HP3000-L Archives

August 2001, Week 1

HP3000-L@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
David T Darnell <[log in to unmask]>
Reply To:
David T Darnell <[log in to unmask]>
Date:
Mon, 6 Aug 2001 08:22:34 -0700
Content-Type:
text/plain
Parts/Attachments:
text/plain (103 lines)
Leonard,

Gavin already issued a definitive answer to the question, but in explanation of my foible:

First of all, let me say that I gave out bad advice when I recommended this type of logging (ILR.) This is showing my HP1000 roots and the fact that I haven't done a lot of DBA work in the last few years!

I can not recommend Intrinsic Level Recovery (so I retract my post recommending it.) HP recommends against ILR.

First let me explain the logging modes that cover the conditions for which I was suggesting ILR:

Default Recovery Mode (XM) guards against physical inconsistency from program aborts. Here's the explanation from the HP 6.0 TurboIMAGE manual -
Default Recovery Mode. In default recovery mode, TurboIMAGE/XL uses an MPE/iX
 ,le system service, Transaction Management (XM), to ensure the structural integrity of
the database following a system failure. All modi ,cations to the database (DBPUTs,
DBDELETEs, and DBUPDATEs) are automatically logged to an MPE/iX XM log  ,le.
However, this XM log  ,le is only written to disk when one of the following situations is true:
A system-speci ,ed time has elapsed.
A request is made by a process to ush the log  ,le to disk, for example, a call to DBEND
mode 2 or 4.
The XM bu +er is full.
XM ensures that the TurboIMAGE/XL intrinsics are applied to the log  ,le in a serial manner.
If a system failure occurs, those completed intrinsics that have not been written to disk are
not recovered. Thus, one or more completed DBPUTs, DBDELETEs, and DBUPDATEs
can be lost, but the internal structure of the database remains consistent. Recovery with
DBRECOV must be performed at system startup time before anyone modi ,es the database.
==

So, after a system failure when database modifications could have been in progress, DBRECOV should be run. (Experts, how do we ensure that database access is not allowed until DBRECOV has been run?)

DBAs, it looks like this defauilt logging scheme will not help if the system fails before the transaction gets posted from the XM buffer to the log file. Is this correct?  If so, does Dynamic Roll-Back fill the gap?

From the manual:
Dynamic Roll-Back Recovery
Dynamic roll-back allows a more timely recovery of databases than is possible with
DBRECOV. Dynamic roll-back eliminates the overhead incurred when a database is enabled
for user logging and permits database access to continue, even while other users are accessing
a database.
===
If logging is speci ,ed and DBBEGIN/DBEND (static transactions) or DBXBEGIN/DBXEND
(dynamic transactions) are not used, TurboIMAGE/XL considers each DBPUT, DBDELETE,
and DBUPDATE to be a single logical transaction. While a transaction is executing, the
database is considered to be in an inconsistent state. Thus, each transaction takes the
database from one consistent state to another.
==

I must also ask, if all logging is turned off (like in the dbutil answer below) are we denying ourselves the benefit of even the XM default logging?

>>show advdb flags
  For database ADVDB

  Access is enabled.
  Autodefer is disabled.
  Dumping is disabled.
  Rollback recovery is disabled.
  Recovery is disabled.
  ILR is disabled.
  Mustrecover is disabled.
  Logging is disabled.
  Prefetch is disabled.
  Indexing is disabled.
  HWMPUT is disabled.
  Add entry into lock table for deadlock detect is enabled.
  Restart is disabled.
  Database last stored using True-Online Backup on
     SUN, AUG  5, 2001, 10:01 PM.
  Database has not been modified since last store date.
  Shadowing is disabled.
  Using Dependency SEMaphore(DSEM) is disabled.
  Subsystem access is READ/WRITE.
>>



-Dave "shot my mouth off on a topic in which I am not current" Darnell






[log in to unmask] on 08/03/2001 10:16:00 AM
To:     David T Darnell
cc:
Subject:        Re: [HP3000-L] Does ABORTJOB cause Image broken chains?



Can you explain what intrinsic level logging is? Is is different from Data
 Base
logging? Perhaps you ought to post the reply to the list.

Thanks.
===================
Leonard S. Berkowitz
Perot Health Care Systems
(Harvard Pilgrim Health Care account)
voice: 617-509-1212
fax:   617-509-3737
pager: 781-226-2431

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *

ATOM RSS1 RSS2