Major Description
In MB-56547 - KV has changed it such that OSO is not guaranteed to occur for a single collection.
In XDCR, this breaks the oso tests as we now see the following messages in source cluster's memcached log:
2023-09-19T13:54:07.341123-07:00 INFO 69: (B1) DCP (Producer) eq_dcpq:xdcr:dcp_backfill_ff74ddcd03879f28cb8c4bc57a662345/B1/B2_127.0.0.1:12000_0:7dSoVX_2oV-8cf6euYQbOw== - (vb:31) Skipping OSO backfill for cid:0xd as collection item count (12) is too large a percentage of the vBucket item count (36) - (33.33%)
|
2023-09-19T13:54:07.341187-07:00 INFO 69: (B1) DCP (Producer) eq_dcpq:xdcr:dcp_backfill_ff74ddcd03879f28cb8c4bc57a662345/B1/B2_127.0.0.1:12000_0:7dSoVX_2oV-8cf6euYQbOw== - (vb:31) Scheduling backfill from 1 to 45, uid:2049, reschedule flag : False
|
2023-09-19T13:54:07.341246-07:00 INFO 69: (B1) DCP (Producer) eq_dcpq:xdcr:dcp_backfill_ff74ddcd03879f28cb8c4bc57a662345/B1/B2_127.0.0.1:12000_0:7dSoVX_2oV-8cf6euYQbOw== - (vb:397) Creating stream with start seqno 0 and end seqno 30; requested end seqno was 30, flags:80, snapshot:{0,0} collections-filter-size:1 sid:none
|
2023-09-19T13:54:07.341258-07:00 INFO 69: (B1) DCP (Producer) eq_dcpq:xdcr:dcp_backfill_ff74ddcd03879f28cb8c4bc57a662345/B1/B2_127.0.0.1:12000_0:7dSoVX_2oV-8cf6euYQbOw== - (vb:397) ActiveStream::scheduleBackfill_UNLOCKED register cursor with name "eq_dcpq:xdcr:dcp_backfill_ff74ddcd03879f28cb8c4bc57a662345/B1/B2_127.0.0.1:12000_0:7dSoVX_2oV-8cf6euYQbOw==" wantedSeqno:0, result{tryBackfill:true, seqno:31}, op:empty, lastReadSeqno:0
|
MB-56601 showcases a diag/eval command to force the old behavior but that doesn't seem to work. The right thing to do for XDCR test is to load the bucket large enough such that the above condition is satisfied.
Attachments
Issue Links
- is caused by
-
MB-56547 Selectively enable out-of-seqno backfill only when expected to be faster than seqno
-
- Closed
-