Re:Killing processes that don't want to die
by emilfihlman on Tuesday September 25, 2018 @05:41AM
(#18064795)
It's absolutely infuriating that you can, with a simple NFS, cause an unkillable, unremovable, always there process to exist. There is simply no reason to require this.
[ Parent
| on HN ]
Re:Killing processes that don't want to die
by giovannibajo1 on Tuesday September 25, 2018 @06:21AM
(#18064999)
For non vital filesystems, I use "-o intr,soft,timeo=5" at NFS mount time. This causes syscalls to return EINTR and timeout without hard locking. It can sometimes confuse userspace, but at least you don't get a stuck process.
Re:Killing processes that don't want to die
by pjc50 on Tuesday September 25, 2018 @06:38AM
(#18065100)
I always preferred "intr,hard", because having filesystem operations fail really stresses out a lot of userspace applications. Most developers haven't thoroughly tested their handling of "what if this disk operation fails?", because it's a pain to do and a rare case.