Systemd programming, 30 months later

Posted Oct 11, 2016 9:30 UTC (Tue) by neilbrown (subscriber, #359)
In reply to: Systemd programming, 30 months later by Darkstar
Parent article: Systemd programming, 30 months later

> I think "mount -o soft" should still work

"-o soft" never worked for any useful definition of "worked" - i.e. one where you could trust that you data was safe. I once heard NFS described as Nulls Frequently Substituted. If you use -o soft and have bad latency on your network, you can get holes in files.

autofs is by far the best solution to handle disconnects well.

Systemd programming, 30 months later

Posted Oct 11, 2016 16:12 UTC (Tue) by Darkstar (guest, #28767) [Link] (1 responses)

> If you use -o soft and have bad latency on your network, you can get holes in files.
I think this only applies to "-o soft,udp" but not "-o soft,tcp". But then again it's been years and you might be correct. I agree that autofs is probably the better option

Systemd programming, 30 months later

Posted Oct 11, 2016 21:45 UTC (Tue) by neilbrown (subscriber, #359) [Link]

soft,tcp would certainly be different than soft,udp but the same risks are there - just different probabilities and different patterns.. I suspect it would be harder to demonstrate a problem with tcp, but not impossible.