Struct QueryEngine 

pub struct QueryEngine { /* private fields */ }

Read-only query interface to the storage database.

Implementations

impl QueryEngine

pub fn clients(&self, params: &ClientsParams) -> Result<Vec<ClientRow>, Error>

Aggregate client usage across sessions, sorted by total calls descending.

impl QueryEngine

pub fn logs(&self, params: &LogsParams) -> Result<Vec<LogRow>, Error>

Fetch recent request logs, newest first.

pub fn logs_since( &self, params: &LogsParams, after_ts: i64, ) -> Result<Vec<LogRow>, Error>

Fetch logs newer than a given timestamp, oldest first.

Used for --follow mode: poll every 500ms with the last seen timestamp.

impl QueryEngine

pub fn schema(&self, params: &SchemaParams) -> Result<Vec<SchemaRow>, Error>

Fetch all captured schema snapshots, optionally filtered by proxy name and/or MCP method.

pub fn schema_changes( &self, params: &SchemaChangesParams, ) -> Result<Vec<SchemaChangeRow>, Error>

Fetch schema change history, optionally filtered by proxy name and/or MCP method.

pub fn schema_status( &self, upstream_url: &str, ) -> Result<SchemaStatusRow, Error>

Compute the schema status for a given upstream URL.

pub fn latest_schema_row( &self, proxy: &str, method: &str, ) -> Result<Option<LatestSchemaRow>, Error>

Fetch the latest persisted schema row for (proxy, method) — used at proxy startup to hydrate SchemaManager from disk.

Returns None when no row matches. server_schema is an UPSERT-keyed-by-(proxy, upstream_url, method) table, so the “latest” is simply the single row per key; we pick by highest captured_at to be robust against multiple upstream URLs that share the same proxy name (unlikely today but possible).

pub fn schema_unused( &self, params: &SchemaUnusedParams, ) -> Result<Vec<SchemaToolUsageRow>, Error>

Cross-reference captured tools/list schema with actual request logs. Returns all listed tools with their usage stats — unused tools have calls = 0.

impl QueryEngine

pub fn session_detail( &self, session_id: &str, ) -> Result<Option<SessionDetail>, Error>

Fetch a single session by ID, along with all its request logs.

Returns None if the session doesn’t exist.

impl QueryEngine

pub fn sessions( &self, params: &SessionsParams, ) -> Result<Vec<SessionRow>, Error>

List sessions for a proxy, most recently seen first.

impl QueryEngine

pub fn slow(&self, params: &SlowParams) -> Result<Vec<LogRow>, Error>

Fetch the slowest requests above a latency threshold, slowest first.

pub fn slow_since( &self, params: &SlowParams, after_ts: i64, ) -> Result<Vec<LogRow>, Error>

Fetch slow calls newer than a given timestamp (for --tail live stream).

impl QueryEngine

pub fn stats(&self, params: &StatsParams) -> Result<StatsResult, Error>

Compute per-tool aggregated stats for a proxy within a time window.

Percentiles are computed in Rust (load latency values into a Vec and sort) because SQLite has no native percentile function. This is fine for the expected data volumes (<1M rows per proxy).

impl QueryEngine

pub fn store_stats(&self, db_path: &Path) -> Result<StoreStats, Error>

Get database-level statistics.

pub fn vacuum(&self, params: &VacuumParams) -> Result<VacuumResult, Error>

Delete old requests and orphaned sessions, optionally scoped to one proxy. In dry-run mode, returns counts without deleting.

impl QueryEngine

pub fn open(db_path: &Path) -> Result<Self, Error>

Open a query connection to the database at the given path.

Runs migrations on open. The writer also migrates on Store::open, but a user who upgrades the binary and runs a read command before restarting the proxy would otherwise hit “no such column” errors against a stale schema. Migrations are idempotent + WAL-safe, so it’s fine for the reader to bump the schema if the writer hasn’t caught up yet.