fix: only log config on debug log level, fix trtllm warnings

[Aphoh]

Small fix that

1. only grabs/logs the config on debug or when the --dump-config-to flag is specified
2. properly dumps some trtllm objects that use pydantic, avoiding some warnings

Summary by CodeRabbit

• New Features

  • Optional support for serializing Pydantic models in config dumps.
  • New encoder hook for Config objects to ensure consistent serialization.
  • Config dump now available in DEBUG logs without writing to a file.

• Bug Fixes

  • More robust config dump writing: on file errors, automatically falls back to stdout with clear logging.
  • Ensures config payload is computed before attempting file output to prevent partial or missing dumps.

• Refactor

  • Expanded internal preprocessing to better handle diverse config object types without changing public APIs.

[coderabbitai]

Walkthrough

Refactors config dump logic to compute payload conditionally, add robust file-write fallback to stdout with detailed logging, and emit dumps at DEBUG without files. Extends encoding to optionally process Pydantic models. Adds an encoder registration for TRT-LLM Config objects to serialize via dict. Public APIs remain unchanged.

Changes

Cohort / File(s)	Summary
Config dump flow and encoding enhancements components/src/dynamo/common/config_dump/config_dumper.py	- Compute payload only when needed; handle file write OSError/IOError and general Exception by logging and dumping to stdout. - If logger is DEBUG, compute and log dump without writing to file. - Conditional Pydantic support: attempt BaseModel -> dict via model_dump(); fallback to existing paths when unavailable. - Preprocessor extended to use optional Pydantic handling. - Potential inclusion of additional version info (sglang, trtllm, vllm) preserved. - Public signatures unchanged.
Encoder registration for TRT-LLM Config components/src/dynamo/trtllm/utils/trtllm_utils.py	- Import encoder registration utility and add @register_encoder(Config) with _preprocess_for_encode_config returning obj.__dict__ to participate in encoding pipelines.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  actor U as Caller
  participant CD as ConfigDumper
  participant FS as FileSystem
  participant Log as Logger

  U->>CD: dump_config(dump_config_to=path?)
  alt DEBUG logging with no file
    Note over CD,Log: Logger at DEBUG
    CD->>CD: compute config_dump_payload
    CD->>Log: debug(CONFIG_DUMP payload)
  else File path provided
    CD->>CD: compute config_dump_payload
    CD->>FS: write(payload)
    alt Write fails (OSError/IOError)
      CD->>Log: error("Failed to write... dropping to stdout")
      CD-->>U: stdout(CONFIG_DUMP payload)
    else Other Exception
      CD->>Log: exception("Failed to write... dropping to stdout")
      CD-->>U: stdout(CONFIG_DUMP payload)
    end
  else No path and non-DEBUG
    CD-->>U: return payload
  end

Loading

sequenceDiagram
  autonumber
  participant TR as trtllm_utils
  participant REG as EncoderRegistry
  participant EP as EncodingPipeline
  participant C as Config

  TR->>REG: register_encoder(Config, _preprocess_for_encode_config)
  Note over TR,REG: Registration occurs at import time

  EP->>C: needs encoding
  EP->>REG: lookup encoder for Config
  REG-->>EP: _preprocess_for_encode_config
  EP->>C: call encoder -> __dict__
  EP-->>EP: proceed with downstream encoding

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

I thump my paws—configs leap to light,
From files to stdout in the hush of night.
Pydantic hops in, if carrots align,
TRT-LLM’s Config joins the line.
Logs nibble errors, tidy and neat—
Another merge, another beat. 🥕

Pre-merge checks

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Description Check	⚠️ Warning	The PR description is too brief and does not follow the repository’s required template, which expects "Overview", "Details", "Where should the reviewer start?", and "Related Issues" sections. Without these headings and structured content, important context, change scope, file pointers, and issue references are missing. As a result, the description does not provide sufficient guidance for reviewers or document related issues.	Please update the PR description to adhere to the provided template by adding the "Overview", "Details", "Where should the reviewer start?", and "Related Issues" sections, filling each with the relevant information such as high-level summary, specific code changes, key file locations, and any linked issue numbers.
Docstring Coverage	⚠️ Warning	Docstring coverage is 55.56% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Title Check	✅ Passed	The title clearly highlights the two key changes introduced by this PR: limiting config logging to debug level and resolving warnings in trtllm via improved pydantic handling. It is concise, directly related to the main modifications, and uses specific terminology tied to the code changes. This makes it easy for reviewers and future readers to quickly understand the PR’s intent.

---

📜 Recent review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between d4ff6f0 and d681f80.

📒 Files selected for processing (2)

• components/src/dynamo/common/config_dump/config_dumper.py (3 hunks)
• components/src/dynamo/trtllm/utils/trtllm_utils.py (2 hunks)

🧰 Additional context used

🧬 Code graph analysis (1)

components/src/dynamo/trtllm/utils/trtllm_utils.py (3)

components/src/dynamo/common/config_dump/config_dumper.py (2)

• add_config_dump_args (175-187)
• register_encoder (224-234)

components/src/dynamo/sglang/args.py (2)

• Config (130-146)
• _preprocess_for_encode_config (151-160)

components/src/dynamo/vllm/args.py (2)

• Config (40-68)
• _preprocess_for_encode_config (72-73)

🪛 Ruff (0.14.0)

components/src/dynamo/common/config_dump/config_dumper.py

200-200: Unused function argument: obj

(ARG001)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (6)

• GitHub Check: trtllm (amd64)
• GitHub Check: trtllm (arm64)
• GitHub Check: vllm (amd64)
• GitHub Check: vllm (arm64)
• GitHub Check: sglang
• GitHub Check: Build and Test - dynamo

🔇 Additional comments (7)

components/src/dynamo/trtllm/utils/trtllm_utils.py (2)

11-11: LGTM! Import updated to support encoder registration.

The addition of register_encoder to the imports is necessary for the new Config encoder function below.

---

101-105: LGTM! Config encoder follows established pattern.

The encoder registration for Config is consistent with similar implementations in sglang and vllm modules (as seen in the relevant code snippets). Returning __dict__ is appropriate for serializing this configuration class and should help avoid pydantic-related warnings when the config is dumped.

components/src/dynamo/common/config_dump/config_dumper.py (5)

103-103: LGTM! Conditional payload computation reduces overhead.

Moving the payload computation inside the conditional block ensures it's only generated when actually needed (when dumping to file or at DEBUG level). This optimization aligns with the PR objective to reduce unnecessary config processing.

---

111-117: LGTM! Improved error handling with better context.

The updated error messages provide clearer information about what went wrong and where the config dump is being redirected. The fallback to stdout ensures the config is not lost even when file writing fails.

---

118-121: LGTM! DEBUG-level conditional aligns with PR objectives.

The config is now only computed and logged when the logger is at DEBUG level or lower. This prevents unnecessary overhead in production environments where debug logging is typically disabled, which directly addresses the PR objective to reduce config logging.

---

190-201: LGTM! Optional pydantic support avoids hard dependency.

The try/except ImportError pattern correctly handles pydantic as an optional dependency. When available, it uses model_dump() to serialize pydantic models, which should resolve the trtllm warnings mentioned in the PR objectives.

Note: The static analysis warning about the unused obj parameter in the fallback function (line 200) is a false positive—the signature must match the version with pydantic available.

---

215-216: LGTM! Pydantic integration with appropriate fallback.

The walrus operator efficiently checks if pydantic processing succeeded and returns the result if available, otherwise falling through to the existing __dict__ or dataclass handling. This maintains backward compatibility while adding pydantic support.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}