| acoles | mattoliver: clayg: I just wrote a comment on https://review.opendev.org/c/openstack/swift/+/805607 and on reflection came up with a 'principle' ( "axiom" if you like :) ) that might help us reason about the reconciler/replicator behaviour i.e. if you take a chronological sequence of events that happened across multiple nodes and play them back on a single node, then after replication the state on all nodes should eventually | 09:25 |
|---|---|---|
| acoles | become consistent with the outcome on the single node. In almost all cases the most recent event determines the final outcome, but with container policy index changes it doesn't because newer policy index changes to an existing container are ignored. | 09:25 |
| *** jokke_ is now known as jokke | 11:07 | |
| *** jokke is now known as jokke_ | 11:07 | |
| clayg | IIRC with container SPI we also have to deal with handoffs created with the wrong policy. The older db should win, but handoffs also look like they’ve never been deleted so they loose against recreated too. If you could get a handoff to accept a recreate you might be able to change an SPI! | 11:36 |
| clayg | Shards should still resolve to root tho. Even if we don’t like root behavior or change it later. | 11:37 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!