Details
-
Bug
-
Resolution: Fixed
-
Critical
-
3.1.6, 4.6.0
-
None
-
Untriaged
-
-
No
-
KV Spock Beta
Description
The issue was seen with XDCR + graceful failover.
Issue seen on node 172.23.98.224. All missing items are one vbucket, vb 998.
2017-05-19T17:15:36.209243-07:00 NOTICE (Test1) DCP (Producer) eq_dcpq:replication:ns_1@172.23.98.224->ns_1@172.23.98.225:Test1 - (vb 998) Creating takeover stream with start seqno 848 and end seqno 18446744073709551615
|
2017-05-19T17:15:36.690043-07:00 NOTICE (Test1) DCP (Producer) eq_dcpq:replication:ns_1@172.23.98.224->ns_1@172.23.98.225:Test1 - (vb 998) Vbucket marked as dead, last sent seqno: 848, high seqno: 865
|
2017-05-19T17:15:36.691534-07:00 NOTICE (Test1) DCP (Producer) eq_dcpq:replication:ns_1@172.23.98.224->ns_1@172.23.98.225:Test1 - (vb 998) Stream closing, sent until seqno 848 remaining items 0, reason: The stream ended due to all items being streamed
|
2017-05-19T17:15:36.692372-07:00 NOTICE (Test1) DCP (Producer) eq_dcpq:replication:ns_1@172.23.98.224->ns_1@172.23.98.225:Test1 - (vb 998) Stream closed, 0 items sent from backfill phase, 0 items sent from memory phase, 848 was last seqno sent
|
We are closing the takeover stream before all items are streamed (865-848). This is causing the data loss.
Attachments
For Gerrit Dashboard: MB-24817 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
79426,2 | MB-24817: Upon DCP stream creation, log end_seqno more accurately | master | kv_engine | Status: MERGED | +2 | +1 |
79500,3 | MB-24817, WIP: Check thread sanitizer | watson | ep-engine | Status: ABANDONED | 0 | -1 |
79532,4 | MB-24817: During takeover, hold stream lock until vb is set to dead | watson | ep-engine | Status: MERGED | +2 | +1 |
79681,1 | MB-24817: During takeover, hold stream lock until vb is set to dead | master | kv_engine | Status: ABANDONED | -2 | +1 |
79783,3 | Merge remote-tracking branch 'couchbase/watson_ep' | master | kv_engine | Status: MERGED | +2 | +1 |