Hey Ludo,
> Now, the only way ‘narinfo-string’ can take too long these days is (I > think) if the store GC lock is held (we should check that hypothesis, > but I believe that if the GC lock is held, then the ‘query-path-info’ > RPC made from ‘narinfo-string’ might block until the lock is released). > The GC lock is no longer held for hours on berlin, so there’s less > pressure to address that. Thanks for having a look to that issue. On my side, I'd like to take some time to look closely at the situation. > To summarize: I think ‘guix publish’ is okay as-is but we should > fiberize it sometime. I think however that it would be nice to create a wip branch for those changes, deploy it and monitor it closely on Berlin before pushing. From experience, the publish code is really fragile and the I/O pressure on Berlin can exhibit a lot of unexpected issues. Mathieu