This question is probably better suited for the user@ group. It doesn't sound to me like you've uncovered a bug, but rather you're engaging in highly contentious paxos, which is rarely going to have a favorable outcome. Likely you're overwhelming your cluster (or more specifically the replicas for your single partition - in that sense for your test as designed cluster size would be basically irrelevant).
It also sounds like you're building something like a queue, which is a significant classic anti-pattern in the Cassandra realm; there are much better purpose built tools for this. See http://www.datastax.com/dev/blog/cassandra-anti-patterns-queues-and-queue-like-datasets for more information on why queue-like datasets aren't a good fit for Cassandra. On Fri, Jan 20, 2017 at 5:25 AM Jaroslav KamenĂk <jaros...@kamenik.cz> wrote: > Hi, > > I would like to ask here before posting new bug. I am trying to make a > simple system > for distribution preallocated tickets between concurrent clients using C* > LWTs. > It is simply one partition containing tickets for one domain, client reads > the first one > and tries to delete it conditionally, success = it owns it, fail = try > again.. > > It works well, but it starts to fail with WTEs under load. So I tried to > make simple > test with 16 concurrent threads competing for one row with 1000 cols. It > was running > on cluster with 5 C* 3.0.9 with default configuration, replication factor > 3. > > Surprisingly, it failed immediately after few requests. It takes longer > time with > less threads, but even 2 clients are enough to crash it. > > I am wondering, if it Is problem in Cassandra or normal behaviour or bad > use of LWT? > > Thanks, > > Jaroslav >