2. When with multiple projects and local preferences entered and a project is detached, for whatever reason, the local prefs are lost/overwritten by the global prefs from another attached project (don't know which since 12 are attached]. The signs were that the LP connects was set to 0.01 and buffer 0.75. The values bacame 0.1 connect and 0.25 days buffer which may ring a bell. Network time changed from 12:45-13:00 to 00:00-00:00)

This one at least isn't Windows 7 related.

07-Jun-2009 17:53:48 [---] [wfd] target work buffer: 864.00 + 0.00 sec
07-Jun-2009 17:54:38 [SETI@home] Detaching from project
07-Jun-2009 17:54:43 [---] [wfd] target work buffer: 8640.00 + 21600.00 sec
... (now this is odd)
07-Jun-2009 17:55:58 [SETI@home] [wfd] request: CPU (0.00 sec, 0) CUDA (30240.00 sec, 1)
... (before initialisation?)
07-Jun-2009 17:56:03 [SETI@home] Master file download succeeded
07-Jun-2009 17:56:08 [SETI@home] [wfd] request: CPU (0.00 sec, 0) CUDA (30240.00 sec, 1)
07-Jun-2009 17:56:08 [SETI@home] Sending scheduler request: Project initialization.
... (some comms errors)
07-Jun-2009 18:00:33 [SETI@home] New computer location: school
07-Jun-2009 18:00:33 [SETI@home] General prefs: from SETI@home (last modified 04-Apr-2009 20:08:29)
07-Jun-2009 18:00:33 [---] General prefs: using separate prefs for school
07-Jun-2009 18:00:33 [---] Reading preferences override file
07-Jun-2009 18:00:39 [---] [wfd] target work buffer: 864.00 + 0.00 sec

Looking back at my report to boinc_alpha the following day ("target work buffer wrong after project detach"), that was with BOINC v6.6.31 running under Windows XP.

I don't think it reverts to preferences from another project: I think those are hard-wired defaults in the client itself.

Tested on WIndows XP SP3 with 6.10.19:

19-Nov-09 12:30:41 DrugDiscovery General prefs: from DrugDiscovery (last modified 19-Nov-2009 12:29:23)
19-Nov-09 12:30:41 DrugDiscovery Host location: none
19-Nov-09 12:30:41 DrugDiscovery General prefs: using your defaults
19-Nov-09 12:30:41 Reading preferences override file
19-Nov-09 12:30:41 Preferences limit memory usage when active to 2047.23MB
19-Nov-09 12:30:41 Preferences limit memory usage when idle to 2047.23MB
19-Nov-09 12:30:41 Preferences limit disk usage to 5.00GB

19-Nov-09 12:31:00 DrugDiscovery Resetting project
19-Nov-09 12:31:00 [cpu_sched_debug] Request CPU reschedule: exit_tasks
19-Nov-09 12:31:01 DrugDiscovery Detaching from project
19-Nov-09 12:31:01 [cpu_sched_debug] Request CPU reschedule: Detach
19-Nov-09 12:31:01 [work_fetch_debug] Request work fetch: Detach
19-Nov-09 12:31:01 [cpu_sched_debug] Request CPU reschedule: project detached by user
19-Nov-09 12:31:01 [work_fetch_debug] Request work fetch: project detached by user

19-Nov-09 12:31:07 SETI@home Scheduler request completed: got 1 new tasks
19-Nov-09 12:31:07 SETI@home [sched_op_debug] Server version 611
19-Nov-09 12:31:07 SETI@home Project requested delay of 11 seconds
19-Nov-09 12:31:07 SETI@home General prefs: from SETI@home (last modified 30-Oct-2009 01:14:02)
19-Nov-09 12:31:07 SETI@home Host location: none
19-Nov-09 12:31:07 SETI@home General prefs: using your defaults
19-Nov-09 12:31:07 Reading preferences override file
19-Nov-09 12:31:07 Preferences limit memory usage when active to 2047.23MB
19-Nov-09 12:31:07 Preferences limit memory usage when idle to 2047.23MB
19-Nov-09 12:31:07 Preferences limit disk usage to 5.00GB

As you can see, it's re-reading the local preferences automatically at any detach.

---

That's why I posted (and luckily still had available) the Work Fetch Debug logs: that part, at least, was running on the hard-wired preferences until I reattached. Didn't look at the GUI while it was all going on: I only wanted to detach/reattach briefly so the server would know about the 100-odd tasks that had been trashed by a big bug in v6.6.34 a few days before.

v6.10.19 is available for download (Windows only), and it should contain some important fixes I've been pestering for. I also want to do some serious research on debt accountancy in this version: one bug is among those fixed this time (subject to confirmation testing), but I suspect others remain.

Part of my plan for starting debts with a clean slate includes detaching (permanently) from LHC and Orbit, since neither project is showing any signs of life. I'll watch to see how long it takes before it notices the prefs override file this time.

As you can see, it's re-reading the local preferences automatically at any detach.

Well, it's re-reading at the next scheduler contact with another project, at least: might be a long time if you're running AQUA.

(I was going to say CPDN, but the trickle upload might be enough).

I doubt it forgets in the mean time that you used local preferences. In the least, the global_prefs_override.xml file isn't deleted between detaches, so they'll be overriding the global_prefs.xml at all times.

But I'll forward this thread to the developers anyway. Together with some other lingering Windows 7 troubles which may or may not have been fixed in .19

Edit: Just noticed that upon detaching from the project that holds the web-preferences, the global_prefs.xml file will be deleted. It won't return until the next project has been contacted and the previous preferences have been downloaded.

---

Installed v6.10.19, and carried out the 'detach from project' test - results much as previously described.

I fiddled some general preferences at Orbit, updated that and then a few other projects to check that they replicated properly - they did, once I'd remembered to ensure the host was set on the right venue.

Then, cleared local preferences (which had a smaller cache setting), refilled the cache according to venue settings, and re-established (restrictive) local prefs. Finally, detached from Orbit.

Confirmed that global_prefs.xml file was deleted (though global_prefs_override.xml remains). Also, the fields in the local preferences GUI now show the hardwired defaults, matching neither my web preferences nor my override preferences. And to clinch it, the host suspended CUDA processing, although I always choose to allow CUDA while the computer is in use. Sure enough, CUDA restarted automatically 3 minutes later - it's surprising how long that is, when you're looking at the screen but can't allow yourself to touch the mouse!

I've retreated upstairs, where I can monitor task switches etc. remotely (BoincView) without the computer being 'in use'. I'll examine the logs more closely when I next see that it's updated a project automatically.

Also, the fields in the local preferences GUI now show the hardwired defaults, matching neither my web preferences nor my override preferences.

What "hardwired defaults"?
Which values show? What should they be, according to you (or your global_prefs_override.xml file)?

For me it shows my zero day connect and 0.1 day additional, just as ordered.

---

Also, the fields in the local preferences GUI now show the hardwired defaults, matching neither my web preferences nor my override preferences.

What "hardwired defaults"?
Which values show? What should they be, according to you (or your global_prefs_override.xml file)?

For me it shows my zero day connect and 0.1 day additional, just as ordered.

Normally, that page defaults to the set in use: for the purposes of this test, 0.01 days CI / 0.75 days additional (web, venue 'school'), or 0.05 days CI / 0.3 days additional (local override). After I detached Orbit, the GUI was defaulting to 0.1 / 0.25, matching that log from 7 June (I think - I don't actually want to touch the GUI again for at least another hour, to see how it handles AQUA task switches without "request reschedule CPUs - active state change", or that interruption to CUDA work. I deliberately chose the settings to potentially delay the 'Reading preferences override file' event for several hours, if my theory about it only happening at project update is correct).

So far, it has handled the first few task switches from CPU to MT correctly, keeping all cores busy, although it has preempted an Einstein task after 4m 25s and a SETI (CPU) task after 2m 14s, both in favour of AQUA.

OK, second task switch completed successfully, so I think we have a winner.

I sacrificed 5 minutes of CUDA work:



According to the log, the preferences override file hasn't been read since 15:54, so I've been running on hardwired for well over 2 hours.

Yes! Almost a full six hours after

19-Nov-2009 15:57:51 [orbit@home] Detaching from project

I finally got

19-Nov-2009 21:38:25 [CPDN Beta] [wfd] request: 0.00 sec CPU (0.00 sec, 0) NVIDIA GPU (0.00 sec, 0)
19-Nov-2009 21:38:25 [CPDN Beta] Sending scheduler request: To send trickle-up message.
19-Nov-2009 21:38:25 [CPDN Beta] Not reporting or requesting tasks
19-Nov-2009 21:38:30 [CPDN Beta] Scheduler request completed
19-Nov-2009 21:38:30 [---] General prefs: from http://orbit.psi.edu/oah/ (last modified 19-Nov-2009 15:36:13)
19-Nov-2009 21:38:30 [---] Computer location: school
19-Nov-2009 21:38:30 [---] General prefs: using separate prefs for school
19-Nov-2009 21:38:30 [---] Reading preferences override file

and everything is back to normal - GUI shows 0.05 / 0.3, and I can use the mouse / keyboard without unloading CUDA tasks. And I can finally

19-Nov-2009 21:39:41 [SETI@home] Sending scheduler request: Requested by user.
19-Nov-2009 21:39:41 [SETI@home] Reporting 65 completed tasks, not requesting new tasks

without interrupting the experiment.

[OK. 'fess up time. Searching the log reveals

19-Nov-2009 20:05:21 [GPUGRID] Sending scheduler request: Requested by project.
19-Nov-2009 20:05:21 [GPUGRID] Not reporting or requesting tasks
19-Nov-2009 20:05:26 [GPUGRID] Scheduler request completed
19-Nov-2009 20:05:26 [SETI@home] General prefs: from SETI@home (last modified 04-Apr-2009 20:08:29)
19-Nov-2009 20:05:26 [SETI@home] Computer location: school
19-Nov-2009 20:05:26 [---] General prefs: using separate prefs for school
19-Nov-2009 20:05:26 [---] Reading preferences override file

but I was out for a beer at the time, so it doesn't count.]

No, to be fair the GUI (was) (would have been, if I'd chosen to look) showing the settings that were in effect.

The trouble is, those settings - for over four hours - bore no relation to my preferences, whether web preferences or local preferences.

I'll leave it to somebody else to do the Account Manager tests!

Ah - and of course, if I'd clicked 'OK' instead of 'cancel' to dismiss the GUI preferences dialog in the Manager, I'd have locked in the hardwired prefs set (including the unwanted settings on other tabs, like 'suspend GPU while computer in use') as my new local override set.

Re point 2 of the original post, the low allocation of CPU time on the W7-32 Centrino Duo...

That would be point 3 of the original post ;-)