From: Bastian F. <ba...@ba...> - 2007-09-02 10:33:53
|
Hi, On Sonntag 02 September 2007, Kern Sibbald wrote: [...] > > FIFOs. This feature is crucial for our needs: Backing up process > > output instead of files. [...] > > The big problems are that this project is currently rather ill > defined, and it is rather low in user priority.=20 The votes obviously show this low priority; I frankly don't understand=20 that, though. Every organization/company/unit where a database is set=20 up (including, but not restricted to SQL DBMS) should need a database=20 backup system - something that is not currently available in Bacula,=20 and possibly the most significant difference from the commercial=20 competitors. YMMV. > It suffers from some=20 > of the same problems that plugins face The reason I used the subject I used is that I regard the two subjects=20 as potentially equivalent: A set of two executables (binary, shell=20 scripts, ...) in fact defines an interface for a plugin. When I wrote my initial mail, I had not yet read James' posting on the=20 topic. His concerns about incremental and differential backups possibly=20 result in a third "boolean executable" for the decision on whether to=20 back up a file/fifo/item. > -- that is how to make the=20 > Volume self contained after adding this kind of feature. The Volume > format currently is totally self contained in that a restore requires > only the FD -- it doesn't even need a current definition of a FileSet > to work. This of course is correct for all machines of interest; with a little=20 abstraction, though, it is only valid for systems that have a similar=20 semantics on their file systems. Adding plugins adds a level of=20 semantics. Unfortunately, the current level of file system semantics as required by=20 Bacula is used by approximately 100% of the systems out there (side=20 note: the file system available on my PalmOS PDA would _not_ be=20 sufficient for a Bacula restore), while adding the plugins would lower=20 this level to, well, approximately 0% :-/ > 1. Some important design decisions that permit the Volume to remain > self contained and independent of FileSet definitions. In my oppinion, "if plugins (executables) are used during backup,=20 equivalent plugins need to be in place during restore, or else a=20 fallback (e.g. writing stdout to a certain file) jumps in" would be a=20 valid decision. Independence of a FileSet definition can be achieved by storing the=20 fact "is plugin/executable based, and that plugin is..." along with the=20 data - e.g. by using a different file name schema, similarly to how=20 it's done in afbackup. > 2. A programmer. Er. Yes. :) Thx, Bastian =2D-=20 Bastian Friedrich ba...@ba... Adress & Fon available on my HP http://www.bastian-friedrich.de/ \~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\ \ FATAL ERROR; SYSTEM HALTED; Press any key to do nothing. |