Applying patches ahead of time...

[Rich Hearn]

Hi all,

For the record... I know you "normally" apply the patche(s), then re-boot as soon as possible (if not sooner :^)

Now for my "insane" question - I'd like to install Update v9.0 to my live systems (2 node cluster - each has a system disk and a common disk between them for non-sys specific stuff), but don't know when I'll get to re-boot them (could be months).

My thought being if the systems should "crash" before I get down-time, the new Update v9.0 will be installed automatically on re-boot.

What would be the down-side? Could the application of the patch somehow cause a crash by some newer non-re-boot required code being used, but the re-boot required code not accepting it causing a crash?

Thoughts welcome,
Rich

[Robert Gezelter · Accepted solution]

Rich,

The "downside" is that your configuration has a problem that prevents the system from restarting.

The "half-applied" scenario mentioned is also a possibility.

- Bob Gezelter, http://www.rlgsc.com

[John Gillings]

Rich,

"It depends". Many patches don't need an immediate reboot, but occasionally they do (noted in the "reboot required" comment).

The effect of not rebooting obviously depends on the nature of the patch. You could end up with existing processes running against the old image, and new processes running against a new one. This may, or may not cause problems.

To satisfy your requirement without actually installing the patches, you could stage them in a known area and put code in your startup and shutdown procedures which check for pending patches, install them, then delete or move the files that were just installed, then initiate a reboot. Obviously you'd need some fairly extensive testing and safeguards to prevent a reboot loop.

On the other hand, if your system crashes, do you really want it to be rebooting into a new and untested environment? Maybe it would be better to just stage the patches. If you have an unexpected outage, manually install them and reboot.

(there's some accepted wisdom that complex systems should not be allowed to reboot by themselves, especially after a crash. Let them restart to a minimal system, then have the system manager manually initiate application startups, after first validating that the system is in a consistent state - it all depends on what your system does...)

[Rich Hearn]

Bob,

Thanks for your viewpoint. I won't argue the fact that having successfully re-booted many times in the past, I tend to *assume* that the trend will continue. BTW: How does one calculate the odds for the probability of the "half-applied" scenario :^)


John

Regarding "staging" the patches, I can only think of a shutdown *.com that works *flawlessly* on the test cluster, but I have yet to figure out what's happening/not happening that causes problems on the live cluster when it's used. BTW, other than the hardware (DS20's vs ES47's) the systems are "identical" - uh... right! we've all been there... :^)

As far as re-booting into a new & untested environment, *anytime* patches are applied and a re-boot is done it becomes an untested environment - test clusters just find "the obvious" major issues :^) (better than nothing tho')

fyi: on a re-boot, only VMS gets started - as you stated, we manually start the App.

Gentlemen, "Thank You" both for your time and thoughts that provided a "sanity check" for me.

Rich