Fix unexpected max latency in output of stat
#349
Conversation
# Conflicts: # src/Service/Metrics.h
src/Service/ConnectionHandler.cpp
Outdated
| Int64 elapsed = Poco::Timestamp().epochMicroseconds() / 1000 - response->request_created_time_ms; | ||
| auto current_time = getCurrentTimeMilliseconds(); | ||
| Int64 elapsed = current_time - response->request_created_time_ms; | ||
| if (elapsed < 0) |
There was a problem hiding this comment.
Why elapsed can be -1? Caused by different clock? If it is, we must use stedy_lock in all places.
There was a problem hiding this comment.
Maybe different clock, so I use getCurrentTimeMilliseconds which is steady_clock for request in all places.
There was a problem hiding this comment.
Maybe caused by clock diff in different servers. When leader append_entries the create_time is attached to the log and when the follower takes the log the timestamp in leader is used. If there is a clock diff in the servers -1 may raise.
There was a problem hiding this comment.
Let's add a comment to help readers understand the code.
There was a problem hiding this comment.
Maybe caused by clock diff in different servers. When leader append_entries the
create_timeis attached to the log and when the follower takes the log the timestamp in leader is used. If there is a clock diff in the servers-1may raise.
When followers takes the log from leader, them won't updateLatency for the log. We used to use system_clock to measure request time cost. Clock rollback can cause this issue with system_clock.
JackyWoo
changed the title
stat
src/Service/ConnectionHandler.cpp
Outdated
| response->xid, | ||
| Coordination::toString(response->getOpNum()), | ||
| elapsed); | ||
| LOG_WARNING( |
There was a problem hiding this comment.
Just delete the logging, for it is confused for the system mantainers.
Which issues of this PR fixes:
This PR try to fix #348
Change log: