Hi everyone,

I'm writing this request to propose a merge of HBASE-26913[1] "Replication
Observability Framework" to master and branch-2.
The goal is to persist replication related metadata to newly created hbase
tables to reduce MTTD (Mean Time To Detect) and
MTTR (Mean Time To Repair) replication inconsistencies between Primary and
DR clusters.
The design doc is here[2].

We have created 2 new hbase tables, REPLICATION.WALEVENTTRACKER and
REPLICATION.SINK_TRACKER. The first table
will store all the WAL events (ACTIVE, ROLLING, ROLLED) along with metadata
(wal name, wal length, region server name, timestamp)
from all region servers. We have introduced a new chore called
ReplicationMarkerChore which will create special marker/sentinel rows
periodically (which is configurable) and inject them directly to WAL. This
marker rows will be handled specially and replicated to
sink cluster and will be persisted to REPLICATION.SINK_TRACKER table.

Highlights
* The entire feature is configurable. Defaults to false.
* A new section is added to the HBase book which covers the feature and how
to use it.

The vote will open for at least 72 hours.
Please vote:
[+1]    Merge the changes from HBASE-26913 to master/branch-2
[+/-0]  Neutral
[-1]    Disagree (please include actionable feedback)

1. https://issues.apache.org/jira/browse/HBASE-26913
2.
https://docs.google.com/document/d/14oZ5ssY28hvJaQD_Jg9kWX7LfUKUyyU2PCA93PPzVko/edit#heading=h.9oum2kn0zj5r


Thanks,
Rushabh

Reply via email to