Output reloading: ensure all events get eventually published #17657
Conversation
| fmt.Println("reloading output...") | ||
| pipeline.output.Set(out) | ||
|
|
||
| //time.Sleep(4 * time.Millisecond) |
There was a problem hiding this comment.
Interestingly, if some time is allowed between output reloads the chances of all events eventually being published are MUCH higher. Obviously this is not a solution, just an observation that might give a clue as to what's happening.
|
Based on debugging statements I just added, it appears that sometimes the memory queue consumer is returning events but the output worker is somehow not getting them for publishing. This is causing the test to fail (eventually all events are not being published). Investigating deeper... |
andresrc
added
[zube]: Inbox
Team:Services
|
Pinging @elastic/integrations-services (Team:Services) |
ycombinator
force-pushed
the
lb-output-reloading-unit-test
branch
from
0601071
to
b15a1c2
Compare
…TRYER BLOCKING CONSUMER!
ycombinator
force-pushed
the
lb-output-reloading-unit-test
branch
2 times, most recently
from
b124ca3
to
dabff5e
Compare
ycombinator
force-pushed
the
lb-output-reloading-unit-test
branch
from
14057ae
to
2d82abc
Compare
|
Discussed the test and experimental code changes made in this PR with @urso off-PR. We decided to "move" the test implemented in this PR into #17381, as the latter PR contains some code changes that might help make this test pass. Some of the experimental code changes made in this PR here might also need to be moved over to help make the test pass. So I've closed this PR unmerged and will continue further development in #17381. |
NOTE: This PR is VERY MUCH a WIP. It is not intended to be reviewed at this point. At this point, it is primarily being used for debugging the event publishing flow with output reloading.
What does this PR do?
This PR adds a unit test for output reloading. The test reloads the output multiple times, very quickly, and checks if all events are eventually published.
Why is it important?
If the output is reloaded multiple times very quickly, not all events seem to get published eventually. This PR adds a test that proves it so we can then work towards a fix.