This PR enables all PIE rules on ruff, there are already some enabled rules from this family, the new added rules are
```
PIE796 Enum contains duplicate value: {value}
PIE808 Unnecessary start argument in range
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165814
Approved by: https://github.com/ezyang
This PR enables all PIE rules on ruff, there are already some enabled rules from this family, the new added rules are
```
PIE796 Enum contains duplicate value: {value}
PIE808 Unnecessary start argument in range
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165814
Approved by: https://github.com/ezyang
Fixes#154849
This change addresses the request to add support for SIGUSR1 and SIGUSR2 signals in torchrun for SLURM environments. Changes supports these signals through the configurable `TORCHELASTIC_SIGNALS_TO_HANDLE` environment variable and signals_to_handle parameter from laucher api
Tests:
For validations purpose:
test_signal_handling.py,
simple_test_api_signal_handling.py,
Unit Tests:
for launcher changes:launcher/test_api.py
for api changes: multiprocessing/test_api.py
E2E: test_run.py
Pull Request resolved: https://github.com/pytorch/pytorch/pull/160690
Approved by: https://github.com/fduwjj
# Context
This is an extension of #149334.
# This PR
Add support for NUMA bindings with Callable entrypoints, such as `do_train` instead of `/usr/local/bin/python`.
Most notably, we utilize a hack in order to force `Process.start()` to use custom NUMA bindings for each subprocess. Please search for `HACK:` in the code to see a description of the implementation we chose, and #160006 for discussion of alternatives and why this is necessary.
Other changes:
* Remove unnecessary `--preferred` option from all binding strategies. By default, Linux already allocates memory to the NUMA node local to the CPU which triggered the allocation. (See [MPOL_LOCAL](https://man7.org/linux/man-pages/man2/set_mempolicy.2.html).)
* Refactor so that the main API is `maybe_wrap_command_with_numa_bindings`, which computes bindings for a single rank at a time, rather than `maybe_wrap_with_numa_bindings` which computed bindings for all ranks at once. This allowed for more code sharing between `Callable` and `str` entrypoints.
# Test Plan
## Automated
`$ pytest test/test_numa_binding.py`
## Manual
Using [this benchmark,](https://gist.github.com/pdesupinski/bbe01ade455d86e989794f2c612e2d91), ran
```
$ PYTHONUNBUFFERED=1 LOGLEVEL=INFO perf stat -e ls_dmnd_fills_from_sys.dram_io_far,ls_dmnd_fills_from_sys.dram_io_near -- python -m torch.distributed.run --standalone --nproc-per-node=8 --numa-binding=node --run-path mlp_train.py 2>&1 | tee node_callable.txt && PYTHONUNBUFFERED=1 LOGLEVEL=INFO perf stat -e ls_dmnd_fills_from_sys.dram_io_far,ls_dmnd_fills_from_sys.dram_io_near -- python -u -m torch.distributed.run --standalone --nproc-per-node=8 --run-path mlp_train.py 2>&1 | tee none_callable.txt
```
and observed
* 6.6% remote memory accesses with 'node' bindings
* 11.6% remote without bindings
I also ran similar with `str` entrypoints as before just to be sure it's still working.
NOTE: [--run-path triggers the code to be run inside a `Callable`.](017259f9c6/torch/distributed/run.py (L870))
Pull Request resolved: https://github.com/pytorch/pytorch/pull/160163
Approved by: https://github.com/d4l3k
Summary:
In multiprocessing, signal handling is not possible if the thread is not the main thread. This resulted in the following error:
> "ValueError('signal only works in main thread of the main interpreter')"
To address this issue, the diff checks whether the thread is the main thread and, if not, skips signal handling.
Test Plan:
Before this change, MAST job failed:
https://fburl.com/mlhub/iq2m10v8
With this change, MAST job succeeded:
https://fburl.com/mlhub/q6kb8343
Differential Revision: D62166943
Pull Request resolved: https://github.com/pytorch/pytorch/pull/135088
Approved by: https://github.com/d4l3k
The current call passes in `['/actual/path']` to os.walk which is a string pointing to no path and thus silently leads to and empty traversal.
There is an unused function just above that handles that, so I guess this is what was supposed to be called.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/126103
Approved by: https://github.com/suo
Observed Problem
---------------------
When `torchrun` has finished running the main trainer function (aka entrypoint/user function) successfully, I noticed that sometimes it SIGTERMS the child processes. Then `torchrun` exits successfully.
This results in misleading warning log messages towards the end of the job like the one below:
```
W0510 14:52:48.185934 672413 api.py:513] Closing process 675171 via signal SIGTERM
W0510 14:52:48.185984 672413 api.py:513] Closing process 675172 via signal SIGTERM
W0510 14:52:48.186013 672413 api.py:513] Closing process 675174 via signal SIGTERM
# <---- ^^^ ??? everything runs successfully but child still SIGTERM'ed? ^^^ --->
I0510 14:52:48.229119 672413 api.py:877] [main] worker group successfully finished. Waiting 300 seconds for other agents to finish.
I0510 14:52:48.229161 672413 api.py:922] Local worker group finished (WorkerState.SUCCEEDED). Waiting 300 seconds for other agents to finish
I0510 14:52:48.229395 672413 api.py:936] Done waiting for other agents. Elapsed: 0.0001709461212158203 seconds
I0510 14:52:48.257544 672413 dynamic_rendezvous.py:1131] The node 'localhost_672413_0' has closed the rendezvous 'torchrun_qpfd'.
I0510 14:52:48.568198 672413 distributed.py:200] Deleting temp log directory: /tmp/torchrun_udgp8zoq
I0510 14:52:48.568989 672413 distributed.py:202] Finished running `main`
```
Root Cause
------------------
I noticed that this was due to the incorrect usage of `torch.multiprocessing.ProcessContext.join()` in `torch.distributed.elastic.multiprocessing.api.MultiprocessingContext`.
`torch.multiprocessing.ProcessContext.join()` does not actually wait for ALL child procs to exit, but rather waits for **at-least-one** child proc to exit. If only a subset of the child procs have exited, it returns `False` and if all child procs have exited it returns `True`.
`torch.distributed.elastic.multiprocessing.api.MultiprocessingContext` was assuming that `torch.multiprocessing.ProcessContext.join()` blocks indefinitely until all child procs have exited.
Fix
---------
The fix is simple, just loop, while continuing to call `pc.join()` until it returns `True`
> **NOTE**: that the indefinite blocking is NOT an issue since by the time `torch.distributed.elastic.multiprocessing.api.MultiprocessingContext` calls `pc.join()` it already did all the checking to validate that the entrypoint functions either return successfully or that one of them has failed. So we are really just waiting for the unix process to exit after running the entrypoint function.
> **NOTE**: since `pc.join()` already blocks until at-least-one child proc exits, there is no need to add a polling interval in the body of the loop and the debug logging will show at most `nproc_per_node` times so no log spamming is observed.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/125969
Approved by: https://github.com/d4l3k
Summary:
Minor logging cleanup in distributed library
1. Don't use "f" formatted strings - address linter issues.
2. Nits: Make use of unused `e` (error) in a few logs.
3. Change info->debug as asked in issue #113545
4. Nit: rename log -> logger in a few files for consistency
5. Fix a linter error.
Test Plan:
1. Local build passes.
2. Linter is happy.
Reviewers: wanchaol
Pull Request resolved: https://github.com/pytorch/pytorch/pull/122921
Approved by: https://github.com/wanchaol
Summary:
## No Functional Change
- Refactor Subprocess Handler into a separate folder for easier subclassing
- SubprocessHandler
- added `local_rank_id` in `SubprocessHandler` to make it available as a field in the class
- pass in `local_rank_id` from subprocess start
Test Plan: No functional changes.
Differential Revision: D54038627
#suppress-api-compatibility-check
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120373
Approved by: https://github.com/kurman
Summary:
Expose an option to users to specify name of the LogsSpec implementation to use.
- Has to be defined in entrypoints under `torchrun.logs_specs` group.
- Must implement LogsSpec defined in prior PR/diff.
Test Plan: unit test+local tests
Reviewed By: ezyang
Differential Revision: D54180838
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120942
Approved by: https://github.com/ezyang
Summary:
Pulling out logging parameters into a logging specs that can be overridden (follow-up changes on possible mechanism)
Why?
Right now the logging approach is quite rigid:
- Requires for log directory to exist and not be empty
- Will create tempdir otherwise,
- Creates subdir for a run
- creates subdir for each attempt
- creates files named as stdout.log, stderr.log, error.json
In some instances some of the users would like to customize the behavior including file names based on context. And we do have right now a mechanism to template multiplexed teed output prefix.
With current changes, users can create custom log spec that can use env variables to change the behavior.
Notes:
Made `LaunchConf.logs_specs` as an optional field that will be bound to `DefaultLogsSpecs` instance. There are large number of clients (code) that use the API directly without using torchrun API. For those cases, we have to explicitly pass LogSpecs implementation if we would like to override the implementation. For the regular torchrun users, we can use pluggable approach proposed in the follow up change.
Test Plan: CI + unit tests
Differential Revision: D54176265
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120691
Approved by: https://github.com/ezyang
Simplifies and optimizes dict construction using the `fromkeys` classmethod ctor. This also makes it really obvious when all the keys will have the same static value, which could be a bug if unintentional. It is also significantly faster than using a dict comprehension. The rule is in preview, but I am adding a forward fix for when it becomes stable.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/118637
Approved by: https://github.com/albanD
When torchelastic notices that one rank has failed, it will sent a SIGTERM
signal to other trainer ranks to tear them down before restarting. However,
if the trainer itself launches subprocesses, or is launched by a non-python
wrapper script, then the SIGTERM will be delivered only to the direct child of
torch eleastic and not all descendants. This opens subprocesses in a new
linux 'session' which starts a new process group with the pgid the same
as the trainers pid. Then when we send signals, we deliver them to the
process group rather than just the direct child.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/113231
Approved by: https://github.com/H-Huang
Summary:
Add an ability to customize log lines and addtional template like behavior to enrich log information.
Motivation:
a) Log stream processing/aggregation gains additional value when it includes information about the global rank. Extension to that is that it will be easier to map ranks to hosts from log stream information (less relevant at the moment)
b) Users can easily map the failure to the right rank without matching node rank offset+local rank.
Implementation
- BC change - keeps the logs line prefix as `[<role name><local rank>]:`
- Optional env variable TORCHELASTIC_LOG_LINE_HEADER that will be used as a prefix when specified and currently exposes `role_name`, `rank` and `local_rank` variables that will be bound when agent assigns the ranks.
Test Plan:
CI
https://fburl.com/mlhub/mzx5xspv
Differential Revision: D50584590
Pull Request resolved: https://github.com/pytorch/pytorch/pull/112357
Approved by: https://github.com/kiukchung
Summary:
Fixes https://github.com/pytorch/pytorch/issues/71946
This commit removes some bits of code that were hard coded for Python 3.6 support from the `.circleci` and `torch` folders. It should only be merged if https://github.com/pytorch/pytorch/issues/66462 is complete.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/72211
Reviewed By: dagitses, seemethere
Differential Revision: D33982604
Pulled By: musebc
fbshipit-source-id: 8f453bf9909df615addd59538adb369c65484044
(cherry picked from commit 944a9970fe68a40999b5c8af731e632c28fd15c5)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62823
The diff makes sure that the warning message is printed only when the child processes are stuck after sending termination code.
Test Plan:
sandcastle
buck build mode/dev-nosan //caffe2:run
buck-out/gen/caffe2/run.par --nnodes 1 --nproc_per_node 1 main.py
P435691445
Differential Revision: D30046695
fbshipit-source-id: c59170b297f4a0e530906fa5069234303deee938
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61602
The diff introduces signal handlers and SignalException that is raised when the agent process receives SIGTERM or SIGINT.
When any of these signals received, the termination handler will raise the `SignalException`. The exception will then be processed by the main agent loop. The `shutdown(signum)` will be invoked, that would propagate the received signal to the child processes. The default 30 seconds timeout introduced: if child processes will not be able gracefully terminate during this timeout, the agent process would kill the processes via SIGKILL.
Test Plan: unittests, sandcastle
Reviewed By: cbalioglu
Differential Revision: D29671783
fbshipit-source-id: 3dbca2125676dc18d417cc3e3bb0301fdd42737a
