Windows S5R2 App 4.33 available for Beta Test

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820
Topic 13574

A new Windows App is available from our Beta Test Page.

This is our new "release candidate" that, if the Beta test doesn't reveal any serious problems, will be made "official" in the next few days. It should then fix at least the serious "Access Violations" in the "houghmap" module.

There is not much change to the 4.32 App, except for even more paranoid error checking and reporting to track down the remaining problems (including the fancy "empty or vanished checkpoint file" (resulting in an exit code 10)).

Please test and report.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

Windows S5R2 App 4.33 available for Beta Test

Thanks for spotting this!

I fixed the app_info.xml and updated the package (and the md5 sum).

Sorry for this.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

RE: Here is the results

Quote:
Here is the results list of a machine that was crashing on previous apps but then had success when switched to 4.33. The success was shortlived as there have now been two crashes since the "success" result whilst still on 4.33. Hopefully Bernd might be able to make some sense from this.


Thank you!

This is indeed very helpful, though it was not what I was hoping for. It means that the code we inserted to fix the access violation addresses the right problem, but doesn't do everything that is needed to fix it completely.

Any clue why you didn't get the symbols from the PDB ("einstein_S5R2_4.33_windows_intelx86.exe (-nosymbols- Symbols Loaded)")? Was the Einstein@Home main server not accessible at that time?

Please try to put a file named "EAH_MSC_BREAKPOINT" (w/o extension) into the BOINC directory of that machine (maybe suspend running tasks first and start new ones) and restart BOINC. Each task should fail with a client error, but it should download the PDB from the symbol store (which may take some seconds), the stderr should list a "breakpoint encountered". Stop the BOINC Client, remove the file, and start it again for (hopefully) normal operation.

EDIT: I put up the PDB here for manual download. Putting it into the project directory beside the App file should also lead to useful stackdump.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

RE: I've now downloaded the

Quote:
I've now downloaded the .pdb and deployed it and stopped and restarted BOINC on that machine. The current result has clocked up over 20 hours so far without incident. About another 35 hours to go to completion.


Thanks. You'll probably understand that I'm hoping that the error occurs again :-)

Quote:
The following snippet comes from the messages tab of Boinc Manager at the time of one of the crashes. There doesn't seem to be any attempt to download the .pdb from the server.


The PDB is not downloaded by the BOINC Client, but by the debugger embedded in the App.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

RE: K, I guess I just

Quote:
K, I guess I just assumed that BOINC would be used for all transfers like this. You can tell that I wasn't really paying attention earlier when you talked about the app "phoning home" to download the debugging symbols when needed :). So why didn't it do what it was supposed to do? Is it some misconfiguration with my LAN? It would be really galling if there are no more crashes to analyse :).


Don't know yet. I'll do a bit more of testing myself when I have more internet access than just a browser again.

There is a small chance that something is wrong with the symbol store on the server (http://einstein.phys.uwm.edu/symstore/einstein_S5R2_4.33_windows_intelx86.pdb/2545763533B44450B32318278AD17A631/einstein_S5R2_4.33_windows_intelx86.pd_)
or has been at that time. Apparently a DNS server (possibly @UWM) has occasional problems. There is also a small chance that for whatever reason the checksum in the symbol store doesn't match the one encoded in the App, e.g. if something went wrong during the build or the transfer.

The "Breakpoint" feature I described had been put into the App for testing the symbol store communication, you may want to give it a shot. If you get a -nosymbols- line again, then something went wrong again, otherwise you should get a proper stackdump, listing source code lines in the "worker thread" section instead of the "SymGetLineFromAddr()" calls of the previous dumps.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

Gary, your host #704557 has

Gary, your host #704557 has just returned 4 results with access violations and -nosymbols-. No clue what happened there, apparently not the "Fixing..." problem. It's probably not the same host where you adjusted the Firewall or manually added the PDB? Can you do it there, to?

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

RE: Hey all, I think about

Quote:
Hey all, I think about to get the Beta Version. Question: Since the automatic download of this .pdb file is not working in the current version


It works fine on my machines (I just checked). It might indeed be a firewall setting or sth. that blocks it on Garys machines. But it's definitely working in principle.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

As some firewalls etc.

As some firewalls etc. apparently are blocking to download the PDB by the App, we again bundled the PDB file with the official App. This might be an intermediate measurement while we rely so strongly on the debugger feedback.

I don't know for sure, but my guess is that this error when deleting the app_info.xml happens as the client finds the pdb file missing and thus errors the result. To test (and hopefully avoid) this, try to downlad the pdb manually from the link I gave in a message to Gary before removing the app_info.xml.

Gary, are you using a proxy? I'm afraid the App doesn't inherit the proxy settings from the client.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

RE: [8/19/2007 4:39:21

Quote:
[8/19/2007 4:39:21 PM|Einstein@Home|[error] Application file einstein_S5R2_4.33_windows_intelx86.exe missing signature
8/19/2007 4:39:21 PM|Einstein@Home|[error] BOINC cannot accept this file


Surprising. I've never seen that message before. Did switching back from a previous Beta App work before with the same version of the BOINC Client?

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

Probably switching back

Probably switching back doesn't work since we enabled verify_files_on_app_start. Sorry, we didn't think about this side effect for Beta tests. I'll see if there's any non-destructive way of switching back to the official App path from the Beta.

For now I recommend to NOT remove the app_info.xml while you have tasks in progress with more than 5% done.

Probably this week we'll get a new Windows Beta App anyway (still debugging...).

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Joined: 15 Oct 04
Posts: 2,684
Credit: 25,950,161
RAC: 34,820

The new Beta App is here. BM

The new Beta App is here.

BM

BM

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.