This is due to a sext encoding bug present in current riak releases.
When sext and riak has been updated with a fix, revert this commit.
(cherry picked from commit b5a0dd9737)
The http client seems to have a lot of problems and works differently
than the PB client. Index values are not coming back on a get, even
though a PB get returns them, so indexing is not working properly.
Binaries are not encoded properly when the query index URL is created,
and empty binaries are taken to mean "no value".
(cherry picked from commit 1c29f7fb43)
General clean up and organization, with an eye on making this a
template for other EQC tests.
Preparing code to handle more than the PB client.
(cherry picked from commit 02b5e2128d)
It needs more love to match the functionality of the regular 2i tests,
but this is a good beginning. It can show a race in 2i pagination that
has been in the code for years. The race does not show up in run in a
single node with very few vnodes. I tried 32. Ideally we could use
PULSE to make scheduling in the riak nodes less deterministic and help
find this kind of issue, but currently PULSE is a single node joint.
This test uses a different bucket for each EQC command run. It is not
perfect isolation, but it is good enough and very fast compared to
taking down the entire cluster in between each run.
(cherry picked from commit fa49385837)
The http client seems to have a lot of problems and works differently
than the PB client. Index values are not coming back on a get, even
though a PB get returns them, so indexing is not working properly.
Binaries are not encoded properly when the query index URL is created,
and empty binaries are taken to mean "no value".
It needs more love to match the functionality of the regular 2i tests,
but this is a good beginning. It can show a race in 2i pagination that
has been in the code for years. The race does not show up in run in a
single node with very few vnodes. I tried 32. Ideally we could use
PULSE to make scheduling in the riak nodes less deterministic and help
find this kind of issue, but currently PULSE is a single node joint.
This test uses a different bucket for each EQC command run. It is not
perfect isolation, but it is good enough and very fast compared to
taking down the entire cluster in between each run.
