bsodfather/crank
1
0
Fork 0
Code Issues Pull Requests 3 Actions Packages Projects Releases Wiki Activity

feat: harden community production foundation through story 1.5

Browse Source
This commit is contained in:
bsodfather
2026-08-14 00:21:59 +03:00
parent c30461cc92
commit f6fc2e5c9b
161 changed files with 16758 additions and 2515 deletions
Download Patch File Download Diff File Expand all files Collapse all files
apps/mcp-server/src/main.rs
+280 -71
View File
@@ -1,12 +1,17 @@
use std::{env, net::SocketAddr, time::Duration};
use std::{io, process::ExitCode, time::Duration};
use crank_community_mcp::{
auth::CommunityMachineCredentialVerifier, build_app_with_background_workers_and_limits,
session::PostgresTransportSessionStore,
};
use crank_config::{
CacheBackend as ConfigCacheBackend, ConfigSource, DatabaseSettings, DiagnosticCode,
McpProcessConfig, ObservabilitySettings, ProcessKind, parse_process,
};
use crank_core::CacheBackend;
use crank_observability::{
CriticalErrorCategory, MetricsConfig, ObservabilityConfig, ObservabilityLifecycle,
capture_critical_error,
OtlpTraceConfig, RedactionLimits, SentryConfig, ServiceIdentity, capture_critical_error,
};
use crank_registry::{PostgresPoolConfig, PostgresRegistry};
use crank_runtime::{
@@ -19,14 +24,76 @@ use tokio::net::TcpListener;
use tracing::info;
#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
let observability = crank_observability::init(ObservabilityConfig::from_env(
"mcp-server",
env!("CARGO_PKG_VERSION"),
"mcp_server=info,tower_http=info",
)?)?;
async fn main() -> ExitCode {
match main_result().await {
Ok(()) => ExitCode::SUCCESS,
Err(error) => {
eprintln!("{}", safe_startup_diagnostic(error.as_ref()));
ExitCode::FAILURE
}
}
}
fn safe_startup_diagnostic(error: &(dyn std::error::Error + 'static)) -> String {
let mut current = Some(error);
while let Some(cause) = current {
if let Some(config) = cause.downcast_ref::<crank_config::ConfigError>() {
return config.to_json();
}
if let Some(migration) = cause.downcast_ref::<crank_registry::MigrationError>() {
return serde_json::json!({
"status": "error",
"code": migration.code(),
"stage": migration.stage(),
"version": migration.version(),
"recovery": migration.recovery(),
})
.to_string();
}
if let Some(crank_registry::RegistryError::Migration(migration)) =
cause.downcast_ref::<crank_registry::RegistryError>()
{
return serde_json::json!({
"status": "error",
"code": migration.code(),
"stage": migration.stage(),
"version": migration.version(),
"recovery": migration.recovery(),
})
.to_string();
}
current = cause.source();
}
serde_json::json!({
"status": "error",
"code": "startup_failed",
"stage": "startup",
"version": null,
"recovery": "contact_operator",
})
.to_string()
}
async fn main_result() -> Result<(), Box<dyn std::error::Error>> {
let effective = parse_process(ProcessKind::McpServer, ConfigSource::from_os()?)?;
let config = effective
.mcp()
.cloned()
.ok_or_else(|| io::Error::other("MCP configuration projection is unavailable"))?;
preflight_config(&config)?;
let observability = init_observability(&config.observability)?;
for deprecation in effective.deprecations() {
tracing::warn!(
name: "config.deprecated",
field = deprecation.field,
source_class = deprecation.source_class,
replacement = deprecation.replacement,
removal_window = deprecation.removal_window,
"deprecated configuration accepted"
);
}
let mut startup_completed = false;
let result = run(&observability, &mut startup_completed).await;
let result = run(config, &observability, &mut startup_completed).await;
if result.is_err() {
capture_critical_error(if startup_completed {
CriticalErrorCategory::Internal
@@ -38,49 +105,61 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
}
async fn run(
config: McpProcessConfig,
observability: &ObservabilityLifecycle,
startup_completed: &mut bool,
) -> Result<(), Box<dyn std::error::Error>> {
let metrics_config =
MetricsConfig::from_env("CRANK_MCP_METRICS_BIND", "127.0.0.1:9465".parse()?)?;
let metrics_config = MetricsConfig::new(
config.observability.metrics.enabled,
config.observability.metrics.bind_addr,
config
.observability
.metrics
.bearer_token
.as_ref()
.map(|token| token.expose_secret().to_owned()),
)?;
let metrics_enabled = metrics_config.enabled();
let metrics_server = if metrics_config.enabled() {
let pool_config = postgres_pool_config(&config.database)?;
let runtime_limits = RuntimeLimits::try_new(
config.runtime.max_concurrent_unary,
config.runtime.max_concurrent_sessions,
)?;
let cache_config = runtime_cache_config(&config)?;
let cache_stores = RuntimeCacheStores::from_config(&cache_config).await?;
let api_rate_limit = RequestRateLimitConfig::new(
config.rate_limit.requests_per_second,
config.rate_limit.burst,
)?;
let registry = PostgresRegistry::connect_with_options_and_pool_config(
database_options(&config.database)?,
pool_config,
)
.await?;
let metrics_server = if metrics_enabled {
Some(observability.metrics_surface(metrics_config).bind().await?)
} else {
None
};
let bind_addr = env::var("CRANK_MCP_BIND").unwrap_or_else(|_| "0.0.0.0:3002".into());
let base_url = env::var("CRANK_BASE_URL").ok();
let refresh_interval = env::var("CRANK_MCP_REFRESH_MS")
.ok()
.and_then(|value| value.parse::<u64>().ok())
.map(Duration::from_millis)
.unwrap_or_else(|| Duration::from_secs(5));
let socket_addr: SocketAddr = bind_addr.parse()?;
let pool_config = PostgresPoolConfig::from_env()?;
let runtime_limits = RuntimeLimits::from_env()?;
let cache_config = RuntimeCacheConfig::from_env()?;
let cache_stores = RuntimeCacheStores::from_config(&cache_config).await?;
let api_rate_limit = mcp_api_rate_limit_config_from_env()?;
let database_options = database_options_from_env()?;
let registry =
PostgresRegistry::connect_with_options_and_pool_config(database_options, pool_config)
.await?;
if metrics_enabled {
spawn_postgres_pool_metrics(registry.pool().clone());
}
let session_store = PostgresTransportSessionStore::from_pool(registry.pool().clone()).await?;
let secret_crypto = SecretCrypto::new(&env::var("CRANK_MASTER_KEY")?)?;
let runtime = crank_runtime::community_from_env()?
let secret_crypto = SecretCrypto::new(config.runtime.master_key.expose_secret())?;
let outbound_http_policy = crank_runtime::OutboundHttpPolicy::try_new(
config.runtime.outbound.allowed_hosts.clone(),
config.runtime.outbound.denied_hosts.clone(),
config.runtime.outbound.max_response_bytes,
)?;
let runtime = crank_runtime::community_with_outbound_policy(outbound_http_policy)
.with_limits(runtime_limits)
.with_response_cache(cache_stores.response.clone())
.with_coordination_store(cache_stores.coordination.clone())
.build();
let app = build_app_with_background_workers_and_limits(
registry,
refresh_interval,
base_url,
Duration::from_millis(config.refresh_ms),
config.runtime.base_url.clone(),
secret_crypto,
runtime,
if cache_config.backend.is_external() {
@@ -93,7 +172,7 @@ async fn run(
std::sync::Arc::new(CommunityMachineCredentialVerifier),
runtime_limits.max_concurrent_sessions,
);
let listener = TcpListener::bind(socket_addr).await?;
let listener = TcpListener::bind(config.bind_addr).await?;
info!(
name: "mcp.postgres_pool.configured",
@@ -109,11 +188,7 @@ async fn run(
max_lifetime_ms = pool_config.max_lifetime_ms,
"postgres pool configured"
);
info!(
name: "mcp.server.listening",
bind_address = %socket_addr,
"mcp-server listening"
);
info!(name: "mcp.server.listening", bind_address = %config.bind_addr, "mcp-server listening");
*startup_completed = true;
if let Some(metrics_server) = metrics_server {
@@ -124,42 +199,176 @@ async fn run(
} else {
axum::serve(listener, app).await?;
}
Ok(())
}
fn database_options_from_env() -> Result<PgConnectOptions, Box<dyn std::error::Error>> {
if let Ok(database_url) = env::var("CRANK_DATABASE_URL") {
return Ok(database_url.parse::<PgConnectOptions>()?);
fn init_observability(
config: &ObservabilitySettings,
) -> Result<ObservabilityLifecycle, Box<dyn std::error::Error>> {
let identity = ServiceIdentity::try_new(
"mcp-server",
env!("CARGO_PKG_VERSION"),
config.environment.clone(),
)?;
let base = ObservabilityConfig::try_new(
identity,
config.log_filter.clone(),
RedactionLimits::default(),
)?;
let sentry = SentryConfig::parse(
config
.sentry_dsn
.as_ref()
.map(|value| value.expose_secret()),
)?;
let values = &config.otlp;
let otlp = OtlpTraceConfig::from_values(
values.endpoint.clone(),
values.traces_endpoint.clone(),
values.protocol.clone(),
values.traces_protocol.clone(),
values.timeout.clone(),
values.traces_timeout.clone(),
values
.headers
.as_ref()
.map(|value| value.expose_secret().to_owned()),
values
.traces_headers
.as_ref()
.map(|value| value.expose_secret().to_owned()),
values.max_queue_size,
values.max_export_batch_size,
values.schedule_delay.clone(),
values.export_timeout.clone(),
)?;
Ok(ObservabilityLifecycle::init_with_exporters(
base, sentry, otlp,
)?)
}
fn preflight_config(config: &McpProcessConfig) -> Result<(), crank_config::ConfigError> {
let invalid = |field| crank_config::ConfigError::single(DiagnosticCode::InvalidType, field);
database_options(&config.database).map_err(|_| invalid("database.source"))?;
postgres_pool_config(&config.database).map_err(|_| invalid("database.pool"))?;
MetricsConfig::new(
config.observability.metrics.enabled,
config.observability.metrics.bind_addr,
config
.observability
.metrics
.bearer_token
.as_ref()
.map(|v| v.expose_secret().to_owned()),
)
.map_err(|_| invalid("observability.metrics"))?;
RuntimeLimits::try_new(
config.runtime.max_concurrent_unary,
config.runtime.max_concurrent_sessions,
)
.map_err(|_| invalid("runtime.limits"))?;
runtime_cache_config(config).map_err(|_| invalid("cache"))?;
RequestRateLimitConfig::new(
config.rate_limit.requests_per_second,
config.rate_limit.burst,
)
.map_err(|_| invalid("mcp.rate_limit"))?;
SecretCrypto::new(config.runtime.master_key.expose_secret())
.map_err(|_| invalid("runtime.master_key"))?;
crank_runtime::OutboundHttpPolicy::try_new(
config.runtime.outbound.allowed_hosts.clone(),
config.runtime.outbound.denied_hosts.clone(),
config.runtime.outbound.max_response_bytes,
)
.map_err(|_| invalid("runtime.outbound"))?;
let identity = ServiceIdentity::try_new(
"mcp-server",
env!("CARGO_PKG_VERSION"),
config.observability.environment.clone(),
)
.map_err(|_| invalid("observability.environment"))?;
ObservabilityConfig::try_new(
identity,
config.observability.log_filter.clone(),
RedactionLimits::default(),
)
.map_err(|_| invalid("observability.log_filter"))?;
SentryConfig::parse(
config
.observability
.sentry_dsn
.as_ref()
.map(|v| v.expose_secret()),
)
.map_err(|_| invalid("observability.sentry_dsn"))?;
let values = &config.observability.otlp;
OtlpTraceConfig::from_values(
values.endpoint.clone(),
values.traces_endpoint.clone(),
values.protocol.clone(),
values.traces_protocol.clone(),
values.timeout.clone(),
values.traces_timeout.clone(),
values
.headers
.as_ref()
.map(|v| v.expose_secret().to_owned()),
values
.traces_headers
.as_ref()
.map(|v| v.expose_secret().to_owned()),
values.max_queue_size,
values.max_export_batch_size,
values.schedule_delay.clone(),
values.export_timeout.clone(),
)
.map_err(|_| invalid("observability.otlp"))?;
Ok(())
}
fn postgres_pool_config(
config: &DatabaseSettings,
) -> Result<PostgresPoolConfig, crank_registry::PostgresPoolConfigError> {
PostgresPoolConfig::try_new(
config.pool.max_connections,
config.pool.min_connections,
config.pool.acquire_timeout_ms,
config.pool.idle_timeout_ms,
config.pool.max_lifetime_ms,
)
}
fn database_options(
config: &DatabaseSettings,
) -> Result<PgConnectOptions, Box<dyn std::error::Error>> {
if let Some(url) = &config.url {
return url
.expose_secret()
.parse::<PgConnectOptions>()
.map_err(|_| io::Error::other("database URL is invalid").into());
}
let host = env::var("POSTGRES_HOST").unwrap_or_else(|_| "postgres".into());
let port = env::var("POSTGRES_PORT")
.ok()
.and_then(|value| value.parse::<u16>().ok())
.unwrap_or(5432);
let database = env::var("POSTGRES_DB").unwrap_or_else(|_| "crank".into());
let username = env::var("POSTGRES_USER").unwrap_or_else(|_| "crank".into());
let password = env::var("POSTGRES_PASSWORD").unwrap_or_else(|_| "crank".into());
Ok(PgConnectOptions::new()
.host(&host)
.port(port)
.database(&database)
.username(&username)
.password(&password))
.host(&config.host)
.port(config.port)
.database(&config.database)
.username(&config.username)
.password(config.password.expose_secret()))
}
fn mcp_api_rate_limit_config_from_env() -> Result<RequestRateLimitConfig, Box<dyn std::error::Error>>
{
let requests_per_second = env::var("CRANK_MCP_RATE_LIMIT_RPS")
.ok()
.and_then(|value| value.parse::<u32>().ok())
.unwrap_or(60);
let burst = env::var("CRANK_MCP_RATE_LIMIT_BURST")
.ok()
.and_then(|value| value.parse::<u32>().ok())
.unwrap_or(120);
Ok(RequestRateLimitConfig::new(requests_per_second, burst)?)
fn runtime_cache_config(
config: &McpProcessConfig,
) -> Result<RuntimeCacheConfig, crank_runtime::RuntimeCacheConfigError> {
RuntimeCacheConfig::try_new(
match config.runtime.cache.backend {
ConfigCacheBackend::Memory => CacheBackend::Memory,
ConfigCacheBackend::Valkey => CacheBackend::Valkey,
ConfigCacheBackend::Redis => CacheBackend::Redis,
},
config
.runtime
.cache
.url
.as_ref()
.map(|value| value.expose_secret().to_owned()),
)
}