wu-sheng commented on a change in pull request #5407:
URL: https://github.com/apache/skywalking/pull/5407#discussion_r478869479
##########
File path: docs/en/setup/backend/trace-sampling.md
##########
@@ -30,4 +30,8 @@ When you set the rate different, let's say
And we assume the agents reported all trace segments to backend,
Then the 35% traces in the global will be collected and saved in storage
consistent/complete, with all spans.
20% trace segments, which reported to Backend-Instance**B**, will saved in
storage, maybe miss some trace segments,
-because they are reported to Backend-Instance**A** and ignored.
\ No newline at end of file
+because they are reported to Backend-Instance**A** and ignored.
+
+# Note
+When you open sampling, the actual sampleRate will above sampleRate. Because
we want some error segment will be saved, even that segment will abandoned by
server side trace sampling mechanism. may be miss some other trace segments,
but we can analyze this error segment to solve problem.
Review comment:
Try to provide a configuration to disable this feature, this could be
default open as the harm is limited.
##########
File path: docs/en/setup/backend/trace-sampling.md
##########
@@ -30,4 +30,8 @@ When you set the rate different, let's say
And we assume the agents reported all trace segments to backend,
Then the 35% traces in the global will be collected and saved in storage
consistent/complete, with all spans.
20% trace segments, which reported to Backend-Instance**B**, will saved in
storage, maybe miss some trace segments,
-because they are reported to Backend-Instance**A** and ignored.
\ No newline at end of file
+because they are reported to Backend-Instance**A** and ignored.
+
+# Note
+When you open sampling, the actual sampleRate will above sampleRate. Because
we want some error segment will be saved, even that segment will abandoned by
server side trace sampling mechanism. may be miss some other trace segments,
but we can analyze this error segment to solve problem.
Review comment:
```suggestion
When you open sampling, the actual sample rate could be over sampleRate.
Because currently, all error segments will be saved, meanwhile, the upstream
and downstream may not be sampled. This feature is going to make sure you could
have the error stacks and segments, but don't guarantee you would have the
whole trace.
Also, the side effect would be, if most of the accesses are fail, the
sampling rate would be closing to 100%, which could crash the backend or
storage clusters.
```
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
