Details
-
Bug
-
Resolution: Duplicate
-
Blocker
-
3.0
-
Security Level: Public
-
None
-
Untriaged
-
Ubuntu 64-bit
-
Yes
-
June 30 - July 18
Description
Uni-directional XDCR replication gets stuck if the source cluster hits hard out-of-memory (OOM) prior to the start of XDCR replication.
Workaround: It is recoverable by increasing the bucket RAM quota on the source cluster.
Steps to reproduce:
1)Setup uni-directional xdcr with two buckets in source and destination clusters. The buckets are named "high"(high priority bucket) and "low"(low priority bucket).
2)Both source buckets have ~600k items.
3)XDCR is stuck for both the buckets and I see the following messages in babysitter.log in one case
start seqno< end seqno and in another, end seqno is 0.
stream created with start seqno 0 and end seqno 0
memcached<0.77.0>: Tue Jul 1 11:47:57.540429 PDT 3: (high) UPR (Notifier) eq_uprq:xdcr:notifier:ns_1@172.23.105.239:high - (vb 93) stream created with start seqno 0 and end seqno 0
memcached<0.77.0>: Tue Jul 1 11:47:57.560008 PDT 3: (high) UPR (Notifier) eq_uprq:xdcr:notifier:ns_1@172.23.105.239:high - (vb 440) stream created with start seqno 0 and end seqno 0
memcached<0.77.0>: Tue Jul 1 11:56:04.169444 PDT 3: (low) UPR (Notifier) eq_uprq:xdcr:notifier:ns_1@172.23.105.239:low - (vb 88) stream created with start seqno 691 and end seqno 0