Features:

  • Meta node starts follower service if it is a follower
  • Meta node starts leader services if it is a leader
  • Meta node initialises leader services only after it becomes a leader
  • Failover from follower to leader if follower dies
  • unittests check if follower becomes leader on failover

Limitations:

  • No failover from leader to follower (see fencing)
  • Meta clients are not aware that there may be an HA
  • No fencing
  • No service discovery

Related:

  • https://github.com/risingwavelabs/risingwave/issues/6534
0
stream-processing sql serverless postgresql rust distributed-database cloud-native database

When we initialise the hummock manager only after the node became a leader, we will still get an error, however a slightly different one: (see logs/stack trace) CC @yezizp2012

This is still related to stored results generated by the selection in the meta store. Please ENSURE that the leader information in MetaSrvEnv is correctly updated when follower get elected. I guess this is the main cause. 😃

1

This is still related to stored results generated by the selection in the meta store. Please ENSURE that the leader information in MetaSrvEnv is correctly updated when follower get elected. I guess this is the main cause. 😃

Done. See https://github.com/risingwavelabs/risingwave/pull/6937/commits/30b8e4a855d60f257d612c5934255cd6ecf04b5d

0
© 2022 pullanswer.com - All rights reserved.
Privacy policy