sgl-project/sglang · warning

{name} is deprecated; use {replacement} instead.

Error message

{name} is deprecated; use {replacement} instead.

What it means

DeprecationWarning from sglang.srt.layers.dcp.comm: the DCP accessor helpers (dcp_enabled, get_attention_dcp_world_size, get_attention_dcp_rank) are deprecated in favor of newer accessors (the runtime_context get_parallel() state). Each call to these helpers warns once via _warn_deprecated_dcp_accessor.

Source

Thrown at python/sglang/srt/layers/dcp/comm.py:43

import torch

from sglang.kernels.ops.attention.dcp_kernels import (
    CPTritonContext,
    _lse_pack_dim,
    correct_attn_out,
    dcp_lse_combine_triton,
    dcp_pack_a2a_send,
)
from sglang.srt.distributed.device_communicators.pynccl_allocator import (
    use_symmetric_memory,
)
from sglang.srt.distributed.parallel_state import GroupCoordinator
from sglang.srt.runtime_context import get_parallel


def _warn_deprecated_dcp_accessor(name: str, replacement: str) -> None:
    warnings.warn(
        f"{name} is deprecated; use {replacement} instead.",
        DeprecationWarning,
        stacklevel=2,
    )


def dcp_enabled() -> bool:
    """Deprecated: use ``get_parallel().dcp_enabled``."""
    _warn_deprecated_dcp_accessor("dcp_enabled()", "get_parallel().dcp_enabled")
    return get_parallel().dcp_enabled


def get_attention_dcp_world_size() -> int:
    """Deprecated: use ``get_parallel().attn_dcp_size``."""
    _warn_deprecated_dcp_accessor(
        "get_attention_dcp_world_size()", "get_parallel().attn_dcp_size"
    )
    return get_parallel().attn_dcp_size

View on GitHub (pinned to 0132848349)

Solutions

1. Replace calls with the recommended replacements named in the warning (runtime_context get_parallel() DCP fields)
2. Inspect the docstring/source of _warn_deprecated_dcp_accessor callers for the exact replacement names
3. Wrap remaining calls behind your own helper during migration

Example fix

# before
from sglang.srt.layers.dcp.comm import get_attention_dcp_rank
rank = get_attention_dcp_rank()
# after
from sglang.srt.runtime_context import get_parallel
rank = get_parallel().dcp_rank  # per replacement noted in warning

When it happens

Trigger: Calling dcp_enabled(), get_attention_dcp_world_size(), or get_attention_dcp_rank() from model or parallel-state code.

Common situations: DP-attention/DCP model code written before the accessor refactor; checking DCP rank/world size inside custom models after upgrading SGLang.

Related errors

• --dcp-comm-backend fi_a2a delegates the exchange to FlashInf
• --dcp-replicate-q-proj requires --dcp-size > 1.
• --dcp-replicate-q-proj only applies to the a2a/fi_a2a DCP co
• Kimi-K3 DCP with decode_attention_backend='cutedsl_mla' requ
• Kimi-K3 DCP with decode_attention_backend='cutedsl_mla' requ

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/425668473bd6a0f9. Report an issue: GitHub.