IDS Forum

Re: Is there recovery from this?

Posted By: Martin Fuerderer
Date: Tuesday, 30 September 2003, at 3:33 a.m.

In Response To: Is there recovery from this? (Danny Wright)

Hi,

I cannot "confirm that upgrading while instance is up should have
this effect". But if you think about the effect of such actions it will
become more clear why such an effect is plausible ... :

So far there's no special check in the installation script that would
detect a running instance (that was started from the installation to
be upgraded), because this itself is not trivial (there was internal
discussion about this recently, but I don't know the result yet). It is
still assumed that only the administrator does upgrades of the
installation and that he knows what he's doing ... :)

Anyway, therefore when you do the upgrade, it means that binary
executables are copied onto the disk to the same location where
there is/was already one. If from this installation an instance was
started and is still running, then the system has a (potentially partial)
copy of the binary executable in memory for current execution. At
any time may the system (UNIX kernel) decide to load something
from this binary executable (on disk) into memory for execution. The
same applies for shared libs ... Due to the upgrade activity, the
original binary on disk is gone and there's now something different
(the new binary). So you can easily imagine that at this point the
kernel and your running binary executable will get into trouble ...

The effect is different, depending on platforms. E.g. on Solaris the
running binary is generally known to crash in some unspecified way.
On HP-UX the installation will get an error because the OS detected
that a binary is running in memory and will prevent overwriting the
corresponding binary on disk.

In your case (according to online.log excerpt) it looks like the running
binary managed to write some log record in the new format before it
(finally) crashed. Don't ask me how this was possible - I don't know,
but as I tried to explain above, the effects can be non-deterministic
(at least from my perspective as application programmer) ...

So this is a bad situation for the fast recovery, when it expects old
format log records and suddenly there are new format log records,
and that switch happened in the middle of an open transaction.
This is (obviously) not accounted for (and does not happen when
the system is brought down before upgrade).

With that I don't think that there would have been a way to bring the
system back up without the help of IBM Informix Support.

Regards,
Martin
--
Martin Fuerderer
IBM Informix Development Munich
Data Management Solutions





"Danny Wright" <dwright@sherwoodfoods.com>
Sent by: forum.subscriber@iiug.org
30.09.2003 02:14


To: ids@iiug.org
cc:
Subject: Is there recovery from this? [1953]



I upgraded Informix IDS from 7.24.UC7 to 7.31.UD6 over the weekend, and =
everything was fine with our production instance, but I couldn't get our =
Test instance back.

It seems like someone (me) forgot to shut down that instance before =
upgrading....I got errors complaining about shared memory, so I figured =
I'd free it up by rebooting....That didn't work. =20

Since it was only a Test instance that I planned to do a dbimport on =
anyway, I just recreated all the dbspaces, but I was wondering if there =
was anything else I could have done (short of calling Informix and =
hoping they could salvage it).

Can anyone confirm that upgrading while an instance is up should have =
this effect?
(This was the same instance my dbexport was 'hanging' on that I asked =
about a few weeks ago - I never resolved that, but I never had trouble =
bringing it online).

BTW, nobody was in that instance when the upgrade was done.

Thanks,
Danny Wright

Below is output from the online log when I attempted to bring it up.

11:01:39 Event alarms enabled. ALARMPROG =3D '/informix/etc/no_log.sh' =

11:01:44 DR: DRAUTO is 0 (Off)=20
11:01:44 AIX MP latch code enabled=20
11:01:44 Requested shared memory segment size rounded from 2132KB to =
2144KB=20
11:01:44 Informix Dynamic Server Version 7.31.UD6 Software Serial =
Number ......
11:01:45=20
11:01:45 (5) connection rejected - no calls allowed for sqlexec=20
11:01:45 listener-thread: err =3D -27002: oserr =3D 0: errstr =3D : No =
connections are allowed in Dynamic Server quiescent mode. =20
11:01:45 Informix Dynamic Server Initialized -- Shared Memory =
Initialized.=20
11:01:45 Physical Recovery Started.=20
11:01:45 Physical Recovery Complete: 0 Pages Restored.=20
11:01:45 Logical Recovery Started.=20
11:01:45 Open transaction detected when changing log versions.=20
11:01:46 Cannot Rollforward from Checkpoint.=20

Messages In This Thread

• Is there recovery from this?
  Danny Wright -- Monday, 29 September 2003, at 8:14 p.m.
  • Re: Is there recovery from this?
    Martin Fuerderer -- Tuesday, 30 September 2003, at 3:33 a.m.

IDS Forum is maintained by Administrator with WebBBS 5.12.