Link statically #2085
Closed
Link statically #2085
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
meta-cla
bot
added
the
CLA Signed
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 9, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/) ghstack-source-id: 327971955
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 9, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328253183 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 10, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328257055 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 10, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328257535 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 10, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328260415 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 10, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328274952 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 10, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328276528 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 10, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328278142 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
zdevito
added a commit
that referenced
this pull request
Dec 10, 2025
Pull Request resolved: #2085 Link everything into monarch statically, or dlopen it. This makes the rdma related libraries and libc++ linked statically for increased portability. This means monarch built with rdma support and with a different version of libc++ than the current system can still be used. By building rdma-core from source, we can also make sure Dennis' 64-bit patch is applied. Total size of our so is 63MB now with no uncommon deps: ``` (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so linux-vdso.so.1 (0x00007f347a684000) libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007f3477200000) libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007f347a664000) libm.so.6 => /lib64/libm.so.6 (0x00007f3477125000) libc.so.6 => /lib64/libc.so.6 (0x00007f3476e00000) /lib64/ld-linux-x86-64.so.2 (0x00007f347a686000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f347a64a000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f347a645000) libutil.so.1 => /lib64/libutil.so.1 (0x00007f347a640000) (monarch) [zdevito@devgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ du -h ./python/monarch/_rust_bindings.so 63M ./python/monarch/_rust_bindings.so ``` ghstack-source-id: 328288065 Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
Closed
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
```
(monarch) [zdevitodevgpu014 /data/users/zdevito/fbsource/fbcode/monarch]$ ldd ./python/monarch/_rust_bindings.so
linux-vdso.so.1 (0x00007ffdabdf5000)
libpython3.11.so.1.0 => /home/zdevito/local/miniconda3/envs/monarch/lib/libpython3.11.so.1.0 (0x00007fe1fae00000)
libgcc_s.so.1 => /home/zdevito/local/miniconda3/envs/monarch/lib/libgcc_s.so.1 (0x00007fe201a0f000)
libm.so.6 => /lib64/libm.so.6 (0x00007fe201921000)
libc.so.6 => /lib64/libc.so.6 (0x00007fe1faa00000)
/lib64/ld-linux-x86-64.so.2 (0x00007fe201a2b000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fe20191a000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007fe201915000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007fe201910000)
```
Differential Revision: [D88540873](https://our.internmc.facebook.com/intern/diff/D88540873/)
[ghstack-poisoned]
|
This pull request has been merged in be3d6bc. |
This was referenced Dec 10, 2025
Closed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
Draft of how we can link everything into monarch statically.
After looking at binary size it is pretty clear we want to dynamically link libnccl (it can be big if it supports many arches). So I will change that to an approach similar to libcuda where we load the needed functions dynamically.
Differential Revision: D88540873