Closed beerriot closed 10 years ago
Did you decrease the vnode_management_timer? See verify_dt_converge for an example.
Sean Cribbs
On Oct 8, 2013, at 6:52 AM, Bryan Fink notifications@github.com wrote:
The verify_handoff_mixed test from basho/riak_test#400 should be added to all 2.0 (and future) runs.
It operates somewhat like loaded_upgrade, in that it also uses either a previous or a legacy version node, according to the upgrade_version metadata. It would be great to have the test run twice, once with upgrade_version=previous and once =legacy.
I'll warn that the test often takes four minutes to run successfully, and will take over five minutes to fail. It has to wait for handoff to finish, and that can take a while. Raising handoff_concurrency to 8 did not reduce the success time.
— Reply to this email directly or view it on GitHub.
Just tried it. Unfortunately, it increases the likelihood that handoff will happen before the cluster has downgraded the fold req version, preventing the test from triggering the behavior it's intended to test.
Oh, hey - vnode_management_timer
can be changed on the fly. I've just submitted a patch to lower it after we've confirmed that the fold req version has been renegotiated: basho/riak_test#409
The
verify_handoff_mixed
test from basho/riak_test#400 should be added to all 2.0 (and future) runs.It operates somewhat like
loaded_upgrade
, in that it also uses either aprevious
or alegacy
version node, according to theupgrade_version
metadata. It would be great to have the test run twice, once withupgrade_version=previous
and once=legacy
.I'll warn that the test often takes four minutes to run successfully, and will take over five minutes to fail. It has to wait for handoff to finish, and that can take a while. Raising
handoff_concurrency
to 8 did not reduce the success time.