From: Willem v. d. W. <wav...@gm...> - 2011-09-03 19:39:08
|
<html> <head> <meta content="text/html; charset=UTF-8" http-equiv="Content-Type"> </head> <body bgcolor="#FFFFFF" text="#000000"> <br> Case 1: <br> eXist Version: 1.4.1dev<br> eXist Build: 20110522<br> SVN Revision: 14490<br> Operating System: Linux 2.6.18-164.el5 amd64<br> Java<br> Vendor: Sun Microsystems Inc.<br> Version: 1.6.0_21<br> Implementation: Java HotSpot(TM) 64-Bit Server VM<br> <br> Case 2:<br> <br> eXist Version: 1.4.1dev<br> eXist Build: 20110611<br> SVN Revision: 14669<br> Operating System: Linux 2.6.18-194.el5xen amd64<br> Java<br> Vendor: Sun Microsystems Inc.<br> Version: 1.6.0_21<br> Implementation: Java HotSpot(TM) 64-Bit Server VM<br> <br> <br> Willem<br> <br> On 2011/09/03 09:10 PM, Dmitriy Shabanov wrote: <blockquote cite="mid:CADD4p=6sT...@ma..." type="cite">eXist version, please?<br> <div><br> </div> <div>That issues was discuses and it should be fixed for 1.4.1 <br> </div> <div><br> </div> <div class="gmail_quote">On Thu, Sep 1, 2011 at 12:14 PM, Willem van der Westhuizen <span dir="ltr"><<a moz-do-not-send="true" href="mailto:wav...@gm...">wav...@gm...</a>></span> wrote:<br> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">I want to report some issues with backup and restores for your consideration.<br> <br> We have recently had two production databases that developed errors in the dom.dbx (as reported in the report.log). The causes for these errors were external, a hard drive filling up unexpectedly, and a problematic xquery locking up the database repeatedly leading to forced shutdown and restart sequences over a period of time. In both cases the backup / restore functions partially failed.<br> <br> We noticed that in the presence of these dom.dbx errors the backups generated were becoming unreliable. Backups generated from the backup.sh would exit prematurely, excluding large amounts of data, and backups triggered through the user interface failed in two ways:<br> <br> 1. In one case it exited completely before finishing leaving an invalid zip file. Fortunately, in this case the backup.sh backup worked correctly and we could restore the database completely.<br> <br> 2. In the other case the data in the zip file was complete, but there were inconsistencies in the __content__.xml file entries that would omit all sub-collections of a particular name. In this case all sub-collections with the name "workspace" were present in the data backed up, but absent in the __content__.xml, hence not restoring when we restored the database. There were another few sub-collection terms that had the same problem. We were able to restore the database in the end by creating restore shellscripts for each "workspace" sub-collection, and restoring them individually. The bulk of the database consist of collections for which the basic structure is repeated. It seems to happen that if there is a dom.dbx corruption on one of these sub-collections, all the other collection structures with the same sub-collection is affected when creating the __content__.xml file.<br> <br> It is always best to try and prevent dom.dbx errors. But it seems that there might be some ways in which the backup and restore procedures could be improved, particularly not to make the creation of the __content__.xml dependent on the possibly broken indexes. Since the data is still present, it should be possible to create the __content__.xml in a way that would not be affected by index corruptions.<br> </blockquote> </div> <br> -- <br> Dmitriy Shabanov<br> </blockquote> <br> </body> </html> |