Add Jobs and tests for common maintenance operations #95
Conversation
of under-replicated topics if a broker is down
as long as you have one up-to-date replica. The use case is broker outages that are longer than mere re-scheduling, such as zone outage in a multi-zone cluster.
solsson
force-pushed
the
ops-jobs
branch
2 times, most recently
from
914029e
to
59fb4b9
Compare
|
This is tagged automation but so far it only aids manual maintenance, so I suggest merge to v.3.1. My work with this PR reinforced my position that production topics should have three replicas but producers should require no more than two acks (assuming a three-broker cluster over three availability zones). |
|
Got a pretty good test case in #98. Maybe it would be a sensible default for the reassign job to list all topics and all Ready brokers, if we can somehow prompt for human confirmation after echoing the proposal. |
which seems to be equivalent to setting replication.factor 3 at topic create.
|
There's other tooling for stuff like increasing replication factor, and bash scripts can be more elegant than mine, but I found 49649e1 immensely useful and it's quite by the (definitive guide) book. Please be careful with these scripts in production though. They're not tested to the degree that automation should be. |
No description provided.