On Mon, 13 Mar 2017, J. Bruce Fields wrote:
> On Fri, Mar 10, 2017 at 04:39:19PM -0500, Scott Mayhew wrote:
> > sm-notify's early ending of the grace period when it has no hosts to
> > notify can cause problems in some high availability configurations,
> > which may be running one sm-notify per floating IP address in the
> > cluster.
>
> So, there may be no hosts associated with this floating IP, but there
> may be with another?
Exactly. What was happening in my testing was that clients would
receive and respond to a NOTIFY, and would immediately send a LOCK with
reclaim=yes but would get NFSERR_NO_GRACE in return.
BTW I had a cluster with this change running at Connectathon.
-Scott
>
> > This commit makes that behavior configurable via the
> > nfs.conf (I don't think having a corresponding command line option
> > would be particularly useful, hence none was added).
>
> And, I see that you're defaulting it to on to match existing behavior.
>
> Makes sense to me, ACK.--b.
>
> >
> > Signed-off-by: Scott Mayhew <smayhew@xxxxxxxxxx>
> > ---
> > nfs.conf | 1 +
> > utils/statd/sm-notify.c | 6 +++++-
> > utils/statd/sm-notify.man | 18 ++++++++++++++++++
> > 3 files changed, 24 insertions(+), 1 deletion(-)
> >
> > diff --git a/nfs.conf b/nfs.conf
> > index 81ece06..690645c 100644
> > --- a/nfs.conf
> > +++ b/nfs.conf
> > @@ -65,6 +65,7 @@
> > # retry-time=900
> > # outgoing-port=
> > # outgoing-addr=
> > +# lift-grace=y
> > #
> > #[svcgssd]
> > # principal=
> > diff --git a/utils/statd/sm-notify.c b/utils/statd/sm-notify.c
> > index 623213e..0c6766f 100644
> > --- a/utils/statd/sm-notify.c
> > +++ b/utils/statd/sm-notify.c
> > @@ -45,6 +45,8 @@
> >
> > #define NLM_END_GRACE_FILE "/proc/fs/lockd/nlm_end_grace"
> >
> > +int lift_grace = 1;
> > +
> > struct nsm_host {
> > struct nsm_host * next;
> > char * name;
> > @@ -494,6 +496,7 @@ main(int argc, char **argv)
> > opt_max_retry = conf_get_num("sm-notify", "retry-time", opt_max_retry / 60) * 60;
> > opt_srcport = conf_get_str("sm-notify", "outgoing-port");
> > opt_srcaddr = conf_get_str("sm-notify", "outgoing-addr");
> > + lift_grace = conf_get_bool("sm-notify", "lift-grace", lift_grace);
> > s = conf_get_str("statd", "state-directory-path");
> > if (s && !nsm_setup_pathnames(argv[0], s))
> > exit(1);
> > @@ -570,7 +573,8 @@ usage: fprintf(stderr,
> > (void)nsm_retire_monitored_hosts();
> > if (nsm_load_notify_list(smn_get_host) == 0) {
> > xlog(D_GENERAL, "No hosts to notify; exiting");
> > - nsm_lift_grace_period();
> > + if (lift_grace)
> > + nsm_lift_grace_period();
> > return 0;
> > }
> >
> > diff --git a/utils/statd/sm-notify.man b/utils/statd/sm-notify.man
> > index bb7f6e0..cfe1e4b 100644
> > --- a/utils/statd/sm-notify.man
> > +++ b/utils/statd/sm-notify.man
> > @@ -241,6 +241,24 @@ These have the same effect as the command line options
> > .B v
> > respectively.
> >
> > +An additional value recognized in the
> > +.B [sm-notify]
> > +section is
> > +.BR lift-grace .
> > +By default,
> > +.B sm-notify
> > +will lift lockd's grace period early if it has no hosts to notify.
> > +Some high availability configurations will run one
> > +.B sm-notify
> > +per floating IP address. In these configurations, lifting the
> > +grace period early may prevent clients from reclaiming locks.
> > +.RB "Setting " lift-grace " to " n
> > +will prevent
> > +.B sm-notify
> > +from ending the grace period early.
> > +.B lift-grace
> > +has no corresponding command line option.
> > +
> > The value recognized in the
> > .B [statd]
> > section is
> > --
> > 2.7.4
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> > the body of a message to majordomo@xxxxxxxxxxxxxxx
> > More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
