There is a bug in the way StackerDB reconfiguration behaves if the number of slots decreases. The replica won't delete the now-obsolete slots. This in turn leads to problems synchronizing with new nodes that didn't ever materialize those chunks in the first place -- each will declare the other as having sent invalid StackerDBChunkInv messages, since the number of slots will be different.
The fix is to delete unused chunks on the call to StackerDBTx::reconfigure_stackerdb()
jcnelson
commented
2 weeks ago
There is a bug in the way StackerDB reconfiguration behaves if the number of slots decreases. The replica won't delete the now-obsolete slots. This in turn leads to problems synchronizing with new nodes that didn't ever materialize those chunks in the first place -- each will declare the other as having sent invalid
StackerDBChunkInvmessages, since the number of slots will be different.The fix is to delete unused chunks on the call to
StackerDBTx::reconfigure_stackerdb()