While designing something else when TCPStore is needed. I spent some time digging into the codebase of TCPStore and found that the code is a little bit challenging to understand without proper documents. Although people from OSS community must be smarter than me, I still want to document my findings in the code so that devs and users can use them as a reference down the road.
Also for libuv, we need to make private variables with a "_", so it's a pure renaming of private variables such as `tcpServer`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130496
Approved by: https://github.com/wconstab
This replaces the existing TCPStore counters with the new shared wait counters. There's no users of the tcpstore counters so should be completely safe to remove.
Test plan:
Existing tests + build
There's no OSS backend for wait counters so can't write any tests with them currently.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/135283
Approved by: https://github.com/c-p-i-o
This does a round trip request on socket connect -- this allows for detecting connection resets etc and retrying before the non-retryable application requests are sent.
This adds support for PING to both the libuv and legacy backend.
Example error:
```
[trainer85612|12]:W0701 13:41:43.421574 4776 TCPStore.cpp:182] [c10d] recvValue failed on SocketImpl(fd=24, ...): Connection reset by peer
[trainer85612|12]:Exception raised from recvBytes at /mnt/code/pytorch/torch/csrc/distributed/c10d/Utils.hpp:669 (most recent call first):
...
[trainer85612|12]:#9 c10d::TCPStore::incrementValueBy(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, long) from /packages/.../conda/lib/python3.10/site-packages/torch/lib/libtorch_cpu.so:84809637
[trainer85612|12]:#10 c10d::TCPStore::waitForWorkers() from /packages/.../conda/lib/python3.10/site-packages/torch/lib/libtorch_cpu.so:84812868
[trainer85612|12]:#11 c10d::TCPStore::TCPStore(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, c10d::TCPStoreOptions const&) from /packages/.../conda/lib/python3.10/site-packages/torch/lib/libtorch_cpu.so:84814775
```
Test plan:
```
python test/distributed/test_store.py -v
```
```
tristanr@devvm4382 ~/pytorch (d4l3k/tcpstore_ping)> python ~/pt_tests/tcpstore_large_test.py
starting pool
started 90000
started 30000
started 70000
started 20000
started 80000
started 60000
started 0
[W702 16:16:25.301681870 TCPStore.cpp:343] [c10d] Starting store with 100000 workers but somaxconn is 4096.This might cause instability during bootstrap, consider increasing it.
init 20000
set 20000
init 80000
set 80000
init 70000
set 70000
init 60000
set 60000
init 30000
set 30000
init 90000
set 90000
started 40000
init 40000
set 40000
started 50000
init 50000
set 50000
started 10000
init 10000
set 10000
init 0
set 0
run finished 617.2992351055145
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129985
Approved by: https://github.com/rsdcastro, https://github.com/kurman
We've been facing issues where TCPStore can successfully connect but then fail in the validate() function due to resets from listen backlog queue overflow when combined with reset enabled as well as long init times.
This PR does a few things:
* Retry that connect and validate up to the specified timeout.
* Use exponential backoff for the retry logic with jitter instead of a fixed 1s sleep.
* Eliminate the `sleep(std::chrono::milliseconds(numWorkers))` on init which can add significant delays to startup. This is no longer necessary per @XilunWu https://github.com/pytorch/pytorch/pull/116141
Test plan:
```
python test/distributed/test_store.py -v
./build/bin/BackoffTest
```
Will do internal testing with some large scale jobs to ensure TCPStore works correctly.
At 4k scale: 4x improvement
```
tristanr@devvm4382 ~/pt_tests [SIGABRT]> time TORCH_SHOW_CPP_STACKTRACES=1 python tcpstore_large_test.py (pytorch-3.10)
started 0
init 0
set 0
joined all
________________________________________________________
Executed in 1.98 secs fish external
usr time 0.93 secs 91.00 micros 0.93 secs
sys time 1.98 secs 954.00 micros 1.97 secs
tristanr@devvm4382 ~/pt_tests> conda activate torchdrive-3.10 (pytorch-3.10)
tristanr@devvm4382 ~/pt_tests> time TORCH_SHOW_CPP_STACKTRACES=1 python tcpstore_large_test.py (torchdrive-3.10)
started 0
init 0
set 0
joined all
________________________________________________________
Executed in 8.20 secs fish external
usr time 2.15 secs 0.00 micros 2.15 secs
sys time 2.76 secs 843.00 micros 2.76 secs
```
```py
import time
import os
import threading
from multiprocessing import Pool
WORLD_SIZE = 10000
import torch.distributed as dist
def run(rank):
should_log = rank % (WORLD_SIZE // 10) == 0
if should_log:
print(f"started {rank}")
store = dist.TCPStore(
host_name="devvm4382.nao0.facebook.com",
port=29500,
world_size=WORLD_SIZE,
is_master=rank == 0,
use_libuv=True,
)
if should_log:
print(f"init {rank}")
store.set(f"key{rank}", "1234")
if should_log:
print(f"set {rank}")
del store
def noop(rank):
pass
print("starting pool")
with Pool(WORLD_SIZE) as pool:
pool.map(noop, range(WORLD_SIZE), 1)
print("pool hot")
start = time.time()
pool.map(run, range(WORLD_SIZE), 1)
print("run finished", time.time()-start)
```
```
tristanr@devvm4382 ~/pt_tests> python tcpstore_large_test.py (pytorch-3.10)
starting pool
pool hot
started 0
[W624 16:58:09.086081750 TCPStore.cpp:343] [c10d] Starting store with 10000 workers but somaxconn is 4096.This might cause instability during bootstrap, consider increasing it.
started 1000
init 1000
set 1000
started 2000
init 2000
set 2000
started 3000
init 3000
set 3000
started 4000
init 4000
set 4000
started 5000
init 5000
set 5000
started 6000
init 6000
set 6000
started 7000
init 7000
set 7000
started 8000
init 8000
set 8000
started 9000
init 9000
set 9000
init 0
set 0
run finished 0.705092191696167
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129261
Approved by: https://github.com/rsdcastro, https://github.com/wconstab, https://github.com/kurman, https://github.com/XilunWu, https://github.com/c-p-i-o
**Summary**
This PR switches the default TCPStore server backend to a new implementation that utilizes [`libuv`](https://github.com/libuv/libuv) for significantly lower initialization time and better scalability:
<img width="714" alt="image" src="https://github.com/pytorch/pytorch/assets/12968408/18503011-da5d-4104-8ba9-abc456438b02">
We hope this improvement would benefit users from a much shorter startup time in large-scale jobs. Eventually, we hope to fully replace the old TCPStore backend implementation with the libuv one.
**What it changes**
This PR changes the underlying TCPStore server backend to `libuv` if users don't explicitly specify to use the old TCPStore server. This change is not supposed to cause any user notice except significant faster TCPStore startup for large-scale jobs.
One thing to note is, we do not support the initialization approach where user passes in a socket for libuv backend. We plan to support it as a next step but we choose to disable it before fully testing. If you are initializing TCPStore in this approach, you can see the next section to remain using the old TCPStore server.
**Fallback/Remain using the old TCPStore server**
For users who want to stay with the old TCPStore backend, there're 3 ways:
1. If user is directly instantiating TCPStore object, user can pass in argument `use_libuv=False` to use the old TCPStore server backend e.g. `store = torch.distributed.TCPStore(..., use_libuv=False)`.
2. Or, specify the TCPStore backend option in `init_method` when calling default ProcessGroup init, e.g. `torch.distributed.init_process_group(..., init_method="{YOUR_RENDEZVOUS_METHOD}://{YOUR_HOSTNAME}:{YOUR_PORT}?use_libuv=0")`
3. Or, user can set environment variable `USE_LIBUV` to `"0"` when launching.
These 3 approach are in order of precedence. That being said, if user specifies `use_libuv=0` in `init_method` and also sets environment var `USE_LIBUV="1"`, the former will take effect and the TCPStore backend instantiated will be the old one instead of the one using libuv.
**Operating Systems Compatibility**
From the CI signals, we believe the new implementation has the same behavior as the old TCPStore server on all supported platforms. If you notice any behavior discrepancy, please file an issue with `oncall: distributed` label.
**Test Plan**
`pytest test/distributed/test_store.py`
<img width="2548" alt="image" src="https://github.com/pytorch/pytorch/assets/12968408/dc0aebeb-6d5a-4daa-b98c-e56bd39aa588">
note: `TestMultiThreadedWait::test_wait` is a broken test that has been there for some time.
`test/distributed/elastic/utils/distributed_test.py`
<img width="2558" alt="image" src="https://github.com/pytorch/pytorch/assets/12968408/a6a3266d-b798-41c4-94d2-152056a034f6">
**TODO**
1. Update the doc at
- https://pytorch.org/docs/stable/distributed.html#distributed-key-value-store
- https://pytorch.org/docs/stable/distributed.html#tcp-initialization
2. Make torch elastic rendezvous to use libuv TCPStore as well. See `torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py` cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang @kwen2501 @awgu @penguinwu @fegin @wanchaol @fduwjj @wz337 @tianyu-l @wconstab @yf225 @chauhang @d4l3k @kurman
3. Test if libuv backend is okay with initialization with socket. Change `LibUvTCPStoreTest::test_take_over_listen_socket`.
**Test Plan**
`pytest test/distributed/test_store.py`
<img width="2548" alt="image" src="https://github.com/pytorch/pytorch/assets/12968408/dc0aebeb-6d5a-4daa-b98c-e56bd39aa588">
note: `TestMultiThreadedWait::test_wait` is a broken test that has been there for some time.
`test/distributed/elastic/utils/distributed_test.py`
<img width="2558" alt="image" src="https://github.com/pytorch/pytorch/assets/12968408/a6a3266d-b798-41c4-94d2-152056a034f6">
Differential Revision: [D58259591](https://our.internmc.facebook.com/intern/diff/D58259591)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/127957
Approved by: https://github.com/kurman
ghstack dependencies: #127956
This PR continues to clean clang-tidy warnings in torch/csrc/distributed/c10d, following #124701. In addition, libfmt dependency is added in CMake code to enable using it in the headers. The libfmt has to be added as private dependency to torch_cuda and torch_hip because they include torch/csrc/distributed/c10d/Utils.hpp which uses libfmt.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/124987
Approved by: https://github.com/malfet
This PR fixes#106294.
Due to the lack of request validation mechanism, TCPStore in torch mistakenly treats nmap scan messages as valid query messages, which leads to DDP OOM. The simple solution enforces the very first query from a client is a validation query with a predefined magic number. If the validation fails, the server will terminate the connection.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107607
Approved by: https://github.com/cbalioglu, https://github.com/XilunWu
This PR fixes#106294.
Due to the lack of request validation mechanism, TCPStore in torch mistakenly treats nmap scan messages as valid query messages, which leads to DDP OOM. The simple solution enforces the very first query from a client is a validation query with a predefined magic number. If the validation fails, the server will terminate the connection.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107607
Approved by: https://github.com/cbalioglu, https://github.com/XilunWu
The feature was never fully finished and never got any adoption but
TCPStore pays the cost of twice the number of tcp connections anyway.
While the cost of all those idle connections is minimal is doesn't come for free:
- It increases the likelyhood of a connection refused failure during the initialization stampede.
- TCPStore uses poll for checking for socket availability which scales linearly on the number of sockets regardless of their status.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105014
Approved by: https://github.com/fduwjj
Summary:
This diff allows the `TCPStore` server associated with a gloo process group to listen on an existing socket already bound to a port.
Without the functionality in this diff, canonical initialization of a gloo `ProcessGroup` is fundamentally racy: 1) ask the OS for a free port by creating a socket bound to port 0, 2) close the socket, 3) attempt to initialize a `TCPStore` server that listens on the previously free port. Of course, the problem is that in between steps 2 and 3, another process on the host may have claimed the port, causing `TCPStore` and overall process group initialization to fail. With this diff, it is now possible for users to completely avoid this race (see unit test for how this can be achieved).
Test Plan:
Added new unit test:
buck2 test caffe2/test/distributed:store
Differential Revision: D46622317
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103478
Approved by: https://github.com/H-Huang
Headers under torch/csrc/distributed may be referened with relative path, e.g., "<c10d/...>". However, relative path cannot be gracefully handled by Meta internal build when the NCCL PG is hipified to support AMD/RCCL because the "hipified" header files are generated in other directories. Moreover, using absolute path for header inclusion is the state-of-the-art in most components in Pytorch. Thus, this patch refactors all header paths in torch/csrc/distributed to be absolute.
See D39835774 for more details about Meta internal complication.
**How to test**: commit 9e5d199 removes -I./torch/csrc/distributed in compile options. Thus use it to verify we don't miss any relative path use of torch/csrc/distributed headers.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/85780
Approved by: https://github.com/kumpera, https://github.com/huydhn
Headers under torch/csrc/distributed may be referened with relative path, e.g., "<c10d/...>". However, relative path cannot be gracefully handled by Meta internal build when the NCCL PG is hipified to support AMD/RCCL because the "hipified" header files are generated in other directories. Moreover, using absolute path for header inclusion is the state-of-the-art in most components in Pytorch. Thus, this patch refactors all header paths in torch/csrc/distributed to be absolute.
See D39835774 for more details about Meta internal complication.
**How to test**: commit 9e5d199 removes -I./torch/csrc/distributed in compile options. Thus use it to verify we don't miss any relative path use of torch/csrc/distributed headers.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/85780
Approved by: https://github.com/kumpera
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68499
TCP store is actually being accessed by multi-threading (NCCL watch dog thread), but no mutex protection while FileStore and HashStore have. As enabling desync root cause analysis makes store calls more often, the race condition of TCP store was always triggered when creating another process group like gloo. Adding mutex to TCP store, to be the same with FileStore and HashStore.
Test Plan:
DDP benchmark with desync debug enabled, no perf regression
https://www.internalfb.com/intern/fblearner/details/309398285?tab=Outputs
W/o this diff
https://www.internalfb.com/intern/fblearner/details/308379789?tab=Outputs
Reviewed By: mingzhe09088
Differential Revision: D32482254
fbshipit-source-id: e8f466e1c6fdcab6cfa170f44b9be70395935fb8
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/60543
Since now c10d is part of libtorch, it would also be nice if the sources lived all in one place.
ghstack-source-id: 132306292
Test Plan: It builds
Reviewed By: cbalioglu
Differential Revision: D29062002
fbshipit-source-id: d9e1301e9d73e1643fa0f0119cd2d618f1ad52e6
