Workaround for Error 909

[wirthi]

Hi,

I am just curious: When the client gets the error-message 909 and thus displays "909 structure write error", WHY does it delete the buffer-file? I mean, the client seems to KNOW that an error has occured (that the files can't be uploaded) - WHY does it destroy the work?

In my opinion, whenever error 909 occury, just go on folding (and save the buffer on disk). And when the client has finished another 5000 folds, just try again!

I guess it would be better to correct the code on the server that produces this error, but this method seems to be a quick and easy workaround for the client.

Greets,
Wirthi

[Brian the Fist]

This is already the intended behaviour. If ANY error occurs on upload, the data is retained in buffer to try again later. If this does NOT happen, it is a bug, although looking again quickly at the code it looks right. I will look at this more closely to see though if I can figure it out, next week, if you say your work was being deleted.

As for fixing the backend, it cant be fixed, we are completely changing it. It really is a write error.

[wirthi]

Hi,

I am not absolutely sure if the client buffered MY work (my computers are not running 24/7 currently, so it's hard to say), but I saw others write (or assume) their clients didn't buffer the work-units.

(See Thread http://www.free-dc.org/forum/showthread.php3?s=&threadid=1116)

Greets,
Wirthi

[IronBits]

None of mine were buffered...

[Paratima]

Mine had some odd behaviour...

It looked as if each box had one buffer backed-up. When it finished, it would upload a buffer (not sure whether it was the oldest or the newest), get the 909 error, then press on folding.

The work definitely WAS being deleted in spite of the error. Credit was definitely NOT given. I definitely DID shut down as much of my farm as I could conveniently get to, and put those boxen on no-net until the flood waters subsided and the rescue troops arrived. :D

[Brian the Fist]

Ok, like I said Ill take a closer look and if I find it to be a bug it will be fixed.

[Brian the Fist]

Ok, you were right (again). This is a subtle bug which was not trivial. What really happens is your data DOES get buffered the first time round for the structure error. However, when it tried to upload it a second time later on, the server thinks you've already sent this data before (because we check for duplicate data bein uploaded of course :) ). Thus it won't credit you for the data and it will tell your client to wipe it because it has the data. Thus it would always just have one fileset buffered no matter how many you made, as the server would keep accepting the previous ones as duplicate data. So the logic flaw here is that when a 909 structure error occurs, the server should 'undo' its record of ever having received that data set. (Im being purposely vague here on how we track datasets). Anyhoo, Ill make the server more careful about when it actually marks a dataset as received, ensuring none of the many possible errors occur.

Thanks for your patience and for pointing out this hard-to-find bug.

[bubbadog]

And thank you, Howard, for your quick response. This is exactly what makes this project such a pleasure! :thumbs:

[Paratima]

Originally posted by bubbadog
And thank you, Howard, for your quick response. This is exactly what makes this project such a pleasure! :thumbs:

Amen! And it would be good to post something about this particular fix on the DF site, 'cause a lot of folks felt pretty burned by this bug. Need to let everybody know the waters are safe again. ;)

[MAD-ness]

Definitely the loss of results/credit/data angered more than a few people.

However, I think that 99% of participants understand that software is prone to bugs, no matter how well coded or debugged said software is. The fact that Howard is willing and able to find and fix any bugs that are discovered is a large part of why many people who started running the DF client early on have stuck with it so consistently - despite the inevitable bumps along the way.

Keep up the good work. :)

[Jodie]

Yeah, no doubt.

Any of the 'angered' people ever run an operating system? Any operating system? A word processor, maybe? Spreadsheet? Game? Ever turned on a computer?

Software has bugs. Period. End of discussion on that point. What it comes down to is how quickly they get shot.

You're looking at (apparently) one guy shooting a crudload of them in extreme rapidity and testing his own software. Which is a huge no-no in the industry. A programmer should never look at their own software. Not objective enough.

Anyway - hey, *I'm* impressed, if that counts for anything. I have a couple of dozen programmers here. But if Mr. TheFist ever decides he wants a job slingin' code out in Cali - he knows where to call, I hope. :cheers: :|party|: :smoking: