There seems to be a hard limit of maximum number of workers for a given address. Of course, that's not something most people will ever see, but I bumped into it without noticing a few months ago, and I have been mining without having shares fully rewarded.
The symptoms were : approx. 25 Th/s worth of hash which do appeared on the 256 seconds stats were never present on the 3 hours stats, and a number of workers did not appear on the stats list, although they were hashing happily.
I got confirmation by Luke Jr. privately that there is indeed a hard limit, set to 256 workers (although I have more than that in the list, most being defunct at the moment).
The immediate workaround is to reuse old worker names for new miners, of course it makes it necessary to maintain some kind of indirection table for those names on my end, but it can still do it up to, if I understand, 256 workers.
I plan further extension in the middle term, so I wonder if it would be possible to raise that artificial limit to, say, 1024 workers, without being too heavy on the pool infrastructure ?
There is an exploit/DoS that was happening vs the stats servers re: many workers on an address. The reward system (which produces the 256/128 second stats) is immune since it doesn't care about workers. But the stats currently are capped.
Maybe it is possible to decrease the timeout to which unused names fall off the list?
I do find they stay there too long. A few days is enough. I have not hit this limit either and I have over 50Th, so I admit it is rare, but why keep old names around so long?
The way the stats are pulling data based on the full share database it's hard to actually timeout worker names at all without expensive (processing wise) code. A stats overhaul (in the works) should help with this eventually.
Did something change with vardiff? My difficulty keeps alternating between 512 and 1024 and my hashrate keeps yo-yoing as a result. I know the hashrate probably shouldn't yo-yo from that but it totally is. It happens on my S4 and it happens on another address where I have 2 S5's.
EDIT: Actually I am assuming (perhaps incorrectly) that the alternating difficulty is causing the yo-yo on the hashrate graph. My S4 is alternating between 1024 and 2048 and has always exhibited more hashrate variation than the two S5's but in the past few days they are both swinging up and back in a rather wider arc than I usually see.
The only change to difficulty was bumping the minimum to 512. This shouldn't effect you if you're over 500Gh or so. Alternating between difficulties is normal when a device is on the border between vardiff levels. I'm sure your 12 hour hashrate is fine.
