This is an automated email from the ASF dual-hosted git repository.
morningman pushed a commit to branch branch-4.0
in repository https://gitbox.apache.org/repos/asf/doris.git
The following commit(s) were added to refs/heads/branch-4.0 by this push:
new 15eca4c5db1 branch-4.0: [fix](cloud) Skip skewed warmup rowset latency
samples (#62941) (#63132)
15eca4c5db1 is described below
commit 15eca4c5db1c8fcc40ec69c6600b75ef86fe58e1
Author: bobhan1 <[email protected]>
AuthorDate: Thu May 28 10:20:40 2026 +0800
branch-4.0: [fix](cloud) Skip skewed warmup rowset latency samples (#62941)
(#63132)
pick https://github.com/apache/doris/pull/62941
---
be/src/cloud/cloud_internal_service.cpp | 66 ++++++++++++++++++++++-----------
1 file changed, 44 insertions(+), 22 deletions(-)
diff --git a/be/src/cloud/cloud_internal_service.cpp
b/be/src/cloud/cloud_internal_service.cpp
index 7b6514455eb..45cc2016b25 100644
--- a/be/src/cloud/cloud_internal_service.cpp
+++ b/be/src/cloud/cloud_internal_service.cpp
@@ -20,6 +20,7 @@
#include <bthread/countdown_event.h>
#include <algorithm>
+#include <optional>
#include <thread>
#include "cloud/cloud_storage_engine.h"
@@ -89,6 +90,22 @@ FileCacheType cache_type_to_pb(io::FileCacheType type) {
return FileCacheType::NORMAL;
}
+static int64_t current_unix_time_us() {
+ return std::chrono::duration_cast<std::chrono::microseconds>(
+ std::chrono::system_clock::now().time_since_epoch())
+ .count();
+}
+
+static std::optional<int64_t> warm_up_rowset_cross_host_latency_us(int64_t
start_unix_ts_us,
+ int64_t
end_unix_ts_us) {
+ // The start timestamp is generated by the caller BE. Mixed-version
callers may omit it, and
+ // system clocks across BEs are not guaranteed to be ordered.
+ if (start_unix_ts_us <= 0 || end_unix_ts_us < start_unix_ts_us) {
+ return std::nullopt;
+ }
+ return end_unix_ts_us - start_unix_ts_us;
+}
+
void CloudInternalServiceImpl::get_file_cache_meta_by_tablet_id(
google::protobuf::RpcController* controller [[maybe_unused]],
const PGetFileCacheMetaRequest* request, PGetFileCacheMetaResponse*
response,
@@ -436,15 +453,18 @@ void
CloudInternalServiceImpl::warm_up_rowset(google::protobuf::RpcController* c
auto tablet = res.value();
auto tablet_meta = tablet->tablet_meta();
- int64_t handle_ts =
std::chrono::duration_cast<std::chrono::microseconds>(
-
std::chrono::system_clock::now().time_since_epoch())
- .count();
+ int64_t handle_ts = current_unix_time_us();
g_file_cache_warm_up_rowset_last_handle_unix_ts.set_value(handle_ts);
int64_t request_ts = request->has_unix_ts_us() ? request->unix_ts_us()
: 0;
- g_file_cache_warm_up_rowset_request_to_handle_latency << (handle_ts -
request_ts);
- if (request_ts > 0 && handle_ts - request_ts >
config::warm_up_rowset_slow_log_ms * 1000) {
+ auto request_to_handle_latency_us =
+ warm_up_rowset_cross_host_latency_us(request_ts, handle_ts);
+ if (request_to_handle_latency_us.has_value()) {
+ g_file_cache_warm_up_rowset_request_to_handle_latency <<
*request_to_handle_latency_us;
+ }
+ if (request_to_handle_latency_us.has_value() &&
+ *request_to_handle_latency_us > config::warm_up_rowset_slow_log_ms
* 1000) {
g_file_cache_warm_up_rowset_request_to_handle_slow_count << 1;
- LOG(INFO) << "warm up rowset (request to handle) took " <<
handle_ts - request_ts
+ LOG(INFO) << "warm up rowset (request to handle) took " <<
*request_to_handle_latency_us
<< " us, tablet_id: " << rs_meta.tablet_id()
<< ", rowset_id: " << rowset_id.to_string();
}
@@ -478,18 +498,19 @@ void
CloudInternalServiceImpl::warm_up_rowset(google::protobuf::RpcController* c
if (st.ok()) {
g_file_cache_event_driven_warm_up_finished_segment_num
<< 1;
g_file_cache_event_driven_warm_up_finished_segment_size << segment_size;
- int64_t now_ts =
-
std::chrono::duration_cast<std::chrono::microseconds>(
-
std::chrono::system_clock::now().time_since_epoch())
- .count();
+ int64_t now_ts = current_unix_time_us();
g_file_cache_warm_up_rowset_last_finish_unix_ts.set_value(now_ts);
- g_file_cache_warm_up_rowset_latency << (now_ts -
request_ts);
+ auto rowset_latency_us =
+
warm_up_rowset_cross_host_latency_us(request_ts, now_ts);
+ if (rowset_latency_us.has_value()) {
+ g_file_cache_warm_up_rowset_latency <<
*rowset_latency_us;
+ }
g_file_cache_warm_up_rowset_handle_to_finish_latency
<< (now_ts - handle_ts);
- if (request_ts > 0 &&
- now_ts - request_ts >
config::warm_up_rowset_slow_log_ms * 1000) {
+ if (rowset_latency_us.has_value() &&
+ *rowset_latency_us >
config::warm_up_rowset_slow_log_ms * 1000) {
g_file_cache_warm_up_rowset_slow_count << 1;
- LOG(INFO) << "warm up rowset took " << now_ts -
request_ts
+ LOG(INFO) << "warm up rowset took " <<
*rowset_latency_us
<< " us, tablet_id: " << tablet_id
<< ", rowset_id: " <<
rowset_id.to_string()
<< ", segment_id: " << segment_id;
@@ -559,18 +580,19 @@ void
CloudInternalServiceImpl::warm_up_rowset(google::protobuf::RpcController* c
if (st.ok()) {
g_file_cache_event_driven_warm_up_finished_index_num
<< 1;
g_file_cache_event_driven_warm_up_finished_index_size
<< idx_size;
- int64_t now_ts =
-
std::chrono::duration_cast<std::chrono::microseconds>(
-
std::chrono::system_clock::now().time_since_epoch())
- .count();
+ int64_t now_ts = current_unix_time_us();
g_file_cache_warm_up_rowset_last_finish_unix_ts.set_value(now_ts);
- g_file_cache_warm_up_rowset_latency << (now_ts -
request_ts);
+ auto rowset_latency_us =
+
warm_up_rowset_cross_host_latency_us(request_ts, now_ts);
+ if (rowset_latency_us.has_value()) {
+ g_file_cache_warm_up_rowset_latency <<
*rowset_latency_us;
+ }
g_file_cache_warm_up_rowset_handle_to_finish_latency
<< (now_ts - handle_ts);
- if (request_ts > 0 &&
- now_ts - request_ts >
config::warm_up_rowset_slow_log_ms * 1000) {
+ if (rowset_latency_us.has_value() &&
+ *rowset_latency_us >
config::warm_up_rowset_slow_log_ms * 1000) {
g_file_cache_warm_up_rowset_slow_count << 1;
- LOG(INFO) << "warm up rowset took " << now_ts -
request_ts
+ LOG(INFO) << "warm up rowset took " <<
*rowset_latency_us
<< " us, tablet_id: " << tablet_id
<< ", rowset_id: " <<
rowset_id.to_string()
<< ", segment_id: " << segment_id;
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
