You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As part of #4053 we're switching the semantics of the LB throttle for remote bootstraps to be global, in order to not start too many concurrent operations in the cluster. We should however add back a mechanism to support a per-table control.
This could allow us to make more healthy progress, in the case of very large tables otherwise starving the balancing process.
The text was updated successfully, but these errors were encountered:
Summary:
Before #4053, each table was allowed to have up the specified limit, although this was not the intention of the flag.
Now that it has been fixed, it's posible that one or the first few tables that get analyzed by the load balancer consume all the remote bootstrap sessions allowed by the flag. So this change limits the number of RBS sessions per table to give all the tables a chance to make progress.
Test Plan: ybd release --cxx-test integration-tests_remote_bootstrap-itest --gtest_filter RemoteBootstrapITest.TestLimitNumberOfConcurrentRemoteBootstraps -n 100
Reviewers: bogdan
Reviewed By: bogdan
Subscribers: ybase
Differential Revision: https://phabricator.dev.yugabyte.com/D8455
As part of #4053 we're switching the semantics of the LB throttle for remote bootstraps to be global, in order to not start too many concurrent operations in the cluster. We should however add back a mechanism to support a per-table control.
This could allow us to make more healthy progress, in the case of very large tables otherwise starving the balancing process.
The text was updated successfully, but these errors were encountered: