opening of files in order to simulate
passthrough file IO.
It catches the open/creat/fopen calls, has mergerfs do the call, queries
mergerfs for the branch the file exists on, reopens the file on the
underlying filesystem and returns that instead.
Meaning that you will get native read/write performance because mergerfs
is no longer part of the workflow.
Keep in mind that this also means certain mergerfs features that work by
interrupting the read/write workflow, such as \f[C]moveonenospc\f[R],
will no longer work.
.PP
Also understand that this will only work on dynamically linked software.
Anything statically compiled will not work.
Many GoLang and Rust apps are statically compiled.
.PP
The library will not interfere with non-mergerfs filesystems.
The library is written to always fallback to returning the mergerfs
opened file on error.
.PP
While the library was written to account for a number of edgecases there
could be some yet accounted for so please report any oddities.
@ -1754,22 +1777,52 @@ LD_PRELOAD=/usr/lib/mergerfs/preload.so touch /mnt/mergerfs/filename
.SS Docker usage
.PP
Assume \f[C]/mnt/fs0\f[R] and \f[C]/mnt/fs1\f[R] are pooled with
mergerfs at \f[C]/mnt/mergerfs\f[R].
mergerfs at \f[C]/media\f[R].
.PP
Remember that you must bind into the container the original host paths
to the same locations otherwise the preload module will not be able to
find the files.
All mergerfs branch paths \f[I]must\f[R] be bind mounted into the
container at the same path as found on the host so the preload library
can see them.
.IP
.nf
\f[C]
docker run \[rs]
-e LD_PRELOAD=/usr/lib/mergerfs/preload.so \[rs]
-v /usr/lib/mergerfs/preload.so:/usr/lib/mergerfs/preload.so:ro \[rs]
-v /media:/data \[rs]
-v /mnt:/mnt \[rs]
ubuntu:latest \[rs]
bash
\f[R]
.fi
.PP
or more explicitly
.IP
.nf
\f[C]
docker run \[rs]
-e LD_PRELOAD=/usr/lib/mergerfs/preload.so \[rs]
-v /usr/lib/mergerfs/preload.so:/usr/lib/mergerfs/preload.so:ro \[rs]
-v /media:/data \[rs]
-v /mnt/fs0:/mnt/fs0 \[rs]
-v /mnt/fs1:/mnt/fs1 \[rs]
ubuntu:latest \[rs]
bash
\f[R]
.fi
.SS systemd unit
.PP
Use the \f[C]Environment\f[R] option to set the LD_PRELOAD variable.
.IP \[bu] 2
https://www.freedesktop.org/software/systemd/man/latest/systemd.service.html#Command%20lines
.IP \[bu] 2
https://serverfault.com/questions/413397/how-to-set-environment-variable-in-systemd-service
.IP
.nf
\f[C]
[Service]
Environment=LD_PRELOAD=/usr/lib/mergerfs/preload.so
\f[R]
.fi
.SS Misc
.IP \[bu] 2
https://github.com/trapexit/mergerfs-tools

Loading…
Cancel
Save