Open objmagic opened 8 years ago
We should investigate whether this issue still exists after the protobuf performance is improved.
cf. #1217
@objmagic @taishi8117 - any update on this?
not really sure how to start investigating this issue
@kramasamy this should be the most important and the last issue. And I will investigate this issue.
I am getting a bit lost when investigating this issue. Event loop could stop running, but not all the time. When I decrease parallelism to 1, acking latency (complete latency) is much slower, but still growing.
In PyHeron, acking latency increases gradually until it hits timeout. After that, all acking tuples will be dropped.