archive snowblossom slack general
prev 2018-07-06 next
2018-07-06 04:19:57
Fireduck
pool failover coming up
2018-07-06 08:15:22
Clueless
woo, can you have two pools work together with failover?
2018-07-06 08:15:49
Clueless
or, http://example.fun with multiple ips
2018-07-06 14:14:00
some
what if HA will be on the front of 2+ pools in a failover mode? [pool dns name pointed to HA ip] [HA] [pool1] [pool2] [pool-n]
2018-07-06 15:56:35
Fireduck
Bad idea with current setup. Grpc may reconnect randomly and the new connection might be to a system that hasn't heard of your work unit and rejects shares.
2018-07-06 16:31:49
Shoots
Is the client compatible with mac?
2018-07-06 16:54:30
Fireduck
yes
2018-07-06 19:33:33
Fireduck
INFO: Generating snow field at 4244.73 writes per second. Estimated total runtime is 491.86 hours. 0.03 complete.
2018-07-06 19:33:36
Fireduck
field 12 is a bear
2018-07-06 21:18:50
some
@Fireduck failover mode will switch to the next node only in case if the active one has died
2018-07-06 21:19:38
Fireduck
@some that is the plan
2018-07-06 21:20:03
Fireduck
more specifically, my plan was to have the miner maintain work units from all configured pools in memory
2018-07-06 21:20:13
Fireduck
and use the first pool from the list that has a current valid work unit
2018-07-06 21:20:33
Fireduck
that way the failover is very fast and we can report on pool being up or not even when we aren't using it
2018-07-06 21:22:05
some
I’ve meant HA in failover mode on the front of a bunch on nodes
2018-07-06 21:27:06
some
But from miner side looks more reliable
2018-07-06 21:29:50
Fireduck
ah, in the case of doing it on the server side, yeah, that would work
2018-07-06 21:30:03
Fireduck
there would be some rejects when it fails over but should be fine
2018-07-06 21:36:01
Clueless
The kicker for me is I can switch a data center ip to another internal box, but if the work doesn't carry over it may confuse miners on the pool.
2018-07-06 21:36:36
Clueless
Have to update /restart services once a week.
2018-07-06 21:38:19
Fireduck
yeah, it is way better for client side smarts to do this smoothly