SUMMARY: multible local mounts to one filesystem

From: Schwarzkugler, Bernd <bernd_schwarzkugler_at_ibex-ag.de>
Date: Fri, 14 Dec 2001 09:22:19 +0100

My orginal question:

a question about multible mounting storage under Tru64 5.1 .
We have 4 Alpha Servers connected to a HSG80 with much storage connected.
Is the following possible?
1 Server localy mounts the filesystem /data read/write
1 Server localy mounts the filesystem /data readonly for backup
1 Server localy mounts the filesystem /data readonly for something else
1 Server localy mounts the filesystem /data readonly for something else

All Servers are mounting the filesystem at the same time, but only one
server will do write transactions to the filesystem /data

Does the filesystem /data crash or not?


The answers to the question:

>From Lucien HERCAUD:
Yes it does crashes (the 3 servers mounting /data in READONLY mode MAY
CRASH!)
unless you use the Tru64 clustering software and it's cluster file system.
Otherwise, the only thing you can do is MOUNT EVERYTHING READONLY but this
does not helps you a lot.

One more thing:
The /data filesystem will not get corrupted in the environment you described
... only the servers mounting it READONLY may crash at any time because of
the inconsistencies between the metadata on the disks and their copy of it
in the UBC (in memory)

>From Udo de Boer:
If you are in a cluster all the filesystems are available on each system.
And they will be rw.
But if you are on different machines and try to mount the same data you have
a problem. I can recall somewhere from this list that someone else tried to
do the same thing. But even in readonly mode AdvFS still writes something on
the disk. So this cannot be done.

If you are in a cluster look at the advfs clone option.

>From Darryl Cook:
should not be a problem....i mount /usr/local on about 6 different machines
read-only.

.. sounds interessting to me, wo what filesystem are you using?

>From John Francini:
Actually, the HSG80 should prevent all the servers except for the one
that mounts the filesystem read/write from seeing it. We tried just
this sort of scenario and found that unless the systems are clustered
together, only one system would be able to mount the filesystem.

I must ask the obvious question: why not cluster the servers? Then
everyone can have read/write access to the filesystem, since all
access will be arbitrated by the cluster file system (which
intervenes above the physical filesystems).

>From Dr. Udo Grabowski:
The filesystem will not crash, but the reading machines may
see different contents than the writer as they do local caching
and may not be aware that the contents has changed. For absolute
synchronicity, the machines must be clustered and must mount that
system under the Cluster File System (in which case one server
is the principal coordinator, but data flow may go through the
local HBAs [but not under 5.1, this works partly under 5.1A
for direct access files, 5.1 directs all traffic over the
principal coordinator]).


>From Alan:
The particulars depend on the file system. AdvFS can't be
mounted readonly, so there's no issue. You simply can't
do what you want.

UFS will allow you to mount one of more copies read-only,
unless the driver reserves the disks. However, the mounts
have no knowledge of each other and changes by the read-
write copy will look like file system corruption to the
read-only copies. The file system won't be damaged, but
the systems that think the file system is corrupt will crash.

If the systems are members of a cluster, then each can
mount the data read-write and that part isn't an issue.
Finding the data in a consistent still is, but that
would have been a problem the other way as well.

Conclusion:

Trucluster is the magic word. In this environment it is even safe to mount
read/write from more
than one node.


Thank you for your answers

Have fun...

 Bernd
Received on Fri Dec 14 2001 - 08:24:13 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:43 NZDT