Hi Bo-Chun Wang,
I've been having a little look into why this might be happening because there haven't been many (if any) major code changes to cbexport between those two builds; the only thing I can think of is that by default we will now request
that Couchbase Server sequentially backfills the vBuckets. I have my doubts whether this may be the cause of the issue because if we look at the latest backup to blackhole test we can see that the throughput has increased (backup/export share the same code in this respect).
I did some local testing of cbexport (which should be taken with a grain of salt due to the different data sets/machines) which contradict the results that we can see from Leto; most importantly being that in all cases (e.g. with/without threads) 6.6.0 proves to export data faster than 6.5.0 does.
The difference in throughput for cbexport may simply be down to the performance issues that we've been seeing on the Leto. There are some daily tests which are run on triton which includes a backup. We can see the throughput for this backup is significantly higher than on Leto. Further investigation shows that the throughput hasn't degraded on the triton cluster since the daily performance tests began for 7.0.0. It might be interesting if we could run a test comparing 6.5.0/6.6.0 cbexport on an alternative cluster so that we can determine if there is actually an issue in cbexport.
Thanks in advance,