How are the shard leaders distributed across nodes in YugabyteDB?

General
1

  1. And is there some metrics exported or way to verify what the leader counts is in each node overall as well as per table?

  2. For multi-region installations, Google Cloud Spanner (Informazioni sulle istanze  |  Cloud Spanner  |  Google Cloud) has a notion of a “default leader” for performance reasons, and recommends < Place the bulk of your read and write workloads in the default leader region. >. Does YugabyteDB have a similar notion and if so, how is this specified?

2

hi @busybruner

  1. Tablet (shard) leaders in YugabyteDB are distributed as evenly as possible across yb-tserver nodes. This helps spread read/write workloads because tablet leaders are consulted for writes and reads.

  2. Currently we expose tablet leader/followers on the yb-master UI. We can see the overall tablet counts on http://<any-yb-master-ip>:7000/tablet-servers.

tablet_servers
tablet_servers1855×962 90.2 KB

While tablet specific leaders/followers for each table are displayed on the ui in http://<any-tserver-ip>:9000/tablets url of each yb-tserver.

tablets_page
tablets_page1855×962 265 KB

  1. Yes, YugabyteDB supports the ability to specify preferred zone/regions(s) for the leaders. This can be done with set_preferred_zones 1 command in yb-admin cli. As long nodes are alive and healthy all tablet leaders will reside in that region/zone.
1 Like
4

@busybruner please see edited reply above