Return 0 instead of doing another rt:sysread in repl_util:wait_for_reads/5

[jadeallenx]

Previously, wait_until_read function would execute
another rt:systest_read call and return a count of
errors, when we already asserted above that the list
is empty.

If the wait_until function does not return true, or
times out, or hits the retry limit, the running test
ought to fail anyway without needing an additional
read to fail.

So doing another call is redundant and unnecessary;
we just return 0 as the value if rt:wait_until
completes successfully.

[jadeallenx]

This PR is intended to address an issue we looked at while reviewing #931 - this particular function is called all over the repl tests, so some care ought to be used before merging.

[jadeallenx]

repl_aae_fullsync.erl
repl_aae_fullsync_bench.erl
repl_aae_fullsync_custom_n.erl
repl_rt_heartbeat.erl
repl_rt_overload.erl
repl_rt_pending.erl
repl_util.erl
replication2.erl
replication2_dirty.erl
rt_cascading.erl

These modules have calls to repl_util:wait_for_reads

[hazen]

./riak_test -c rtdev -t repl_aae_fullsync \
> -t repl_aae_fullsync_bench \
> -t repl_aae_fullsync_custom_n \
> -t repl_rt_heartbeat \
> -t repl_rt_overload \
> -t repl_rt_pending \
> -t repl_util \
> -t replication2 \
> -t replication2_dirty \
> -t rt_cascading

Test Results:
replication2-bitcask              : pass
repl_rt_pending-bitcask           : pass
repl_rt_overload-bitcask          : pass
repl_rt_heartbeat-bitcask         : pass
repl_aae_fullsync_custom_n-bitcask: pass
repl_aae_fullsync_bench-bitcask   : pass
repl_aae_fullsync-bitcask         : fail
rt_cascading-bitcask              : pass
replication2_dirty-bitcask        : pass
---------------------------------------------
1 Tests Failed
8 Tests Passed
That's 88.88888888888889% for those keeping score

[hazen]

It looks like repl_aae_fullsync fails quite a bit for unrelated reasons.

[jadeallenx]

@JeetKunDoug has been wrestling with this test too - I think its failures are unrelated to this particular change.

[hazen]

@mrallen1 I tested this and it looks good, however, I'm looking at the usage in the tests and they are testing a single node, not an array of nodes, so the assert we expected to happen earlier in https://github.com/basho/riak_test/blob/mra/repl-util-read-fix/src/rt.erl#L675 does not happen. So if there is an error, it now reports it as having no problems.

[jadeallenx]

@javajolt From my reading of rt:wait_until/2 we should still be ok, because we assert the return value of the function is ok on line 198. rt:wait_until will eventually return {fail, Result} if the Reads is not empty over the span of retry attempts.

[hazen]

Fair enough. The length will only be 0 if it succeeds anyway.
👍