feat(config): add --config and --dump-config for file-based configuration#759
feat(config): add --config and --dump-config for file-based configuration#759
Conversation
… configuration Add support for loading gateway configuration from YAML, JSON, or .env files via a new --config flag, reducing the need to pass 130+ CLI args. What changed: - model_gateway/Cargo.toml: add dotenvy dependency for .env file loading - model_gateway/src/config/loader.rs: new module with load_config_file(), load_env_file(), is_env_file(), dump_config_yaml() and format detection from file extension (.yaml/.yml/.json/.env) - model_gateway/src/config/mod.rs: export loader module and public functions - model_gateway/src/config/types.rs: add #[serde(default)] at struct level on RouterConfig so partial YAML/JSON files work (missing fields use defaults) - model_gateway/src/main.rs: add --config and --dump-config to CliArgs, wire .env loading before clap parsing via pre_extract_config_path(), add build_router_config_from_file() that delegates to RouterConfigBuilder, auto-load .env from CWD via dotenvy::dotenv() - bindings/python/src/smg/router_args.py: add config field and --config CLI arg - bindings/python/src/smg/router.py: pass config_path through to Rust Router - bindings/python/src/lib.rs: add config_path field, load config file or .env in start() before building RouterConfig How it works: - YAML/JSON files deserialize directly into RouterConfig via serde - .env files are loaded into process env before clap parsing so existing env = "..." attributes pick up the values automatically - --dump-config serializes effective RouterConfig to YAML for template generation - Config precedence: Defaults -> Config file -> Env vars -> CLI args Refs: config-file-support Signed-off-by: Simo Lin <linsimo.mark@gmail.com>
slin1237
requested review from
CatherineSue,
gongwei-130 and
key4ng
as code owners
📝 WalkthroughWalkthroughAdds optional config file loading (YAML/JSON/.env) across Rust and Python bindings, a config loader module, CLI flags (--config, --dump-config), pre-arg environment loading, and wiring of an optional Changes
Sequence DiagramsequenceDiagram
participant User as CLI/User
participant Pre as PreExtractor
participant Env as EnvLoader
participant Clap as Clap/ArgParser
participant Loader as ConfigLoader
participant CLIBuilder as CLI->RouterBuilder
participant Router as Router/Server
User->>Pre: raw args
Pre->>Env: if path is .env -> load vars
Pre->>Clap: filtered args
Clap->>Loader: if --config provided -> load file
Loader->>Loader: detect (yaml/json/.env) & parse
Clap->>CLIBuilder: explicit CLI values
Loader->>CLIBuilder: file config
CLIBuilder->>CLIBuilder: merge (CLI overrides explicit)
CLIBuilder->>Router: build RouterConfig
Router->>Router: start with final RouterConfig
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Suggested labels
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches
🧪 Generate unit tests (beta)
📝 Coding Plan
Comment |
Summary of ChangesHello, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request enhances the router's configuration capabilities by introducing support for file-based configuration using YAML, JSON, and Highlights
Changelog
Activity
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
github-actions
bot
added
python-bindings
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a valuable feature for file-based configuration using --config and a --dump-config utility, supporting YAML, JSON, and .env formats. The implementation of the file loading and parsing is well-done. However, there are critical flaws in how the loaded configuration is merged with command-line or Python arguments. In both the main binary and the Python bindings, loading a YAML or JSON configuration file causes most other arguments to be ignored, violating the documented precedence where CLI/Python arguments should override file settings. Furthermore, the Python bindings incorrectly handle .env files, rendering them ineffective. The Python bindings also contain duplicated logic that should be refactored into a shared helper function. These issues will lead to unexpected and confusing behavior for users and need to be addressed.
model_gateway/src/main.rs
Outdated
| fn build_router_config_from_file(&self, config_path: &str) -> ConfigResult<RouterConfig> { | ||
| let config = smg::config::load_config_file(config_path)?; | ||
|
|
||
| RouterConfigBuilder::from_config(config) | ||
| .maybe_server_cert_and_key(self.tls_cert_path.as_ref(), self.tls_key_path.as_ref()) | ||
| .maybe_mcp_config_path(self.mcp_config_path.as_ref()) | ||
| .build() | ||
| } |
There was a problem hiding this comment.
This function incorrectly handles configuration precedence. It loads a configuration from a YAML/JSON file but then only applies a few specific overrides (tls_cert_path, mcp_config_path) from the command-line arguments. Most other CLI arguments are ignored, which violates the documented precedence order: Defaults → Config file → Env vars → CLI args. This means users cannot override most settings from the config file via the CLI, which can lead to unexpected behavior and confusion. The --dump-config feature is also affected by this, as it won't dump the true effective configuration.
| let router_config = if let Some(ref path) = self.config_path { | ||
| if config::is_env_file(path) { | ||
| config::load_env_file(path).map_err(|e| { | ||
| pyo3::exceptions::PyValueError::new_err(format!( | ||
| "Failed to load .env file: {e}" | ||
| )) | ||
| })?; | ||
| self.to_router_config().map_err(|e| { | ||
| pyo3::exceptions::PyValueError::new_err(format!("Configuration error: {e}")) | ||
| })? | ||
| } else { | ||
| config::load_config_file(path).map_err(|e| { | ||
| pyo3::exceptions::PyValueError::new_err(format!( | ||
| "Failed to load config file: {e}" | ||
| )) | ||
| })? | ||
| } | ||
| } else { | ||
| self.to_router_config().map_err(|e| { | ||
| pyo3::exceptions::PyValueError::new_err(format!("Configuration error: {e}")) | ||
| })? | ||
| }; |
There was a problem hiding this comment.
The configuration loading logic in the start method is flawed and does not respect the documented precedence, leading to incorrect behavior for Python users.
.envfile handling: The.envfile is loaded into the process environment, but the configuration is then built from the Python arguments passed duringRouterinitialization. This completely ignores the environment variables from the file..envfiles should be loaded in the Python layer before theRouterobject is created to be effective.- YAML/JSON file handling: The configuration is loaded from the file, but this discards all arguments passed from Python. Python arguments should override values from the configuration file.
- Code Duplication: The call to
self.to_router_config()and its error mapping is duplicated. This duplicated logic should be extracted into a shared helper function to improve maintainability and reduce redundancy.
These issues will cause significant confusion for users of the Python bindings.
References
- Extract duplicated logic into a shared helper function to improve maintainability and reduce redundancy.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c4371f32e9
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| RouterConfigBuilder::from_config(config) | ||
| .maybe_server_cert_and_key(self.tls_cert_path.as_ref(), self.tls_key_path.as_ref()) | ||
| .maybe_mcp_config_path(self.mcp_config_path.as_ref()) | ||
| .build() |
There was a problem hiding this comment.
When --config points to YAML/JSON, this code returns the deserialized file config and only reapplies TLS/MCP path handling, so explicit CLI/env values (for example --port, --policy, or other clap/env-backed fields) never override file values. Both startup and --dump-config route through this helper in main.rs, so the advertised precedence of config file < env < CLI is not actually honored.
Useful? React with 👍 / 👎.
bindings/python/src/lib.rs
Outdated
| config::load_config_file(path).map_err(|e| { | ||
| pyo3::exceptions::PyValueError::new_err(format!( | ||
| "Failed to load config file: {e}" | ||
| )) | ||
| })? |
There was a problem hiding this comment.
In the non-.env branch, start() uses load_config_file(path) directly and skips self.to_router_config(), so constructor/CLI arguments are not merged into the loaded config; this breaks per-run overrides promised by the Python CLI help. It also creates inconsistent behavior because ServerConfig is still built later from self fields (e.g., host/port), while router-level settings come from the file.
Useful? React with 👍 / 👎.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@bindings/python/src/lib.rs`:
- Around line 1025-1046: The code currently constructs router_config differently
depending on config_path, causing .env loads to happen after Router construction
and config-file loads to skip builder-populated assets; fix by ensuring the
config file (either via config::load_env_file or config::load_config_file) is
applied to self before calling self.to_router_config() so that builder assets
(TLS/CA/MCP) and .env values are included; specifically, in the branch where
config::is_env_file(path) call config::load_env_file(path) and then always call
self.to_router_config(), and in the non-env branch load the config file into
self (or merge the parsed config into self) and then call
self.to_router_config() as well, and make server::startup(...) use the resulting
router_config rather than reading raw fields like self.host/self.port.
In `@bindings/python/src/smg/router_args.py`:
- Around line 232-245: The new CLI flag is always registered as "--config" which
ignores the add_cli_args(use_router_prefix=True) behavior and can collide with
parent parsers; change the registration in router_args.py so
config_group.add_argument uses the same prefixing logic as the rest of
add_cli_args (i.e., compute the flag name with the router prefix when
use_router_prefix is True instead of hardcoding "--config"), reusing the
existing prefix helper used elsewhere in add_cli_args and ensuring
from_cli_args() will still read the prefixed key.
In `@model_gateway/src/main.rs`:
- Around line 993-1003: build_router_config_from_file currently loads the config
file into RouterConfig but only applies TLS and MCP overrides, causing other
CLI/env values to neither be read from the file nor override it; update
build_router_config_from_file (and/or RouterConfigBuilder::from_config) so the
builder fully merges file values with CLI/env overrides: load the file into
RouterConfig, pass that into RouterConfigBuilder::from_config, then apply every
relevant CLI override (host/port/discovery/metrics/logging flags present on
CliArgs) via the builder’s methods before calling build(), and ensure
to_server_config() no longer re-reads values directly from CliArgs but uses the
merged RouterConfig produced by build_router_config_from_file (adjust
to_server_config or CliArgs usage accordingly).
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: 01b633db-e77c-4020-af75-fe417b582451
📒 Files selected for processing (8)
bindings/python/src/lib.rsbindings/python/src/smg/router.pybindings/python/src/smg/router_args.pymodel_gateway/Cargo.tomlmodel_gateway/src/config/loader.rsmodel_gateway/src/config/mod.rsmodel_gateway/src/config/types.rsmodel_gateway/src/main.rs
…dence
Replace field-by-field override approach with generic JSON deep merge
that requires zero changes when adding new flat config fields.
What changed:
- model_gateway/src/main.rs: Add merge_configs(), deep_merge(), and
cli_arg_to_config_path() functions that serialize both the file-loaded
and CLI-built RouterConfig to JSON, then recursively merge based on
which CLI args were explicitly set via command line or env vars.
Enum/structural fields (mode, policy, connection_mode) are handled
separately since they can't be deep-merged. CLI arg naming mismatches
(cb_* vs circuit_breaker_*, health_* vs health_check_*) are resolved
via an alias table.
- model_gateway/src/main.rs: Change Cli::parse_from to
Cli::command().get_matches_from() + from_arg_matches() to retain
ArgMatches for value_source() checks.
- model_gateway/src/main.rs: Add sync_server_fields() to sync merged
config values back to CliArgs for fields that to_server_config()
reads directly.
- bindings/python/src/lib.rs: Fix YAML/JSON config loading to pass
through RouterConfigBuilder for cert/MCP/CA loading instead of
returning raw config. Extract ServerConfig fields from RouterConfig
so config file values are used instead of Python defaults.
- bindings/python/src/smg/router_args.py: Fix --config arg to use
f"--{prefix}config" like all other args.
Why:
PR review feedback identified that the previous field-by-field override
approach was not scalable — adding a new field required updating the
override list. The JSON deep merge uses naming convention (field_name
in config JSON matches field-name as clap arg ID) to automatically
handle any flat field. Only structural mismatches (enums, prefix
aliases) need explicit handling.
Refs: #759
Signed-off-by: Simo Lin <linsimo.mark@gmail.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 92f908336c
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| )* | ||
| }; | ||
| } | ||
| sync!(host, port, max_payload_size, request_timeout_secs); |
There was a problem hiding this comment.
build_router_config_from_file() merges file values into RouterConfig, but sync_server_fields() only copies host/port/log-related fields, so self.service_discovery remains the CLI default (false) when discovery is set only in YAML/JSON. to_server_config() then skips building ServiceDiscoveryConfig because it gates on self.service_discovery, so a config file with discovery enabled is silently ignored unless --service-discovery is also passed.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Actionable comments posted: 2
♻️ Duplicate comments (2)
model_gateway/src/main.rs (1)
1176-1203:⚠️ Potential issue | 🟠 MajorServer-facing fields are still only partially synced from the merged config.
to_server_config()below still builds service discovery, Prometheus, and request-id handling fromself.service_discovery,self.prometheus_*, andself.request_id_headers. Because this sync only backfills host/port/payload/timeout/logging, file values underdiscovery,metrics, andrequest_id_headersremain inert when--configis used.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@model_gateway/src/main.rs` around lines 1176 - 1203, sync_server_fields only backfills host/port/max_payload_size/request_timeout_secs/logging but misses other server-facing fields used by to_server_config such as service_discovery, prometheus_* (e.g., prometheus_enabled, prometheus_listen_addr, prometheus_namespace or whatever prometheus fields exist), and request_id_headers; update sync_server_fields to also check is_explicitly_set for the corresponding arg names (convert underscores to hyphens like the macro does) and copy config values into self (cloning as needed) for service_discovery, all prometheus-related fields, and request_id_headers so that to_server_config reads the merged values instead of only the CLI ones.bindings/python/src/lib.rs (1)
1025-1062:⚠️ Potential issue | 🔴 Critical
config_pathstill doesn’t produce a merged effective config in the Python startup path.The
.envbranch loads afterRouteris constructed, andself.to_router_config()never re-reads env, so the file is a no-op for router settings. The YAML/JSON branch does the opposite and builds straight fromfile_config, which drops explicitRouter(...)overrides and leaves downstream server settings like discovery/metrics tied toselfinstead of the loaded config.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@bindings/python/src/lib.rs` around lines 1025 - 1062, The config_path handling currently treats the .env branch as a no-op for Router settings and the YAML/JSON branch as ignoring explicit Python Router(...) overrides; fix by (1) in the env-file branch call config::load_env_file(path) first and then build the router config from self (i.e. call self.to_router_config()) so to_router_config() sees the newly loaded env, and (2) in the file-config branch after loading file_config via config::load_config_file(path) use config::RouterConfigBuilder::from_config(file_config) but then apply the explicit overrides from self (the same overrides applied via maybe_server_cert_and_key, maybe_client_cert_and_key, add_ca_certificates, maybe_mcp_config_path, etc.) before calling build() so the resulting RouterConfig merges file settings with any Python-specified fields instead of discarding them.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@model_gateway/src/main.rs`:
- Around line 91-127: The code currently replaces the whole "policy" or
"mode"/"connection_mode" object whenever any subfield listed in POLICY_ARGS or
MODE_ARGS is present in explicit; change the conditions so the entire enum is
only swapped when the top-level key itself was explicitly provided. Concretely,
in the block that handles MODE_ARGS, replace the MODE_ARGS.iter().any(...) check
with a direct check like explicit.contains("mode") ||
explicit.contains("connection_mode") before inserting
override_map["mode"/"connection_mode"]; and in the POLICY_ARGS block replace the
POLICY_ARGS.iter().any(...) check with explicit.contains("policy") before
inserting override_map["policy"]. Use the same variables shown (base_map,
override_map, explicit, "policy", "mode", "connection_mode") so only top-level
keys trigger full replacement while subfield flags continue to be handled
without overwriting the entire enum.
---
Duplicate comments:
In `@bindings/python/src/lib.rs`:
- Around line 1025-1062: The config_path handling currently treats the .env
branch as a no-op for Router settings and the YAML/JSON branch as ignoring
explicit Python Router(...) overrides; fix by (1) in the env-file branch call
config::load_env_file(path) first and then build the router config from self
(i.e. call self.to_router_config()) so to_router_config() sees the newly loaded
env, and (2) in the file-config branch after loading file_config via
config::load_config_file(path) use
config::RouterConfigBuilder::from_config(file_config) but then apply the
explicit overrides from self (the same overrides applied via
maybe_server_cert_and_key, maybe_client_cert_and_key, add_ca_certificates,
maybe_mcp_config_path, etc.) before calling build() so the resulting
RouterConfig merges file settings with any Python-specified fields instead of
discarding them.
In `@model_gateway/src/main.rs`:
- Around line 1176-1203: sync_server_fields only backfills
host/port/max_payload_size/request_timeout_secs/logging but misses other
server-facing fields used by to_server_config such as service_discovery,
prometheus_* (e.g., prometheus_enabled, prometheus_listen_addr,
prometheus_namespace or whatever prometheus fields exist), and
request_id_headers; update sync_server_fields to also check is_explicitly_set
for the corresponding arg names (convert underscores to hyphens like the macro
does) and copy config values into self (cloning as needed) for
service_discovery, all prometheus-related fields, and request_id_headers so that
to_server_config reads the merged values instead of only the CLI ones.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: f92b729e-e36f-41ae-9f8a-1a9ab22cae78
📒 Files selected for processing (3)
bindings/python/src/lib.rsbindings/python/src/smg/router_args.pymodel_gateway/src/main.rs
| fn merge_configs( | ||
| file_config: RouterConfig, | ||
| cli_config: RouterConfig, | ||
| matches: &clap::ArgMatches, | ||
| ) -> ConfigResult<RouterConfig> { | ||
| // Collect explicitly-set CLI arg IDs, normalized to underscore format. | ||
| // Also expand aliases for args whose names don't match config field paths. | ||
| let explicit: HashSet<String> = matches | ||
| .ids() | ||
| .filter(|id| is_explicitly_set(matches, id.as_str())) | ||
| .flat_map(|id| { | ||
| let key = id.as_str().replace('-', "_"); | ||
| let mut keys = vec![key.clone()]; | ||
| if let Some(alias) = cli_arg_to_config_path(&key) { | ||
| keys.push(alias); | ||
| } | ||
| keys | ||
| }) | ||
| .collect(); | ||
|
|
||
| let mut base = serde_json::to_value(&file_config).map_err(|e| { | ||
| ConfigError::ValidationFailed { | ||
| reason: format!("Failed to serialize file config: {e}"), | ||
| } | ||
| })?; | ||
| let overrides = serde_json::to_value(&cli_config).map_err(|e| { | ||
| ConfigError::ValidationFailed { | ||
| reason: format!("Failed to serialize CLI config: {e}"), | ||
| } | ||
| })?; | ||
|
|
||
| // Skip enum/structural fields — they need whole-value replacement, not deep merge | ||
| let skip_keys: &[&str] = &["mode", "policy", "connection_mode"]; | ||
| deep_merge(&mut base, &overrides, &explicit, "", skip_keys); |
There was a problem hiding this comment.
Nested sections still miss the advertised file < env < CLI precedence.
explicit contains arg IDs like prometheus_port, service_discovery_port, retry_max_retries, or selector, but deep_merge() checks paths like metrics_port, discovery_port, retry_config_max_retries, and recurses past explicit parent maps. So explicit overrides in metrics, discovery, retry_config, and selector maps still lose to the file config.
Also applies to: 134-158, 166-195
| // Handle enum/structural fields: replace entirely when any constituent CLI arg is set | ||
| if let (Some(base_map), Some(override_map)) = (base.as_object_mut(), overrides.as_object()) { | ||
| const MODE_ARGS: &[&str] = &[ | ||
| "worker_urls", | ||
| "pd_disaggregation", | ||
| "decode", | ||
| "backend", | ||
| "prefill_policy", | ||
| "decode_policy", | ||
| ]; | ||
| if MODE_ARGS.iter().any(|k| explicit.contains(*k)) { | ||
| for key in ["mode", "connection_mode"] { | ||
| if let Some(val) = override_map.get(key) { | ||
| base_map.insert(key.to_string(), val.clone()); | ||
| } | ||
| } | ||
| } | ||
|
|
||
| const POLICY_ARGS: &[&str] = &[ | ||
| "policy", | ||
| "cache_threshold", | ||
| "balance_abs_threshold", | ||
| "balance_rel_threshold", | ||
| "eviction_interval", | ||
| "max_tree_size", | ||
| "block_size", | ||
| "max_idle_secs", | ||
| "assignment_mode", | ||
| "prefix_token_count", | ||
| "prefix_hash_load_factor", | ||
| ]; | ||
| if POLICY_ARGS.iter().any(|k| explicit.contains(*k)) { | ||
| if let Some(val) = override_map.get("policy") { | ||
| base_map.insert("policy".to_string(), val.clone()); | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
Don’t replace policy/mode from CLI defaults when only subfields were explicit.
POLICY_ARGS / MODE_ARGS currently swap the entire enum with override_map["policy"] / ["mode"], but those values are rebuilt from CliArgs defaults. A file policy: manual plus --max-idle-secs 30 flips back to default cache_aware, and a file PD config plus --decode ... without --pd-disaggregation gets replaced by Regular.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@model_gateway/src/main.rs` around lines 91 - 127, The code currently replaces
the whole "policy" or "mode"/"connection_mode" object whenever any subfield
listed in POLICY_ARGS or MODE_ARGS is present in explicit; change the conditions
so the entire enum is only swapped when the top-level key itself was explicitly
provided. Concretely, in the block that handles MODE_ARGS, replace the
MODE_ARGS.iter().any(...) check with a direct check like
explicit.contains("mode") || explicit.contains("connection_mode") before
inserting override_map["mode"/"connection_mode"]; and in the POLICY_ARGS block
replace the POLICY_ARGS.iter().any(...) check with explicit.contains("policy")
before inserting override_map["policy"]. Use the same variables shown (base_map,
override_map, explicit, "policy", "mode", "connection_mode") so only top-level
keys trigger full replacement while subfield flags continue to be handled
without overwriting the entire enum.
|
Hi @slin1237, this PR has merge conflicts that must be resolved before it can be merged. Please rebase your branch: git fetch origin main
git rebase origin/main
# resolve any conflicts, then:
git push --force-with-lease |
mergify
bot
added
the
needs-rebase
1 participant
Summary
--config <path>flag to load configuration from YAML, JSON, or.envfiles instead of passing 130+ CLI arguments--dump-configflag to serialize effective configuration as YAML for template generationWhat changed
model_gateway/Cargo.tomldotenvydependency for.envfile loadingmodel_gateway/src/config/loader.rs.envloading, YAML serialization, and 6 unit testsmodel_gateway/src/config/mod.rsmodel_gateway/src/config/types.rs#[serde(default)]at struct level onRouterConfigso partial configs workmodel_gateway/src/main.rs--config/--dump-configflags,.envpre-loading before clap,build_router_config_from_file()viaRouterConfigBuilderbindings/python/src/smg/router_args.pyconfigfield and--configCLI argumentbindings/python/src/smg/router.pyconfig_paththrough to RustRouterbindings/python/src/lib.rsconfig_pathfield, load config file or.envinstart()How it works
RouterConfigvia serde, then passes throughRouterConfigBuilderfor cert/MCP file loading.env: loaded into process environment before clap parsing so existingenv = "..."attributes pick up values automatically--dump-config: serializes effectiveRouterConfigto YAML and exitsExample usage
Test plan
loader.rs(YAML roundtrip, partial YAML defaults, JSON loading, format detection, missing file errors,.envdetection)cargo test -p smg)cargo clippy --all-targets --all-features -- -D warningscleansmgandsmg-pythoncrates compileSummary by CodeRabbit
New Features
Tests