This message was deleted.
# troubleshootings
Slackbot
03/08/2023, 10:24 PMThis message was deleted.
b
Ben Krug
03/08/2023, 10:55 PMI'm not sure how you would make sure they don't write to the same segment-cache, or step on each other's feet on the drive.
Ben Krug
03/08/2023, 10:56 PMMaybe if each is started with a different runtime.properties file being passed to the startup command.
Ben Krug
03/08/2023, 10:56 PMWhy not just run one bigger historical (ie, give it the resources you intend to give both)?
n
Nir Bar On
03/08/2023, 11:02 PMquestion - when I set more than single historical .. they act as a replica meaning each node hold a clone of segments ..
or they act as sharding - each node hold a subset of segments in cache .. ?
d
Didip Kerabat
03/09/2023, 12:17 AMReplication is managed by the coordinator, you define those rules using the UI/API.
Historical is just a “dumb” data plane and they are all equivalent to each other.
n
Nir Bar On
03/09/2023, 12:29 AMso if I understand correctly if I have 100 segments and I spin up 2 historical , which historical will load 100 segments ?
d
Didip Kerabat
03/09/2023, 12:30 AMCoordinator will decide on its own. it has its own algorithm: cachingCost, cost, etc.
The final result, where things are, will be viewable inside Zookeeper.
n
Nir Bar On
03/09/2023, 12:31 AMwhat is the benfit to use few historical node vs single ?
Nir Bar On
03/09/2023, 12:32 AMif all historical hold the same segments population ..?
Nir Bar On
03/09/2023, 12:37 AMwhat I am not sure I undersntad is if cluster has N count of segments .. is those N segments split (distributed) on historical node (option A).. OR is each historical node act as a copy of the other nodes (replica) - Option B ..
design wise how druid works .. option A / or option B ?
d
Didip Kerabat
03/09/2023, 12:51 AMthe segments are both sharded and replicated
Didip Kerabat
03/09/2023, 12:52 AMso its both
b
Ben Krug
03/09/2023, 12:54 AMYou set the replication factor, ie, how many copies of each segment you wanted loaded. 2 is common. That way if one is unavailable on one historical, it's on another one, too.
Ben Krug
03/09/2023, 12:56 AMsingle node gives no redundancy. Multiple nodes also spreads the load. Big data nodes are fine, but there's also benefits to sharding, for the usual reasons.