Mailing List Archive

meta-data handling implemented
Hi,

Time for a little DRBD-party !
The latest drbd.c has the cvs revision number "1.100".
----------------------------
revision 1.100
date: 2001/05/15 14:50:50; author: kde; state: Exp; lines: +183 -27
Implemented meta-data handling. Completely untested/undocumented.
DRBD now maintains small files in /var/lib/drbd/ which are used to
store state information.
After a common power failure these information can be used to find
the node which has the good bits.
----------------------------

In the next days I will test (and debug) it, write documentation,
update the web-page etc...

But there is still one important question left. How do we hand over
this information to heartbeat ?

-Philipp
Re: meta-data handling implemented [ In reply to ]
On 2001-05-15T16:58:54,
Philipp Reisner <philipp.reisner@example.com> said:

> In the next days I will test (and debug) it, write documentation,
> update the web-page etc...
>
> But there is still one important question left. How do we hand over
> this information to heartbeat ?

If you could provide a parameter which returned the name of the host with the
good bits, at least FailSafe could use this information to dynamically
construct the current available failover hosts.

(This doesn't answer the question, but it would be good to have ;-)

Sincerely,
Lars Marowsky-Brée <lmb@example.com>

--
Perfection is our goal, excellence will be tolerated. -- J. Yahl
Re: meta-data handling implemented [ In reply to ]
Philipp Reisner wrote:
>
> Hi,
>
> Time for a little DRBD-party !
> The latest drbd.c has the cvs revision number "1.100".
> ----------------------------
> revision 1.100
> date: 2001/05/15 14:50:50; author: kde; state: Exp; lines: +183 -27
> Implemented meta-data handling. Completely untested/undocumented.
> DRBD now maintains small files in /var/lib/drbd/ which are used to
> store state information.
> After a common power failure these information can be used to find
> the node which has the good bits.
> ----------------------------
>
> In the next days I will test (and debug) it, write documentation,
> update the web-page etc...
>
> But there is still one important question left. How do we hand over
> this information to heartbeat ?

This is great!

Hmmm... I'll have to go reread what I had suggested before and see if I can
figure this out. ;-)

What kind of information are you keeping?

Are you keeping the specific kind of info I suggested (a tuple of two
numbers)?

What kind of an interface is your code looking for?

In other words:

What info do you have?

What "executive oversight" do you want heartbeat to perform?

This is really great news, though!!!

Congratulations!


-- Alan Robertson
alanr@example.com
Re: Re: meta-data handling implemented [ In reply to ]
Hi Alan,

I modified your initial ideas (the tuple) a bit...
...you can read about it in the paper I wrote:

http://www.complang.tuwien.ac.at/~reisner/drbd/download/drbd_paper_for_NLUUG_2001.pdf
http://www.complang.tuwien.ac.at/~reisner/drbd/download/drbd_paper_for_NLUUG_2001.ps.gz

I am just in the progress of thinking of a solution which does not need
any support from the cluster manager at all.
(The /etc/init.d/drbd script which is started before the
/etc/init.d/heartbeat script will block until heartbeat can
do what it does ...)

I will post a description tomorrow.

-Philipp

* Alan Robertson <alanr@example.com> [010515 21:49]:
> Philipp Reisner wrote:
> >
> > Hi,
> >
> > Time for a little DRBD-party !
> > The latest drbd.c has the cvs revision number "1.100".
> > ----------------------------
> > revision 1.100
> > date: 2001/05/15 14:50:50; author: kde; state: Exp; lines: +183 -27
> > Implemented meta-data handling. Completely untested/undocumented.
> > DRBD now maintains small files in /var/lib/drbd/ which are used to
> > store state information.
> > After a common power failure these information can be used to find
> > the node which has the good bits.
> > ----------------------------
> >
> > In the next days I will test (and debug) it, write documentation,
> > update the web-page etc...
> >
> > But there is still one important question left. How do we hand over
> > this information to heartbeat ?
>
> This is great!
>
> Hmmm... I'll have to go reread what I had suggested before and see if I can
> figure this out. ;-)
>
> What kind of information are you keeping?
>
> Are you keeping the specific kind of info I suggested (a tuple of two
> numbers)?
>
> What kind of an interface is your code looking for?
>
> In other words:
>
> What info do you have?
>
> What "executive oversight" do you want heartbeat to perform?
>
> This is really great news, though!!!
>
> Congratulations!
>
>
> -- Alan Robertson
> alanr@example.com
>
> _______________________________________________
> DRBD-devel mailing list
> DRBD-devel@example.com
> http://lists.sourceforge.net/lists/listinfo/drbd-devel
Re: Re: meta-data handling implemented [ In reply to ]
On 2001-05-16T12:09:53,
Philipp Reisner <philipp.reisner@example.com> said:

> I am just in the progress of thinking of a solution which does not need
> any support from the cluster manager at all.
> (The /etc/init.d/drbd script which is started before the
> /etc/init.d/heartbeat script will block until heartbeat can
> do what it does ...)

I disagree with blocking until the cluster manager can be started. Other
resources might be able to run on the node already, even if drbd isn't ready
yet.

Sincerely,
Lars Marowsky-Brée <lmb@example.com>

--
Perfection is our goal, excellence will be tolerated. -- J. Yahl
RE: Re: meta-data handling implemented [ In reply to ]
> -----Original Message-----
> From: Lars Marowsky-Bree [mailto:lmb@example.com]
> Sent: Wednesday, May 16, 2001 8:48 AM
> To: Philipp Reisner
> Cc: Alan Robertson; drbd-devel@example.com
> Subject: Re: [DRBD-dev] Re: meta-data handling implemented
>
>
> On 2001-05-16T12:09:53,
> Philipp Reisner <philipp.reisner@example.com> said:
>
> > I am just in the progress of thinking of a solution which
> does not need
> > any support from the cluster manager at all.
> > (The /etc/init.d/drbd script which is started before the
> > /etc/init.d/heartbeat script will block until heartbeat can
> > do what it does ...)
>
> I disagree with blocking until the cluster manager can be
> started. Other
> resources might be able to run on the node already, even if
> drbd isn't ready
> yet.

As a datapoint: Most of the services that we control with Heartbeat are dependent on DRBD being up and ready. We take pains to insure that DRBD is up before allowing any other services to come up.

Just my $0.02.

(By the way Phillipp, it's great to hear of the meta-data progress!)


Tony Willoughby
Broadband Access Systems, Inc.
Re: Re: meta-data handling implemented [ In reply to ]
Lars Marowsky-Bree wrote:
>
> On 2001-05-16T12:09:53,
> Philipp Reisner <philipp.reisner@example.com> said:
>
> > I am just in the progress of thinking of a solution which does not need
> > any support from the cluster manager at all.
> > (The /etc/init.d/drbd script which is started before the
> > /etc/init.d/heartbeat script will block until heartbeat can
> > do what it does ...)
>
> I disagree with blocking until the cluster manager can be started. Other
> resources might be able to run on the node already, even if drbd isn't ready
> yet.

I don't think he means the system won't come up - I think he means
that DRBD won't do anything useful for anyone until it coordinates with
heartbeat. I strongly suspect that this is necessary...

He didn't say that heartbeat would block for drbd, but that drbd would block
for heartbeat.

I'll read your paper.

-- Alan Robertson
alanr@example.com