From 4aae302df58418b773bfae6fe34becda38e92934 Mon Sep 17 00:00:00 2001
From: boqiu <82121246@qq.com>
Date: Tue, 20 Aug 2024 18:50:17 +0800
Subject: [PATCH] add more metrics for serial file sync

---
 node/sync/src/controllers/metrics.rs | 4 +++-
 node/sync/src/controllers/serial.rs  | 8 ++++++++
 2 files changed, 11 insertions(+), 1 deletion(-)

diff --git a/node/sync/src/controllers/metrics.rs b/node/sync/src/controllers/metrics.rs
index 85e231f..4e6df22 100644
--- a/node/sync/src/controllers/metrics.rs
+++ b/node/sync/src/controllers/metrics.rs
@@ -1,8 +1,10 @@
 use std::sync::Arc;
 
-use metrics::{register_timer, Timer};
+use metrics::{register_timer, Counter, CounterUsize, Histogram, Sample, Timer};
 
 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_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_UNEXPECTED_ERRORS: Arc<dyn Counter<usize>> = CounterUsize::register("sync_controllers_serial_sync_unexpected_errors");
 }
diff --git a/node/sync/src/controllers/serial.rs b/node/sync/src/controllers/serial.rs
index cb4f46f..dd23e70 100644
--- a/node/sync/src/controllers/serial.rs
+++ b/node/sync/src/controllers/serial.rs
@@ -428,6 +428,7 @@ impl SerialSyncController {
         let data_len = response.chunks.data.len();
         if data_len == 0 || data_len % CHUNK_SIZE > 0 {
             warn!(%from_peer_id, %self.tx_seq, %data_len, "Invalid chunk response data length");
+            metrics::SERIAL_SYNC_UNEXPECTED_ERRORS.inc(1);
             self.ban_peer(from_peer_id, "Invalid chunk response data length");
             self.state = SyncState::Idle;
             return;
@@ -465,6 +466,7 @@ impl SerialSyncController {
             }
             Err(err) => {
                 warn!(%err, %self.tx_seq, "Failed to validate chunks response");
+                metrics::SERIAL_SYNC_UNEXPECTED_ERRORS.inc(1);
                 self.ban_peer(from_peer_id, "Chunk array validation failed");
                 self.state = SyncState::Idle;
                 return;
@@ -473,6 +475,8 @@ impl SerialSyncController {
 
         self.failures = 0;
 
+        metrics::SERIAL_SYNC_SEGMENT_LATENCY.update_since(since.0);
+
         let shard_config = self.store.get_store().flow().get_shard_config();
         let next_chunk = shard_config.next_segment_index(
             (from_chunk / PORA_CHUNK_SIZE as u64) as usize,
@@ -487,6 +491,7 @@ impl SerialSyncController {
             Ok(true) => self.next_chunk = next_chunk as u64,
             Ok(false) => {
                 warn!(%self.tx_seq, ?self.tx_id, "Transaction reverted while storing chunks");
+                metrics::SERIAL_SYNC_UNEXPECTED_ERRORS.inc(1);
                 self.state = SyncState::Failed {
                     reason: FailureReason::TxReverted(self.tx_id),
                 };
@@ -494,6 +499,7 @@ impl SerialSyncController {
             }
             Err(err) => {
                 error!(%err, %self.tx_seq, "Unexpected DB error while storing chunks");
+                metrics::SERIAL_SYNC_UNEXPECTED_ERRORS.inc(1);
                 self.state = SyncState::Failed {
                     reason: FailureReason::DBError(err.to_string()),
                 };
@@ -527,12 +533,14 @@ impl SerialSyncController {
             }
             Ok(false) => {
                 warn!(?self.tx_id, %self.tx_seq, "Transaction reverted during finalize_tx");
+                metrics::SERIAL_SYNC_UNEXPECTED_ERRORS.inc(1);
                 self.state = SyncState::Failed {
                     reason: FailureReason::TxReverted(self.tx_id),
                 };
             }
             Err(err) => {
                 error!(%err, %self.tx_seq, "Unexpected error during finalize_tx");
+                metrics::SERIAL_SYNC_UNEXPECTED_ERRORS.inc(1);
                 self.state = SyncState::Failed {
                     reason: FailureReason::DBError(err.to_string()),
                 };