Hi all, I recently started seeing zfs chattiness at boot time: "reading zfs config" and something like "mounting zfs filesystems (n/n)". Is this really necessary? I thought with SMF the times where every script announced its'' existance had gone (and good thing, too). Can''t we print something only if it goes wrong? Michael -- Michael Schuster Recursion, n.: see ''Recursion''
On 9/24/07, Michael Schuster <Michael.Schuster at sun.com> wrote:> I recently started seeing zfs chattiness at boot time: "reading zfs config" > and something like "mounting zfs filesystems (n/n)". > > Is this really necessary? I thought with SMF the times where every script > announced its'' existance had gone (and good thing, too). > > Can''t we print something only if it goes wrong?Unfortunately, sometimes something goes wrong and nothing realizes. I have always advocated being able the tune the chattiness of booting. If I had a system (under Solaris <10) that was hanging on boot (or more frequently on shutdown) I would add a bunch of echos to the scripts /sbin/rc0, /sbin/rc2, etc. Then I could clearly see what was hanging. I would love SMF to support a boot (and shutdown) option to tell me what services it is trying to start/stop/restart to the console so I can find the one that misbehaves (and doesn''t know it). -- Paul Kraus
See http://www.opensolaris.org/jive/thread.jspa?messageID=122606 -- richard Michael Schuster wrote:> Hi all, > > I recently started seeing zfs chattiness at boot time: "reading zfs config" > and something like "mounting zfs filesystems (n/n)". > > Is this really necessary? I thought with SMF the times where every script > announced its'' existance had gone (and good thing, too). > > Can''t we print something only if it goes wrong? > > Michael
On Mon, 24 Sep 2007, Michael Schuster wrote:> I recently started seeing zfs chattiness at boot time: "reading zfs config" > and something like "mounting zfs filesystems (n/n)".This was added recently because ZFS can take a while to mount large configs. Consoles would appear to freeze after the initial boot-up messages. 10k filesystems could easily take several minutes to mount. Eric Taylor is working on parallel mounting for ZFS which will speed up things considerably, although his changes currently do not remove the messages. Perhaps the reading/mounting messages should be only displayed if, say, a minute has passed and we''re not done? I agree things should not be needlessly chatty, but I also believe that processes which run a long time (especially when affecting boot time) should provide feedback to users to let them know the box isn''t dead. Regards, markm
Mark J Musante wrote:> On Mon, 24 Sep 2007, Michael Schuster wrote: > >> I recently started seeing zfs chattiness at boot time: "reading zfs config" >> and something like "mounting zfs filesystems (n/n)". > > This was added recently because ZFS can take a while to mount large > configs. Consoles would appear to freeze after the initial boot-up > messages. 10k filesystems could easily take several minutes to mount. > > Eric Taylor is working on parallel mounting for ZFS which will speed up > things considerably, although his changes currently do not remove the > messages. Perhaps the reading/mounting messages should be only displayed > if, say, a minute has passed and we''re not done?That was my suspicion, and I''d vote for a change along the outline you suggest here. A minute may be a bit much, though (YMMV ;-). I''m also quite prepared to see a running tally(?) after an initial timeout (your minute) has gone by and we haven''t finished ... but I guess we''d also have to make sure that the output generated isn''t messed up by other output to the console that''s independant of ZFS. I''d completely do away with the first message ("reading zfs config") for the OK case.> I agree things should not be needlessly chatty, but I also believe that > processes which run a long time (especially when affecting boot time) > should provide feedback to users to let them know the box isn''t dead.indeed - see above. thx Michael -- Michael Schuster Recursion, n.: see ''Recursion''
"Paul Kraus" writes:> On 9/24/07, Michael Schuster <Michael.Schuster at sun.com> wrote: > > > I recently started seeing zfs chattiness at boot time: "reading zfs config" > > and something like "mounting zfs filesystems (n/n)". > > > > Is this really necessary? I thought with SMF the times where every script > > announced its'' existance had gone (and good thing, too). > > > > Can''t we print something only if it goes wrong?It would be sad to return to a world where useful messages about things going wrong were overwhelmed by subsystems simply announcing their existance to the world. I''ve been on the fence about whether this case crosses that threshold.> Unfortunately, sometimes something goes wrong and nothing > realizes. I have always advocated being able the tune the chattiness > of booting. If I had a system (under Solaris <10) that was hanging on > boot (or more frequently on shutdown) I would add a bunch of echos to > the scripts /sbin/rc0, /sbin/rc2, etc. Then I could clearly see what > was hanging. > > I would love SMF to support a boot (and shutdown) option to > tell me what services it is trying to start/stop/restart to the > console so I can find the one that misbehaves (and doesn''t know it).boot -m verbose does this for startup. (Note that both error messaging and startup messaging were improved significantly in u3/u4/NV. There were some bugs before that which made the experience less friendly.) Bug 6195169 asserts it should work for shutdown too. (Which I agree with.) There are also some other "chatty" modes that I think would be useful, but I''ll leave that for a conversation on smf-discuss. liane
On Mon, Sep 24, 2007 at 12:33:00PM -0400, Mark J Musante wrote:> On Mon, 24 Sep 2007, Michael Schuster wrote: > > I recently started seeing zfs chattiness at boot time: "reading zfs config" > > and something like "mounting zfs filesystems (n/n)". > > This was added recently because ZFS can take a while to mount large > configs. Consoles would appear to freeze after the initial boot-up > messages. 10k filesystems could easily take several minutes to mount.I hope there''s some RFE (or bug) CR for implementing a more on-demand mounting strategy (like automounting, effectively). Nico --
On Mon, 24 Sep 2007, Michael Schuster wrote:> I''m also quite prepared to see a running tally(?) after an initial timeout > (your minute) has gone by and we haven''t finished ... but I guess we''d also > have to make sure that the output generated isn''t messed up by other output > to the console that''s independant of ZFS.OK, I''ve written up CR 6608465 to track this issue. Regards, markm