Active/Active Apache and CFS

[Adam Garsha]

Looking at Enterprise Cluster Master Toolkit release notes, it looks like active/active is supported if you put the bits on a shared CFS and conf/logs local to each node. Installation and configuration makes sense to me.

Question: how does patching go? is there a process for rolling upgrades (e.g. if the patches are for apache itself or modules?)

Thanks much! I am a point giving fool.

[Adam Garsha]

To clarify, how do you patch if bits are shared? Do you have to stop all nodes or can you just stop one node, patch and restart apache on the alternate nodes?

But what if shared libs are patched... how can that work?

[Stephen Doud · Accepted solution]

The ECM Release Notes at http://docs.hp.com/en/T1909-90054/ch01s03.html states the following:

HP Apache supports CVM/CFS 4.1 and 5.0 running Serviceguard version A.11.18 on HP-UX 11i v2 and can be configured in the following two modes:

Active-Active: In an active-active configuration, multiple nodes can run an Apache instance while concurrently serving the same document root. To coordinate the startup and shutdown of the Apache instances with cluster node startup and shutdown, create a one-node package for each node that runs an Apache instance.

Active-Passive: In an active-passive configuration an instance of Apache web server can run on only one node at any time. A package of this configuration is a typical failover package. The active-passive support on CVM/CFS comes with a caution/limitation that no attempts should be made to start the same instance of Apache on any other node while an Apache instance is up on one node.

NOTE: In both active-active and active-passive configurations, the HP Apache installation on all nodes must reside on the shared CFS mount point.
For complete details on how to set up an Apache Web server on CVM/CFS, refer the HP Apache white paper, Configuring HP Apache Veritas CFS with HP Serviceguard on CFS/CVM under the generic section HP Apache-based Web Server, v.2.0.x for techtips at http://www.hp.com/products1/unix/webservers/apache/techtips/index.html

---- end paste ----

In the NOTE above, the Apache bits reside on the shared CFS mount point. Since this indicated a single-location for the bits, it would appear to require downtown when it is necessary to update or patch the bits.
It would also seem to be a single point of failure since there is no redundancy in the bits.

However an official statement should be sought from the Apache support group at the HP Support Center.

[Heironimus]

I was curious so I went to look at HP's document. I've worked with Apache on some pretty critical systems, and I really don't think their described approach for active-active clustering is a good idea. The document read like a generic recipe for running an app under ServiceGuard, I could see putting some applications' binaries on a shared filesystem but it just doesn't seem appropriate for Apache.

[Shiny Sebastian]

Hi Adam

Well if my reply is not too late, i would suggest something. Well I am one of the ECMT team members and also, one of the authors for the document that speaks about active-active and active-passive detailed configuration of apache.
As i feel, if rolling upgrade is your concern, then you could also install the binaries locally on each node of the cluster.
Both the configurations have been designed keeping in mind the apache server conf files locations. As per the document, binary remains on the shared storage for both the configurations.

Considering the two cases:

1) Active-active: But if you can place binaries locally, then you could halt the pkg on one node and let it be run on the rest so that high availability is guaranteed and appropriately upgrade the application server.

2) Active-Pssive: You could failover the node from the node where you want to upgrade the application to the adoptive node of the package. Once your up-gradation is done you could failback the package to the former node where up-gradation is complete.

In case of concerns i think you can communicate with the appropriate apache support team.

[Steven E. Protter]

Shalom,

You don't need serviceguard for this.

Apache works just fine against shared storage two systems against the same disk. It will work against NFS, though certain files need to be on local disk.

CFS is unneeded the web content can be on NFS.

A rolling update will work fine.

The only real value SG brings to the table is a floating IP address. That could be done via a script.

Patching will work like any rolling update in SG.

SEP