chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[model-gateway] Add Redis support as a history backend (#16300)

Browse Source
This commit is contained in:
Wenyi Xu
2026-01-11 17:03:00 +08:00
committed by GitHub
parent 7b089ae4e0
commit 3c16c58619
12 changed files with 1215 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
sgl-model-gateway/bindings/python/src/lib.rs
View File

@@ -31,6 +31,7 @@ pub enum HistoryBackendType {
None,
Oracle,
Postgres,
Redis,
}
#[pyclass(eq)]
@@ -270,6 +271,40 @@ impl PyOracleConfig {
}
}
#[pyclass]
#[derive(Debug, Clone, PartialEq)]
pub struct PyRedisConfig {
#[pyo3(get, set)]
pub url: String,
#[pyo3(get, set)]
pub pool_max: usize,
#[pyo3(get, set)]
pub retention_days: Option<u64>,
}
#[pymethods]
impl PyRedisConfig {
#[new]
#[pyo3(signature = (url, pool_max = 16, retention_days = Some(30)))]
fn new(url: String, pool_max: usize, retention_days: Option<u64>) -> PyResult<Self> {
Ok(PyRedisConfig {
url,
pool_max,
retention_days,
})
}
}
impl PyRedisConfig {
pub fn to_config_redis(&self) -> config::RedisConfig {
config::RedisConfig {
url: self.url.clone(),
pool_max: self.pool_max,
retention_days: self.retention_days,
}
}
}
#[pyclass]
#[derive(Debug, Clone, PartialEq)]
pub struct PyPostgresConfig {
@@ -374,6 +409,7 @@ struct Router {
history_backend: HistoryBackendType,
oracle_config: Option<PyOracleConfig>,
postgres_config: Option<PyPostgresConfig>,
redis_config: Option<PyRedisConfig>,
client_cert_path: Option<String>,
client_key_path: Option<String>,
ca_cert_paths: Vec<String>,
@@ -486,6 +522,7 @@ impl Router {
HistoryBackendType::None => config::HistoryBackend::None,
HistoryBackendType::Oracle => config::HistoryBackend::Oracle,
HistoryBackendType::Postgres => config::HistoryBackend::Postgres,
HistoryBackendType::Redis => config::HistoryBackend::Redis,
};
let oracle = if matches!(self.history_backend, HistoryBackendType::Oracle) {
@@ -504,6 +541,14 @@ impl Router {
None
};
let redis_config = if matches!(self.history_backend, HistoryBackendType::Redis) {
self.redis_config
.as_ref()
.map(|cfg| cfg.to_config_redis())
} else {
None
};
config::RouterConfig::builder()
.mode(mode)
.policy(policy)
@@ -558,6 +603,7 @@ impl Router {
.maybe_chat_template(self.chat_template.as_ref())
.maybe_oracle(oracle)
.maybe_postgres(postgres_config)
.maybe_redis(redis_config)
.maybe_reasoning_parser(self.reasoning_parser.as_ref())
.maybe_tool_call_parser(self.tool_call_parser.as_ref())
.maybe_mcp_config_path(self.mcp_config_path.as_ref())
@@ -654,6 +700,7 @@ impl Router {
history_backend = HistoryBackendType::Memory,
oracle_config = None,
postgres_config = None,
redis_config = None,
client_cert_path = None,
client_key_path = None,
ca_cert_paths = vec![],
@@ -737,6 +784,7 @@ impl Router {
history_backend: HistoryBackendType,
oracle_config: Option<PyOracleConfig>,
postgres_config: Option<PyPostgresConfig>,
redis_config: Option<PyRedisConfig>,
client_cert_path: Option<String>,
client_key_path: Option<String>,
ca_cert_paths: Vec<String>,
@@ -834,6 +882,7 @@ impl Router {
history_backend,
oracle_config,
postgres_config,
redis_config,
client_cert_path,
client_key_path,
ca_cert_paths,
@@ -946,6 +995,7 @@ fn sglang_router_rs(m: &Bound<'_, PyModule>) -> PyResult<()> {
m.add_class::<PyControlPlaneAuthConfig>()?;
m.add_class::<PyOracleConfig>()?;
m.add_class::<PyPostgresConfig>()?;
m.add_class::<PyRedisConfig>()?;
m.add_class::<Router>()?;
m.add_function(wrap_pyfunction!(get_version_string, m)?)?;
m.add_function(wrap_pyfunction!(get_verbose_version_string, m)?)?;

20
sgl-model-gateway/bindings/python/src/sglang_router/router.py
View File

@@ -10,6 +10,7 @@ from sglang_router.sglang_router_rs import (
PyJwtConfig,
PyOracleConfig,
PyPostgresConfig,
PyRedisConfig,
PyRole,
)
from sglang_router.sglang_router_rs import Router as _Router
@@ -63,6 +64,8 @@ def history_backend_from_str(backend_str: Optional[str]) -> HistoryBackendType:
return HistoryBackendType.Oracle
elif backend_lower == "postgres":
return HistoryBackendType.Postgres
elif backend_lower == "redis":
return HistoryBackendType.Redis
else:
raise ValueError(f"Unknown history backend: {backend_str}")
@@ -262,6 +265,20 @@ class Router:
)
args_dict["postgres_config"] = postgres_config
# Convert Redis config if needed
redis_config = None
if history_backend == HistoryBackendType.Redis:
retention_days = args_dict.get("redis_retention_days", 30)
# If retention_days is negative, it means persistent storage (None in Rust)
retention_arg = None if retention_days < 0 else retention_days
redis_config = PyRedisConfig(
url=args_dict.get("redis_url"),
pool_max=args_dict.get("redis_pool_max", 16),
retention_days=retention_arg,
)
args_dict["redis_config"] = redis_config
# Build control plane auth config
args_dict["control_plane_auth"] = build_control_plane_auth_config(args_dict)
@@ -278,6 +295,9 @@ class Router:
"oracle_pool_timeout_secs",
"postgres_db_url",
"postgres_pool_max",
"redis_url",
"redis_pool_max",
"redis_retention_days",
# Control plane auth fields (converted to control_plane_auth)
"control_plane_api_keys",
"control_plane_audit_enabled",

30
sgl-model-gateway/bindings/python/src/sglang_router/router_args.py
View File

@@ -117,6 +117,9 @@ class RouterArgs:
oracle_pool_timeout_secs: int = 30
postgres_db_url: Optional[str] = None
postgres_pool_max: int = 16
redis_url: Optional[str] = None
redis_pool_max: int = 16
redis_retention_days: int = 30
# mTLS configuration for worker communication
client_cert_path: Optional[str] = None
client_key_path: Optional[str] = None
@@ -200,6 +203,9 @@ class RouterArgs:
postgres_group = parser.add_argument_group(
"PostgreSQL Database", "PostgreSQL database backend configuration"
)
redis_group = parser.add_argument_group(
"Redis Database", "Redis database backend configuration"
)
tls_group = parser.add_argument_group(
"TLS/mTLS Security", "TLS certificates for server and worker communication"
)
@@ -663,7 +669,7 @@ class RouterArgs:
f"--{prefix}history-backend",
type=str,
default=RouterArgs.history_backend,
choices=["memory", "none", "oracle", "postgres"],
choices=["memory", "none", "oracle", "postgres", "redis"],
help="History storage backend for conversations and responses (default: memory)",
)
@@ -733,6 +739,28 @@ class RouterArgs:
help="Maximum PostgreSQL connection pool size (default: 16, env: POSTGRES_POOL_MAX)",
)
# Redis configuration
redis_group.add_argument(
f"--{prefix}redis-url",
type=str,
default=os.getenv("REDIS_URL"),
help="Redis connection URL (env: REDIS_URL)",
)
redis_group.add_argument(
f"--{prefix}redis-pool-max",
type=int,
default=int(os.getenv("REDIS_POOL_MAX", RouterArgs.redis_pool_max)),
help="Maximum Redis connection pool size (default: 16, env: REDIS_POOL_MAX)",
)
redis_group.add_argument(
f"--{prefix}redis-retention-days",
type=int,
default=int(
os.getenv("REDIS_RETENTION_DAYS", RouterArgs.redis_retention_days)
),
help="Redis data retention in days (-1 for persistent, default: 30, env: REDIS_RETENTION_DAYS)",
)
# TLS/mTLS configuration
tls_group.add_argument(
f"--{prefix}client-cert-path",