наблюдаемость: завершить базовый контур Community
Добавить структурированные журналы, метрики, трассировку и безопасный канал критических ошибок. Усилить границы рантайма, тесты, проверку зависимостей и сценарии развёртывания.
This commit is contained in:
+248
-47
@@ -12,16 +12,18 @@ use crank_core::{
|
||||
};
|
||||
use crank_mapping::{MappingRule, MappingSet};
|
||||
use crank_registry::{
|
||||
AgentSummary, CreateInvocationLogRequest, OperationAgentRef, OperationSummary,
|
||||
OperationUsageSummary, PostgresRegistry, RegistryOperation, UsageBucket,
|
||||
AgentSummary, CreateInvocationLogRequest, InvocationHistoryWriteOutcome, OperationAgentRef,
|
||||
OperationSummary, OperationUsageSummary, PostgresRegistry, RegistryOperation, UsageBucket,
|
||||
};
|
||||
use crank_runtime::{
|
||||
OutboundHttpPolicy, PreparedRequest, ResolvedAuth, RuntimeError, RuntimeExecutor, SecretCrypto,
|
||||
};
|
||||
use crank_schema::{Schema, SchemaKind};
|
||||
use crank_trace::{DbOperation, ErrorCategory, Stage, StageOutcome, observe_db_query};
|
||||
use serde_json::{Value, json};
|
||||
use sha2::{Digest, Sha256};
|
||||
use time::{OffsetDateTime, format_description::well_known::Rfc3339};
|
||||
use tracing::Instrument;
|
||||
use uuid::Uuid;
|
||||
|
||||
mod agents;
|
||||
@@ -74,6 +76,11 @@ pub struct AdminServiceBuilder {
|
||||
pub use crate::dto::*;
|
||||
|
||||
impl AdminService {
|
||||
pub async fn readiness(&self) -> Result<(), ApiError> {
|
||||
self.registry.ping().await?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
pub fn new(
|
||||
registry: PostgresRegistry,
|
||||
@@ -199,16 +206,33 @@ impl AdminServiceBuilder {
|
||||
}
|
||||
|
||||
impl AdminService {
|
||||
pub async fn export_workspace(
|
||||
pub async fn export_workspace_catalog_snapshot(
|
||||
&self,
|
||||
workspace_id: &WorkspaceId,
|
||||
) -> Result<WorkspaceExportResponse, ApiError> {
|
||||
) -> Result<WorkspaceCatalogSnapshotResponse, ApiError> {
|
||||
let workspace = self.get_workspace(workspace_id).await?;
|
||||
let operations = self.list_operations(workspace_id).await?;
|
||||
let agents = self.list_agents(workspace_id).await?;
|
||||
let platform_api_keys = self.registry.list_platform_api_keys(workspace_id).await?;
|
||||
|
||||
Ok(WorkspaceExportResponse {
|
||||
Ok(WorkspaceCatalogSnapshotResponse {
|
||||
kind: "workspace_catalog_snapshot".to_owned(),
|
||||
format_version: "1".to_owned(),
|
||||
restorable: false,
|
||||
included: vec![
|
||||
"workspace_settings".to_owned(),
|
||||
"operation_summaries".to_owned(),
|
||||
"agent_summaries".to_owned(),
|
||||
"platform_api_key_metadata".to_owned(),
|
||||
],
|
||||
excluded: vec![
|
||||
"operation_versions_and_samples".to_owned(),
|
||||
"agent_versions_and_bindings".to_owned(),
|
||||
"secret_metadata_and_values".to_owned(),
|
||||
"secret_values".to_owned(),
|
||||
"invocation_logs_and_usage".to_owned(),
|
||||
"authentication_sessions".to_owned(),
|
||||
],
|
||||
workspace,
|
||||
operations,
|
||||
agents,
|
||||
@@ -239,9 +263,13 @@ impl AdminService {
|
||||
return Ok(None);
|
||||
};
|
||||
|
||||
let auth_profile = self
|
||||
.registry
|
||||
.get_auth_profile(workspace_id, auth_profile_id)
|
||||
let span = Stage::AuthResolve.span();
|
||||
let result = async {
|
||||
let auth_profile = observe_db_query(
|
||||
DbOperation::AuthProfileRead,
|
||||
self.registry
|
||||
.get_auth_profile(workspace_id, auth_profile_id),
|
||||
)
|
||||
.await
|
||||
.map_err(|error| RuntimeError::SecretCrypto {
|
||||
operation: "load auth profile",
|
||||
@@ -251,9 +279,20 @@ impl AdminService {
|
||||
auth_profile_id: auth_profile_id.as_str().to_owned(),
|
||||
})?;
|
||||
|
||||
self.resolve_auth_profile(workspace_id, &auth_profile)
|
||||
.await
|
||||
.map(Some)
|
||||
self.resolve_auth_profile(workspace_id, &auth_profile)
|
||||
.await
|
||||
.map(Some)
|
||||
}
|
||||
.instrument(span.clone())
|
||||
.await;
|
||||
match &result {
|
||||
Ok(_) => StageOutcome::Success.record(&span),
|
||||
Err(_) => {
|
||||
StageOutcome::Error.record(&span);
|
||||
ErrorCategory::Configuration.record(&span);
|
||||
}
|
||||
}
|
||||
result
|
||||
}
|
||||
|
||||
async fn resolve_auth_profile(
|
||||
@@ -265,40 +304,46 @@ impl AdminService {
|
||||
let used_at = OffsetDateTime::now_utc();
|
||||
|
||||
for secret_id in auth_profile.config.secret_ids() {
|
||||
let secret = self
|
||||
.registry
|
||||
.get_secret(workspace_id, secret_id)
|
||||
.await
|
||||
.map_err(|error| RuntimeError::SecretCrypto {
|
||||
operation: "load secret",
|
||||
details: error.to_string(),
|
||||
})?
|
||||
.ok_or_else(|| RuntimeError::MissingSecret {
|
||||
secret_id: secret_id.as_str().to_owned(),
|
||||
})?;
|
||||
let version = self
|
||||
.registry
|
||||
.get_current_secret_version(workspace_id, secret_id)
|
||||
.await
|
||||
.map_err(|error| RuntimeError::SecretCrypto {
|
||||
operation: "load current secret version",
|
||||
details: error.to_string(),
|
||||
})?
|
||||
.ok_or_else(|| RuntimeError::MissingSecretVersion {
|
||||
secret_id: secret_id.as_str().to_owned(),
|
||||
version: secret.secret.current_version,
|
||||
})?;
|
||||
let secret = observe_db_query(
|
||||
DbOperation::SecretRead,
|
||||
self.registry.get_secret(workspace_id, secret_id),
|
||||
)
|
||||
.await
|
||||
.map_err(|error| RuntimeError::SecretCrypto {
|
||||
operation: "load secret",
|
||||
details: error.to_string(),
|
||||
})?
|
||||
.ok_or_else(|| RuntimeError::MissingSecret {
|
||||
secret_id: secret_id.as_str().to_owned(),
|
||||
})?;
|
||||
let version = observe_db_query(
|
||||
DbOperation::SecretRead,
|
||||
self.registry
|
||||
.get_current_secret_version(workspace_id, secret_id),
|
||||
)
|
||||
.await
|
||||
.map_err(|error| RuntimeError::SecretCrypto {
|
||||
operation: "load current secret version",
|
||||
details: error.to_string(),
|
||||
})?
|
||||
.ok_or_else(|| RuntimeError::MissingSecretVersion {
|
||||
secret_id: secret_id.as_str().to_owned(),
|
||||
version: secret.secret.current_version,
|
||||
})?;
|
||||
let plaintext = self.secret_crypto.decrypt(
|
||||
&version.secret_version.key_version,
|
||||
&version.secret_version.ciphertext,
|
||||
)?;
|
||||
self.registry
|
||||
.touch_secret(workspace_id, secret_id, &used_at)
|
||||
.await
|
||||
.map_err(|error| RuntimeError::SecretCrypto {
|
||||
operation: "touch secret",
|
||||
details: error.to_string(),
|
||||
})?;
|
||||
observe_db_query(
|
||||
DbOperation::SecretTouch,
|
||||
self.registry
|
||||
.touch_secret(workspace_id, secret_id, &used_at),
|
||||
)
|
||||
.await
|
||||
.map_err(|error| RuntimeError::SecretCrypto {
|
||||
operation: "touch secret",
|
||||
details: error.to_string(),
|
||||
})?;
|
||||
secrets.insert(secret_id.clone(), plaintext);
|
||||
}
|
||||
|
||||
@@ -412,7 +457,7 @@ impl AdminService {
|
||||
async fn record_invocation(
|
||||
&self,
|
||||
request: InvocationRecordRequest<'_>,
|
||||
) -> Result<(), ApiError> {
|
||||
) -> InvocationHistoryWriteOutcome {
|
||||
let log = InvocationLog {
|
||||
id: InvocationLogId::new(new_prefixed_id("log")),
|
||||
workspace_id: request.workspace_id.clone(),
|
||||
@@ -432,11 +477,70 @@ impl AdminService {
|
||||
created_at: OffsetDateTime::now_utc(),
|
||||
};
|
||||
|
||||
self.registry
|
||||
.create_invocation_log(CreateInvocationLogRequest { log: &log })
|
||||
.await?;
|
||||
let history_span = crank_trace::Stage::HistoryWrite.span();
|
||||
let (outcome, db_span) = async {
|
||||
let db_span = crank_trace::Stage::DbQuery
|
||||
.db_span(crank_trace::DbOperation::InvocationHistoryWrite)
|
||||
.expect("database stage");
|
||||
let outcome = self
|
||||
.registry
|
||||
.create_invocation_log(CreateInvocationLogRequest { log: &log })
|
||||
.instrument(db_span.clone())
|
||||
.await;
|
||||
(outcome, db_span)
|
||||
}
|
||||
.instrument(history_span.clone())
|
||||
.await;
|
||||
match outcome {
|
||||
InvocationHistoryWriteOutcome::Recorded => {
|
||||
crank_trace::StageOutcome::Success.record(&db_span);
|
||||
crank_trace::StageOutcome::Success.record(&history_span);
|
||||
}
|
||||
InvocationHistoryWriteOutcome::Lost(_) => {
|
||||
crank_trace::StageOutcome::Error.record(&db_span);
|
||||
crank_trace::ErrorCategory::Database.record(&db_span);
|
||||
crank_trace::StageOutcome::Error.record(&history_span);
|
||||
crank_trace::ErrorCategory::History.record(&history_span);
|
||||
}
|
||||
}
|
||||
drop(db_span);
|
||||
drop(history_span);
|
||||
observe_invocation_history_outcome(
|
||||
outcome,
|
||||
request.request_id,
|
||||
request.status,
|
||||
"admin_test_run",
|
||||
);
|
||||
outcome
|
||||
}
|
||||
}
|
||||
|
||||
Ok(())
|
||||
fn observe_invocation_history_outcome(
|
||||
outcome: InvocationHistoryWriteOutcome,
|
||||
request_id: Option<&str>,
|
||||
status: crank_core::InvocationStatus,
|
||||
source: &'static str,
|
||||
) {
|
||||
let Some(loss) = outcome.loss() else {
|
||||
return;
|
||||
};
|
||||
crank_observability::record_operational_incident(
|
||||
crank_observability::OperationalIncident::InvocationHistoryLost,
|
||||
);
|
||||
tracing::warn!(
|
||||
name: "admin.invocation_history.lost",
|
||||
request_id = request_id.unwrap_or_default(),
|
||||
source,
|
||||
invocation_status = invocation_status_label(status),
|
||||
error_category = loss.category.as_str(),
|
||||
"invocation history was not recorded"
|
||||
);
|
||||
}
|
||||
|
||||
fn invocation_status_label(status: crank_core::InvocationStatus) -> &'static str {
|
||||
match status {
|
||||
crank_core::InvocationStatus::Ok => "ok",
|
||||
crank_core::InvocationStatus::Error => "error",
|
||||
}
|
||||
}
|
||||
|
||||
@@ -543,6 +647,10 @@ fn runtime_error_code(error: &RuntimeError) -> &'static str {
|
||||
RuntimeError::ConfirmationRequired { .. } => "confirmation_required",
|
||||
RuntimeError::InvalidConfirmationToken { .. } => "invalid_confirmation_token",
|
||||
RuntimeError::ConfirmationStoreUnavailable { .. } => "confirmation_unavailable",
|
||||
RuntimeError::IdempotencyStoreUnavailable { .. } => "idempotency_unavailable",
|
||||
RuntimeError::IdempotencyInProgress { .. } => "idempotency_in_progress",
|
||||
RuntimeError::IdempotencyConflict { .. } => "idempotency_conflict",
|
||||
RuntimeError::IdempotencyOutcomeUnknown { .. } => "idempotency_outcome_unknown",
|
||||
RuntimeError::RestAdapter(_) => "rest_error",
|
||||
RuntimeError::ProtocolAdapter(_) => "adapter_error",
|
||||
RuntimeError::UnsupportedProtocol { .. } => "unsupported_protocol",
|
||||
@@ -761,7 +869,25 @@ fn tool_quality_mapping_rule(rule: &MappingRule) -> ToolQualityMappingRule {
|
||||
#[cfg(test)]
|
||||
#[allow(clippy::items_after_test_module)]
|
||||
mod tests {
|
||||
use super::{validate_profile_display_name, validate_profile_email};
|
||||
use std::{
|
||||
io,
|
||||
sync::{Arc, Mutex},
|
||||
};
|
||||
|
||||
use crank_core::InvocationStatus;
|
||||
use crank_observability::{
|
||||
ObservabilityConfig, OperationalIncident, RedactionLimits, ServiceIdentity,
|
||||
operational_incident_total,
|
||||
};
|
||||
use crank_registry::{
|
||||
InvocationHistoryLoss, InvocationHistoryLossCategory, InvocationHistoryWriteOutcome,
|
||||
};
|
||||
use serde_json::Value;
|
||||
use tracing_subscriber::fmt::MakeWriter;
|
||||
|
||||
use super::{
|
||||
observe_invocation_history_outcome, validate_profile_display_name, validate_profile_email,
|
||||
};
|
||||
|
||||
#[test]
|
||||
fn validates_profile_identity_fields() {
|
||||
@@ -782,6 +908,81 @@ mod tests {
|
||||
assert!(validate_profile_display_name(&"x".repeat(81)).is_err());
|
||||
assert!(validate_profile_email("owner <html>@crank.local").is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn emits_bounded_history_loss_incident() {
|
||||
let writer = SharedLogWriter::default();
|
||||
let subscriber = crank_observability::build_subscriber(
|
||||
ObservabilityConfig::new(
|
||||
ServiceIdentity::try_new("admin-api", "test", "test").unwrap(),
|
||||
"info",
|
||||
RedactionLimits::default(),
|
||||
),
|
||||
writer.clone(),
|
||||
)
|
||||
.unwrap();
|
||||
let before = operational_incident_total(OperationalIncident::InvocationHistoryLost);
|
||||
let dispatch = tracing::Dispatch::new(subscriber);
|
||||
let _guard = tracing::dispatcher::set_default(&dispatch);
|
||||
|
||||
observe_invocation_history_outcome(
|
||||
InvocationHistoryWriteOutcome::Lost(InvocationHistoryLoss {
|
||||
category: InvocationHistoryLossCategory::InvalidRecord,
|
||||
}),
|
||||
Some("req_admin_dc08"),
|
||||
InvocationStatus::Error,
|
||||
"admin_test_run",
|
||||
);
|
||||
|
||||
let output = writer.output();
|
||||
assert!(!output.contains("dc08-canary-secret"));
|
||||
let event: Value = output
|
||||
.lines()
|
||||
.map(|line| serde_json::from_str(line).unwrap())
|
||||
.find(|event: &Value| event["event"] == "admin.invocation_history.lost")
|
||||
.unwrap();
|
||||
assert_eq!(event["request_id"], "req_admin_dc08");
|
||||
assert_eq!(event["fields"]["source"], "admin_test_run");
|
||||
assert_eq!(event["fields"]["invocation_status"], "error");
|
||||
assert_eq!(event["fields"]["error_category"], "invalid_record");
|
||||
assert!(operational_incident_total(OperationalIncident::InvocationHistoryLost) > before);
|
||||
}
|
||||
|
||||
#[derive(Clone, Default)]
|
||||
struct SharedLogWriter {
|
||||
buffer: Arc<Mutex<Vec<u8>>>,
|
||||
}
|
||||
|
||||
impl SharedLogWriter {
|
||||
fn output(&self) -> String {
|
||||
String::from_utf8(self.buffer.lock().unwrap().clone()).unwrap()
|
||||
}
|
||||
}
|
||||
|
||||
impl<'a> MakeWriter<'a> for SharedLogWriter {
|
||||
type Writer = SharedLogGuard;
|
||||
|
||||
fn make_writer(&'a self) -> Self::Writer {
|
||||
SharedLogGuard {
|
||||
buffer: Arc::clone(&self.buffer),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
struct SharedLogGuard {
|
||||
buffer: Arc<Mutex<Vec<u8>>>,
|
||||
}
|
||||
|
||||
impl io::Write for SharedLogGuard {
|
||||
fn write(&mut self, bytes: &[u8]) -> io::Result<usize> {
|
||||
self.buffer.lock().unwrap().extend_from_slice(bytes);
|
||||
Ok(bytes.len())
|
||||
}
|
||||
|
||||
fn flush(&mut self) -> io::Result<()> {
|
||||
Ok(())
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
fn enrich_operation_summary(
|
||||
|
||||
Reference in New Issue
Block a user