add timeout metrics for segment sync

This commit is contained in:
boqiu 2024-08-22 09:56:48 +08:00
parent e7f2d46cd4
commit 84a3cc3cbb
2 changed files with 2 additions and 0 deletions

View File

@ -6,5 +6,6 @@ lazy_static::lazy_static! {
pub static ref SERIAL_SYNC_FILE_COMPLETED: Arc<dyn Timer> = register_timer("sync_controllers_serial_sync_file_completed"); pub static ref SERIAL_SYNC_FILE_COMPLETED: Arc<dyn Timer> = register_timer("sync_controllers_serial_sync_file_completed");
pub static ref SERIAL_SYNC_CHUNKS_COMPLETED: Arc<dyn Timer> = register_timer("sync_controllers_serial_sync_chunks_completed"); pub static ref SERIAL_SYNC_CHUNKS_COMPLETED: Arc<dyn Timer> = register_timer("sync_controllers_serial_sync_chunks_completed");
pub static ref SERIAL_SYNC_SEGMENT_LATENCY: Arc<dyn Histogram> = Sample::ExpDecay(0.015).register("sync_controllers_serial_sync_segment_latency", 1024); pub static ref SERIAL_SYNC_SEGMENT_LATENCY: Arc<dyn Histogram> = Sample::ExpDecay(0.015).register("sync_controllers_serial_sync_segment_latency", 1024);
pub static ref SERIAL_SYNC_SEGMENT_TIMEOUT: Arc<dyn Counter<usize>> = CounterUsize::register("sync_controllers_serial_sync_segment_timeout");
pub static ref SERIAL_SYNC_UNEXPECTED_ERRORS: Arc<dyn Counter<usize>> = CounterUsize::register("sync_controllers_serial_sync_unexpected_errors"); pub static ref SERIAL_SYNC_UNEXPECTED_ERRORS: Arc<dyn Counter<usize>> = CounterUsize::register("sync_controllers_serial_sync_unexpected_errors");
} }

View File

@ -684,6 +684,7 @@ impl SerialSyncController {
debug!(%self.tx_seq, "No peer to continue downloading and try to find other peers to download"); debug!(%self.tx_seq, "No peer to continue downloading and try to find other peers to download");
self.state = SyncState::Idle; self.state = SyncState::Idle;
} else if since.elapsed() >= self.config.peer_chunks_download_timeout { } else if since.elapsed() >= self.config.peer_chunks_download_timeout {
metrics::SERIAL_SYNC_SEGMENT_TIMEOUT.inc(1);
self.handle_response_failure(peer_id, "RPC timeout"); self.handle_response_failure(peer_id, "RPC timeout");
} else { } else {
completed = true; completed = true;