Tuesday, May 05, 2009

runInstaller problems: a tale of no support

Granted, I am in a rather cynical mood; after painfully working my way through several SRs this past week (about 6 on Enterprise Manager and/or agents alone), I was not in the mood for any more games. So I gritted my teeth rather loudly when I had to open another SR about the Oracle Universal Installer. Or maybe we should call it the "Oracle Undetermined Installer"? Oracle Unbeknowst Installer?

We were doing a lot of Oracle Home installs. Yes, I could simply have clone a Home (choose your poison), but this was for a class and I wanted students to go through the experience. And we were using the silent installer, because that is just so much cooler. And much faster too boot. So half-way through, we start seeing something really strange - we would get the normal banner, the "Please Wait...." message and then... nothing. No matter how long you waited, nothing. The annoying part is that runInstaller likes to kick things off in the background, so it is not unusual to see some messages flash by, but one can still navigate. So if you see "Please Wait..." and yet can still type, one thinks the installer is still running. One would be wrong.

The Installer was simply aborting. The logs had absolutely no meaningful help in them whatsoever. So I pleaded my case with the smart folks on oracle-l and filed a case with Oracle. As always, I received some interesting information from oracle-l quickly, and Oracle Support just sorta... slowly... asked for an RDA (why?!? For all that is sacred and worthy, why on earth do you need an RDA for this?!?), asked for versions, and without blinking an eye told me my OS was unsupported.

Flustered, I fired up the Installer in graphical mode. Lo and behold, I get an error message saying /tmp was full (or rather, only 50mb was free, which is apparently not enough room for the temp JRE it wants to kick off). So after deleting some crap that the installer left behind from previous runs, I can now run the silent installer.

Thanks, Oracle Support, for being so helpful.

The OUI seems really antiquated. No, I take that back. It does some really awesome stuff when dealing with clusters! Amazing! But simple things like having a fully documented silent installer with cookbook response files are quite lacking. And the total absense of helpful error messages? But like many things with Oracle, they concentrate on the "bright shiny new things" and often do not turn back to polish the older things. No, that is not fair either - that is quite wrong. Oracle does go back and fix some things rather well. But the things that get left behind or fall through the cracks, they cause one to stumble, and are thus more noticeable. =) Unfortunately.

2 comments:

Amit said...

Charles,

I would recommend running rda or hcve scripts from metalink for performing pre-install checks before invoking runInstaller.

One good documentation for silent install can be found on Pythian website. Please refer to following link

http://www.pythian.com/news/1035/oracle-silent-mode-part-110-installation-of-102-and-111-databases

Cheers
Amit

Charles Schultz said...

Grégory Guillou's article is good, thanks. I am bit concerned that one has to run a different set of scripts as a pre-install step. Humans want simplicity, and a "simple" installer would do all pre-install checks inherently. A part of the problem is that we are running on a "unsupported" OS, so we want to skip _some_ of the checks, but not all of them.

One thing the Pythian post does not point out is that the silent installer suppresses a majority of the information (hence, _silent_). This suppression is not helpful when you hit problems and you want to know what the error was.