nickva commented on pull request #2952: URL: https://github.com/apache/couchdb/pull/2952#issuecomment-650584859
> The first thing I realized is that by using the centralized metadataVersion, I think we're basically causing all in-flight write transactions to retry We are doing md version reads at the snapshot level isolation https://github.com/apache/couchdb/blob/prototype/fdb-layer/src/fabric/src/fabric2_fdb.erl#L1273 so don't think they should retry. But it is worth checking if it is not too difficult. > we should be going back all the way to the directory layer to check that our directory hasn't been renamed and what not. That means each time the metadataVersion key we're now adding an extra ~5-10 KV lookups in those cases. That is legit it. Good point. I don't think we technically have any place where we programmatically bump the md version when directory layer updates (we'd do it manually), but still for correctness and completeness we should revalidate it. We already do it for full re-opens (when db version and md versions are stale: https://github.com/apache/couchdb/blob/prototype/fdb-layer/src/fabric/src/fabric2_fdb.erl#L134) so we should do it on md changes as well. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
