SOLVED [Bitcoins.lc] Issue with invalid shares (10 BTC bounty)

[Phil21]

This has been bugging the crap out of me since I saw you refused to release the actual problem.

I'm offering to pay you 5BTC should you release a detailed explanation of the problem, and the fix.  I don't even mine in your pool (I actually hopped on IRC the first day you started to give you a couple suggestions/features that I would have needed to join).  This way, you are now splitting the cost and the entire community as a whole benefits.

Or heck, just PM me the problem and I'll write up the problem/solution myself after I test it in a dev environment.

-Phil

[jgarzik]

I'm planning to keep that to me self, due to i think the other pools based on pushpoold will eventually face the same problem.
This issue costed the pool/me personally 10 BTC + a whole day of hard work with debugging, tcpdump and ~10 people that tried in all ways to help us.

So...  you won't even tell the author of pushpool?

pushpool certainly cost me a lot more than "10 BTC + a whole day of hard work."

Zip for donations or thanks, just people asking for free support all the time, too.

[lizthegrey]

I'm offering to pay you 5BTC should you release a detailed explanation of the problem, and the fix.  I don't even mine in your pool (I actually hopped on IRC the first day you started to give you a couple suggestions/features that I would have needed to join).  This way, you are now splitting the cost and the entire community as a whole benefits.

Code:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

As one of the people who was trying to help track the bug down, I'm a little offended too with the "I'm going to withhold the fix", especially given that Pushpool is open source and cost a lot more than a day to develop.

I'm going to raise the bounty by another 2 btc for releasing the fix, but encourage you to just do the right thing instead of being mercenary and taking the money.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)

iEYEARECAAYFAk30xMYACgkQaj+LAPvd0qQCtQCfQjbjz0HLky0RwIO6j1lqLw36
3fwAmwSbONdoFibFZ/jbZ/bjGE8RIf53
=m0/A
-----END PGP SIGNATURE-----

[padrino]

Did he pay for the fix, if so I find myself conflicted about whether or not it should be released. Maybe cover his cost plus what he lost and he should step up in the spirit of the community. You may want to go to the source of the fix and see if he will discuss since he posted his name earlier.

[Jine]

The problem lies in poclbm's support for X-Roll-NTime.
There are 2 patches for this applied to our source, first disable the header that allows X-Roll-Ntime.
And secondly disabling the support completely.

Seems to have fixed all issues we had.

jgarzik: The patch you merged that added support that, consisted of 2 commits for those two - we undid them.

Now - back to coding, our pool is growing extremely fast.
Sorry about the withhold of the solution. I'm all in for the sprit of the community, and all further patches will be publicly available at my own github.

We're _STILL_ having problems with low number of duplicates, but not even close to 5-6%.

Once again, sorry.
If you want to, donate the "bounty" to ius which works hard to help us out with both bugs and improvements.

[lizthegrey]

If you want to, donate the "bounty" to ius which works hard to help us out with both bugs and improvements.

http://blockexplorer.com/tx/6647483f1329e5fab259b5d5619a7c20a28d6b440a1d866d25971dba78001912

[kjj]

The remaining duplicates are probably coming from the clock problem ArtForz found.  Since you are running NTP on your server, I recommend patching bitcoind to disable the crappy clock adjustment.

The file is util.cpp, look for function GetAdjustedTime().

Code:

int64 GetAdjustedTime()
{
    return GetTime() + nTimeOffset;
}

Change to:

Code:

int64 GetAdjustedTime()
{
    return GetTime();
}

That should probably end up in the official client, in case any devs are watching.  If they really want to keep the clock adjustment even though NTP does a much better job, the getwork() function in rpc.cpp should be changed.  pBlock->nTime needs to be set to nPrevTime, which increases monotonically, after the call to IncrementExtraNonce().

[padrino]

The remaining duplicates are probably coming from the clock problem ArtForz found.  Since you are running NTP on your server, I recommend patching bitcoind to disable the crappy clock adjustment.

The file is util.cpp, look for function GetAdjustedTime().

Code:

int64 GetAdjustedTime()
{
    return GetTime() + nTimeOffset;
}

Change to:

Code:

int64 GetAdjustedTime()
{
    return GetTime();
}

That should probably end up in the official client, in case any devs are watching.  If they really want to keep the clock adjustment even though NTP does a much better job, the getwork() function in rpc.cpp should be changed.  pBlock->nTime needs to be set to nPrevTime, which increases monotonically, after the call to IncrementExtraNonce().

Great point, I was wondering about that myself and can definitely see why it causes issues. Applying this to the official client with an NTP flag for those running right time may be worthwhile. Some PCs are great at it while others are poor because of hardware and/or software.

[kjj]

The client will already complain if the clock is wrong by a lot.  I think that should be enough.  NTP is trivial to install, configure, and run on any OS.

[Jobbernowl]

The problem lies in poclbm's support for X-Roll-NTime.
There are 2 patches for this applied to our source, first disable the header that allows X-Roll-Ntime.
And secondly disabling the support completely.

Seems to have fixed all issues we had.

jgarzik: The patch you merged that added support that, consisted of 2 commits for those two - we undid them.

Now - back to coding, our pool is growing extremely fast.
Sorry about the withhold of the solution. I'm all in for the sprit of the community, and all further patches will be publicly available at my own github.

We're _STILL_ having problems with low number of duplicates, but not even close to 5-6%.

Once again, sorry.
If you want to, donate the "bounty" to ius which works hard to help us out with both bugs and improvements.

You talk well. But I do not understand anything. We must quit this business. In this I do not understand much.

[ius]

I have opened a pull request to revert the offending commits: https://github.com/jgarzik/pushpool/pull/24

[jgarzik]

Oh.  This is an old and well known problem.  Nothing to do with pushpool at all.

luke-jr worked around it in bitcoind with http://luke.dashjr.org/programs/bitcoin/w/bitcoind/luke-jr.git/shortlog/refs/heads/getwork_dedupe but there are other options for varying the coinbase as well.

[kjj]

Ugh.  Why wasn't it merged in a month ago?