From: Kern S. <ke...@si...> - 2005-08-11 17:30:19
|
On Thursday 11 August 2005 16:53, Michael Mackey wrote: > Kern and all, > > Here are the results of my tests (using an HP A4846A 4-drive, 100 slot > tape library and a fresh copy of cvs sources on 8/9/2005): > > I have configured four pools, one for each drive. Configs and log can > be found in http://darwin.ecn.uiowa.edu/bacula > > When starting up from all empty drives, everything works fine. > > However, when I start a job with the drives loaded, bacula checks only > drives 0 and 1 for the desired volume. When the desired tape is in > drive 2 or 3 it is not seen. Instead, it tries to load from the library > slots, where of course there is no tape, and an error is > generated. Yes, I think you have pointed out a failure that I haven't yet addressed. That is if the SD sees that a tape is not in use and it wants that tape, it assumes that it is in the slot, and is not intelligent enough to know that it may be in another drive. I don't think that the problem is not really that it only checks drives 0 and 1 as you indicate. Rather it is that the SD finds that drive 1 (or perhaps 2, I not at home so I haven't looked in detail) is available, so it starts using it. If I make it a bit smarter, it can either know to unload drive 3 where the tape is loaded (for example) then load it into drive 1. Or it can continue looking and find that drive 3 is a suitable drive AND has the tape loaded as well. The logic here is a bit complicated so I need to carefully think about it. I think I will need to increase the number of passes through the available drives. The first pass will require an *exact* match, and subsequent passes will do what they currently do, with the exception that it needs to check if a slot is already loaded in a drive, which it currently does not do. Thanks for the analysis and the tests. Unless you see some major flaw in my logic above (not worrying about the exact drives...), I have enough information to go on for improving the search algorithm. If all goes well, I will be home this weekend and will work on it. Again, thanks for the testing and analysis -- it is invaluable. I'll look over your log in detail when I get home and can print it out ... Best regards, Kern > > Note: machine pisello-fd was not running at 1 am on 8/10, which explains > the errors at that time. Later (~10 am), a manual start of the pisello > job failed for the reason discussed above. > > Also, unmount / mount needs to be changed to support multiple drives (it > currently only sees one - drive 0) > > Thanks > > Michael Mackey |