<    March 2017    >
Su Mo Tu We Th Fr Sa  
          1  2  3  4  
 5  6  7  8  9 10 11  
12 13 14 15 16 17 18  
19 20 21 22 23 24 25  
26 27 28 29 30 31
05:10 bturker joined
06:53 __alex joined
07:02 petro_ joined
07:11 bturker joined
08:28 bturker joined
08:55 smcvey joined
10:00 <russelldb> wouldn't it be great if riak_test was declarative? I have a really simple eqc counter example from a simple model/simulation of riak, and I want to try and write it as a test
10:00 smcvey_ joined
10:00 <russelldb> it involves some partitions and a disk error, but even using the rt:partition and rt_intercepts code it is very hard
10:01 <russelldb> I feel like there must be a better way with riak-test, especially for partitions (changing the cookie doesn't seem right)
10:58 tombriden joined
11:16 bturker joined
11:20 petro_ joined
12:24 andyt1 joined
13:11 nfo joined
13:20 andyt1 joined
13:24 al-damiri joined
13:38 bsparrow joined
14:08 nfo joined
14:22 <russelldb> what can stop a node from handing off?
14:23 <russelldb> I can see some data for key K is on node 3, and it was written to 3 as a fallback, but it _never_ gets handed off. n_val=2, one node is down, 4 nodes are up, calls to riak_core_vnode_manager:force_handoffs/0 do nothing, riak_core_status:transfers/0 shows everyone "waiting to handoff"
14:24 <russelldb> calls to riak_core_handoff_manager:status/0 on the fallback node in question (3) always returns []
14:24 <russelldb> so…how can I get 3 to handoff?
14:24 <russelldb> all the primaries for this key are up
14:31 <russelldb> handoff_concurrency is set to 100
14:32 <russelldb> so far the only thing that works is killing a primary, then the fallback is used in the preflist for a read, and read repair gets the data from the fallback to the primary…but shouldn't handoff do that?
14:49 <mrallen1> I didn't know fallback data gets migrated in handoff
14:49 <mrallen1> I guess the handoff fold doesn't care if the key is primary or fallback
14:49 <mrallen1> So you'd think it should
14:58 <russelldb> mrallen1: I just mean there is data on the fallback that it never getting handed off to the primary that should own it
14:58 <mrallen1> OK
14:59 <russelldb> so hinted, not ownership, handoff
14:59 <mrallen1> I thought that was one of the reasons for read repair / AAE syncs?
15:00 <russelldb> AAE is only for primaries
15:00 <russelldb> and if all your primaries are up, then read repair won't do it, as the fallback node will never be in the read preflist
15:00 <russelldb> so…that is exactly what hinted handoff is for
15:01 <russelldb> but in this instance, it never happens, I wonder if it is weird n=2 edge thing…or…I have no idea. It just never happens
15:07 <russelldb> tbqf, this is just to get a riak test for a bug I've fixed, and already I spent 3 days fighting riak/riak_test, so i think I will just submit a patch without and test, and when the reviewer says "where is the test" I'll have a ready open can'o'worms for them
15:28 <mrallen1> haha
15:50 smcvey_ joined
16:01 <russelldb> holy s*** I finally got a failing case. OK. Now I can make a PR
16:39 jplock joined
16:40 cemerick joined
17:46 <mrallen1> russelldb++
19:12 kleptocroc- joined
20:33 irclogger_com joined
20:33 Topic for
20:35 bturker joined