Resiliency

Stable docs: @docs/training/resiliency.md, @docs/training/checkpointing.md Card: @skills/nemo-mbridge-resiliency/card.yaml

Enablement

Fault tolerance (Slurm only)

Option 1: NeMo Run plugin (recommended)

from megatron.bridge.recipes.run_plugins import FaultTolerancePlugin
import nemo_run as run

task = run.Script(...)
run_plugins = [
    FaultTolerancePlugin(
        enable_ft_package=True,
        calc_ft_timeouts=True,
        num_in_job_restarts=3,
        num_job_retries_on_failure=2,
        initial_rank_heartbeat_timeout=1800,
        rank_heartbeat_timeout=300,
    )
]
run.run(task, plugins=run_plugins, executor=executor)

Option 2: Direct config + ft_launcher

from megatron.bridge.training.config import FaultToleranceConfig

cfg.ft = FaultToleranceConfig(
    enable_ft_package=True,
    calc_ft_timeouts=True,
    simulate_fault=False,
    simulated_fault_type="random",
)

Launch with ft_launcher (not torchrun):

export GROUP_RANK=0  # required for non-Slurm
ft_launcher \
    --rdzv_backend=c10d --rdzv_endpoint=${MASTER_ADDR}:${MASTER_PORT} \
    --nnodes=${NUM_NODES} --nproc-per-node=${NUM_GPUS_PER_NODE} \
    --ft-rank_section_timeouts=setup:600,step:180,checkpointing:420 \
    --ft-rank_out_of_section_timeout=300 \
    your_training_script.py

Section-based timeout monitoring covers setup, training steps, checkpointing, and out-of-section time independently. Timeouts are saved to ftstate.json for subsequent runs when calcft_timeouts=True.

NVRx straggler detection

from megatron.bridge.training.config import NVRxStragglerDetectionConfig

cfg.nvrx_straggler = NVRxStragglerDetectionConfig(
    enabled=True,
    report_time_interval=300.0,
    calc_relative_gpu_perf=True,
    calc_individual_gpu_perf=True,
    num_gpu_perf_scores_to_print=5,
    gpu_relative_perf_threshold=0.7,
    gpu_individual_perf_threshold=0.7,
    stop_if_detected=False,
    enable_logging=True,
)

Preemption

Plugin (Slurm)

from megatron.bridge.recipes.run_plugins import PreemptionPlugin

plugins = [
    PreemptionPlugin(
        preempt_time=60,
        enable_exit_handler=True,
        enable_exit_handler_for_data_loader=False,
    )
]

Direct config

import signal
cfg.train.exit_signal_handler = True
cfg.train.exit_signal = signal.SIGTERM
cfg.train.exit_signal_handler_for_dataloader = False

Re-run state machine (experimental)

from megatron.bridge.training.config import RerunStateMachineConfig

cfg.rerun_state_machine = RerunStateMachineConfig(
    rerun_mode="validate_results",
    check_for_nan_in_loss=True,
    check_for_spiky_loss=False,
    spiky_loss_factor=10.0,
)

Exit codes: 16 = resume to disambiguate, 17 = failed validation.

In-process restart (experimental)

from megatron.bridge.training.config import InProcessRestartConfig

cfg.inprocess_restart = InProcessRestartConfig(
    enabled=True,
    granularity="node",
    soft_timeout=60.0,
    hard_timeout=90.0,
)

Required environment variables:

export TORCH_CPP_LOG_LEVEL=error
export TORCH_NCCL_RETHROW_CUDA_ERRORS=0
export NCCL_NVLS_ENABLE=0

The PyTorch NCCL watchdog timeout must exceed hard_timeout. NeMo-Run's Slurm Executor is not supported; launch directly with srun --kill-on-bad-exit=0.

Async checkpoint save

cfg.checkpoint.async_save = True
cfg.checkpoint.ckpt_format = "torch_dist"

Local checkpointing (NVRx)

cfg.checkpoint.non_persistent_local_ckpt_dir = "/local/scratch/ckpt"
cfg.checkpoint.non_persistent_local_ckpt_algo = "fully_parallel"

Code Anchors

Fault tolerance

• Config: src/megatron/bridge/training/config.py — FaultToleranceConfig
• Runtime: src/megatron/bridge/training/fault_tolerance.py
• Plugin: src/megatron/bridge/recipes/run_plugins.py — FaultTolerancePlugin
• Perf plugin: scripts/performance/nemo-mbridge-resiliency_plugins.py
• Tests: tests/unittests/training/testfault_tolerance.py
• Example: examples/trainingfeatures/nemo-mbridge-resiliency/faulttolerance/

Straggler detection

• Config: src/megatron/bridge/training/config.py — NVRxStragglerDetectionConfig
• Runtime: src/megatron/bridge/training/nvrx_straggler.py
• Train loop: src/megatron/bridge/training/train.py — checknvrxstraggler_detection
• Tests: tests/unittests/training/testnvrxstraggler.py, tests/functionaltests/training/testnvrxstraggler.py
• Example: examples/trainingfeatures/nemo-mbridge-resiliency/stragglerdetection/

In-process restart

• Config: src/megatron/bridge/training/config.py — InProcessRestartConfig
• Runtime: src/megatron/bridge/training/inprocess_restart.py
• Entry point: src/megatron/bridge/training/pretrain.py — maybewrapforinprocessrestart
• Tests: tests/unittests/training/testinprocessrestart.py, tests/functionaltests/training/testinprocessrestart.py

Preemption

• Plugin: src/megatron/bridge/recipes/run_plugins.py — PreemptionPlugin
• Signal handler: src/megatron/bridge/training/utils/sig_utils.py
• Tests: tests/unittests/recipes/testrun_plugins.py

Re-run state machine

• Config: src/megatron/bridge/training/config.py — RerunStateMachineConfig
• Init: src/megatron/bridge/training/initialize.py — initrerunstate

Checkpointing

• Async save: src/megatron/bridge/training/checkpointing.py — scheduleasyncsave
• Local ckpt: src/megatron/bridge/training/checkpointing.py — LocalCheckpointManager
• Tests: tests/functionaltests/training/testlocal_checkpointing.py

Pitfalls

1. ft_launcher, not torchrun: Direct FaultToleranceConfig requires

ftlauncher. Using torchrun silently disables FT. For non-Slurm, set GROUPRANK=0.

1. Async save requires torchdist: asyncsave=True only works with

ckptformat="torchdist". Other formats silently fail or error.

1. IPR + NeMo-Run: In-process restart is not compatible with NeMo-Run

or Slurm preemption plugins. Requires specific PyTorch/NCCL versions and env vars.

1. NVRx vs legacy straggler: Two detectors exist. Use NVRx

(nvrx_straggler); do not enable both.

1. stopifdetected default: NVRx logs but does not stop training by

default. Set stopifdetected=True for automatic termination.

1. NCCL watchdog vs hard_timeout: For IPR, NCCL watchdog timeout must

exceed hard_timeout or PyTorch kills the process before recovery.

1. Rerun state machine is alpha: Use checkfornaninloss=True for

NaN detection, but don't rely on full rerun workflows yet.

Verification

Fault tolerance

./examples/training_features/nemo-mbridge-resiliency/fault_tolerance/run_fault_tolerance.sh
./examples/training_features/nemo-mbridge-resiliency/fault_tolerance/run_fault_tolerance.sh --simulate-fault

Look for [FaultTolerance] / [RankMonitorServer] log lines with section timeouts. Simulated fault should trigger restart from checkpoint.

Straggler detection

uv run python -m torch.distributed.run --nproc_per_node=2 \
    examples/training_features/nemo-mbridge-resiliency/straggler_detection/straggler_detection_example.py

Look for GPU relative performance and GPU individual performance reports with per-rank scores.

Async checkpoint

Look for Scheduling async checkpoint save in logs. Training iterations should continue while checkpoint files are being written.

In-process restart

pytest tests/functional_tests/training/test_inprocess_restart.py -v

Requires compatible PyTorch/NCCL versions.