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:
@@ -0,0 +1,117 @@
|
||||
use std::{sync::Arc, time::Duration};
|
||||
|
||||
use crank_core::{
|
||||
InvocationLevel, InvocationLog, InvocationLogId, InvocationSource, InvocationStatus,
|
||||
};
|
||||
use crank_registry::{
|
||||
CreateInvocationLogRequest, InvocationHistoryWriteOutcome, PublishedAgentTool,
|
||||
};
|
||||
use crank_trace::{DbOperation, ErrorCategory, Stage, StageOutcome};
|
||||
use serde_json::Value;
|
||||
use time::OffsetDateTime;
|
||||
use tracing::{Instrument, warn};
|
||||
|
||||
use super::AppState;
|
||||
|
||||
pub(crate) struct InvocationRecord<'a> {
|
||||
pub(crate) request_id: Option<&'a str>,
|
||||
pub(crate) tool_name: &'a str,
|
||||
pub(crate) status: InvocationStatus,
|
||||
pub(crate) level: InvocationLevel,
|
||||
pub(crate) message: &'a str,
|
||||
pub(crate) status_code: Option<u16>,
|
||||
pub(crate) error_kind: Option<&'a str>,
|
||||
pub(crate) duration: Duration,
|
||||
pub(crate) request_preview: Value,
|
||||
pub(crate) response_preview: Value,
|
||||
}
|
||||
|
||||
pub(crate) async fn persist_invocation(
|
||||
state: &Arc<AppState>,
|
||||
tool: &PublishedAgentTool,
|
||||
record: InvocationRecord<'_>,
|
||||
) -> InvocationHistoryWriteOutcome {
|
||||
let log = InvocationLog {
|
||||
id: InvocationLogId::new(format!("log_{}", uuid::Uuid::now_v7().simple())),
|
||||
workspace_id: tool.workspace_id.clone(),
|
||||
agent_id: Some(tool.agent_id.clone()),
|
||||
operation_id: tool.operation.id.clone(),
|
||||
source: InvocationSource::AgentToolCall,
|
||||
level: record.level,
|
||||
status: record.status,
|
||||
tool_name: record.tool_name.to_owned(),
|
||||
message: record.message.to_owned(),
|
||||
request_id: record.request_id.map(ToOwned::to_owned),
|
||||
status_code: record.status_code,
|
||||
duration_ms: u64::try_from(record.duration.as_millis()).unwrap_or(u64::MAX),
|
||||
error_kind: record.error_kind.map(ToOwned::to_owned),
|
||||
request_preview: record.request_preview,
|
||||
response_preview: record.response_preview,
|
||||
created_at: OffsetDateTime::now_utc(),
|
||||
};
|
||||
|
||||
let history_span = Stage::HistoryWrite.span();
|
||||
let (outcome, db_span) = async {
|
||||
let db_span = Stage::DbQuery
|
||||
.db_span(DbOperation::InvocationHistoryWrite)
|
||||
.expect("database stage");
|
||||
let outcome = state
|
||||
.registry
|
||||
.create_invocation_log(CreateInvocationLogRequest { log: &log })
|
||||
.instrument(db_span.clone())
|
||||
.await;
|
||||
(outcome, db_span)
|
||||
}
|
||||
.instrument(history_span.clone())
|
||||
.await;
|
||||
match outcome {
|
||||
InvocationHistoryWriteOutcome::Recorded => {
|
||||
StageOutcome::Success.record(&db_span);
|
||||
StageOutcome::Success.record(&history_span);
|
||||
}
|
||||
InvocationHistoryWriteOutcome::Lost(_) => {
|
||||
StageOutcome::Error.record(&db_span);
|
||||
ErrorCategory::Database.record(&db_span);
|
||||
StageOutcome::Error.record(&history_span);
|
||||
ErrorCategory::History.record(&history_span);
|
||||
}
|
||||
}
|
||||
drop(db_span);
|
||||
drop(history_span);
|
||||
observe_invocation_history_outcome(
|
||||
outcome,
|
||||
record.request_id,
|
||||
record.status,
|
||||
"agent_tool_call",
|
||||
);
|
||||
outcome
|
||||
}
|
||||
|
||||
pub(super) fn observe_invocation_history_outcome(
|
||||
outcome: InvocationHistoryWriteOutcome,
|
||||
request_id: Option<&str>,
|
||||
status: InvocationStatus,
|
||||
source: &'static str,
|
||||
) {
|
||||
let Some(loss) = outcome.loss() else {
|
||||
return;
|
||||
};
|
||||
crank_observability::record_operational_incident(
|
||||
crank_observability::OperationalIncident::InvocationHistoryLost,
|
||||
);
|
||||
warn!(
|
||||
name: "mcp.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: InvocationStatus) -> &'static str {
|
||||
match status {
|
||||
InvocationStatus::Ok => "ok",
|
||||
InvocationStatus::Error => "error",
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user