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
We currently persist the coordinate subsystem's state to disk, making it much harder to clean up after errors and incorrect values (like NaN and large adjustment values). This feature was added as an optimization to prevent a restart from causing Serf to take another few cycles before it converges.
Instead, we want to remove this persistence mechanism and reset coordinates to the origin on restart. This prevents operators from having to do complex recovery mechanisms like hand editing the local snapshot file on errors, and it should still converge fast after a few rounds of gossip.
The text was updated successfully, but these errors were encountered:
Follow up on #487 after discussion with @slackpad
We currently persist the coordinate subsystem's state to disk, making it much harder to clean up after errors and incorrect values (like NaN and large adjustment values). This feature was added as an optimization to prevent a restart from causing Serf to take another few cycles before it converges.
Instead, we want to remove this persistence mechanism and reset coordinates to the origin on restart. This prevents operators from having to do complex recovery mechanisms like hand editing the local snapshot file on errors, and it should still converge fast after a few rounds of gossip.
The text was updated successfully, but these errors were encountered: