Does anyone know the current status of INGRES clustering support, not
the HA solutions supported for Solaris and VMS, but specifically
GNU/Linux? There was something "in the works" with Red Hat Enterprise
Linux but I don't see anything about it in the documentation or on the
Ingres website.

I ask because I'm working on a few variations of a clustered iSCSI/Samba
storage implementation and we're using RHEL5.3AP and AFAIK the work that
was/is being done was specifically using the Red Hat Cluster Suite. At
the risk of starting an unproductive argument, MySQL offers integration
of DRBD (iSCSI Target block-device mirroring), and we're using DRBD for
mirroring, which prompted the recollection of the above.

--
***
Mark R. Winston
***

On Feb 21, 2009, at 1:50 PM, Mark R. Winston wrote:

This is not an "inside" answer, but ...

The VMS work nearing completion is for true active/active clustering;
in fact, a good deal of the most recent work centered around
fixes for distributed deadlock detection. The last time true clustering
worked in production was 6.4, and it had a totally different deadlock
detection architecture, completely incompatible with the way Ingres
does things now. There has been an awful lot of stress testing, and
fixing of bugs that would take a week or so of running to put in
an appearance. (Ouch.)

Anyway, I don't think any effort has been put into the linux side of
things for a while. I am reasonably sure that if and when linux
clustering is revived, most of the work will be integrating with a
different distributed lock manager. The old linux cluster effort
used OpenDLM, which turned out to be a dead end. I doubt
that OpenDLM is worth rescuing, since there is at least one DLM
out there now as part of one of the cluster filesystems (not sure
which one at the moment); one might assume that a DLM
associated with a working filesystem might itself actually work,
unlike OpenDLM.

Anyone wanting an active/active cluster solution on linux today
would probably start with the latest open source "main", and would
need to figure out what to do on the DLM side of things. The
Ingres/DLM interface is reasonably well isolated into the CX
area of the CL, but who knows what adaptations might be needed.

It's entirely possible that stuff has been going on behind the curtain
that I don't know about; the above is my personal best guess at
the status.

Karl

Karl's assessment is pretty accurate (as usual).

Active-Active clustering is available on Alpha OpenVMS at this time,
and should be available on Itanium OpenVMS in the near future. Other
platforms (Linux and Solaris) are Active-Passive (i.e., failover)
cluster configurations.

Something to keep in mind relative to clustering is that there are
really two distinct aspects: Managing client access; and managing
server access.

Ultimately it comes down to availability and service levels (relative
to performance and unplanned outages). There are so many ways to
measure availability, and so many components that can affect it (e.g.,
local network connection outage, single application or server failure,
regional issues, etc.) that the key is determining what a company
needs from a business perspective, rather than from a product check-
box perspective. High-availability solutions can be implemented with
replication products like GoldenGate and HVR on any platform,
regardless of the official clustering capabilities supported on those
platforms.

Clusters are usually used for failover (active-active or active-
passive) or scalability (active-active). From a scalability
perspective the overhead of the distributed lock manager (DLM) can and
usually does limit the benefits of having multiple active cluster
nodes (with almost any product on any platform). What this really
buys you is having "cluster aware" software. That same level of
awareness can be introduced into an active-passive environment (I've
done it many, many times), but it is more of a manual process to
create automated procedures to manage the environment. Not a big deal
since they really only have to be developed once.

Ingres is very robust so the need to failover is minimized, but there
are times when that is desirable from a hardware perspective. Below
are links to two white papers that outline two types of clusters that
I've implemented with Ingres in the past.

http://www.comp-soln.com/HA2_whitepaper.pdf
http://www.comp-soln.com/HA_whitepaper.pdf

So, a little bit of a tangent based on the original question, but
hopefully this information will be useful. There are many ways to
create a high-availability Ingres configuration that will meet your
business and technical needs. The ideal solution will be based on
business requirements, platforms involved, and budget.

Hope this helps.

Chip