Update test server node generation for riak 1.4

[randysecrist]

In addition to making the test server work with riak 1.4; this PR makes the following high level changes:

• Simplify the generation by coping everything except the data directory.
• Remove support for a symlinked basedir.
• Depend less on the control script.
• Add support for the env script.
• Move protobuf config to riak_api for 1.4 only
• Make backward compatible with riak 1.3.

[dalibor]

I'm running Ubuntu 12.10 and getting the following error when running specs:

#<Errno::ENOENT: No such file or directory - /usr/releases/start_erl.data>

I've installed riak with apt-get and the location of start_erl.data file is /usr/lib/riak/releases/start_erl.data.

So, maybe this should be config or ENV variable?

[randysecrist]

will carve out some time shortly to try and reproduce & resolve this; locating this file (and other things) this is part of what the test server was trying to accomplish with all the symlink checking & poking around the riak scripts; something I'm hoping to move away from.

[seancribbs]

Why not optionally read the lib/env.sh instead of assuming the basedir will always be the parent of the source?

[randysecrist]

I don't see a reason to read config files for base_dir if we require test code to specify a source. I think we can continue working through the ubuntu differences ; except for the following issue:

TL/DR: I think we may have a few packaging issues that need to be fixed before we can support ubuntu.

Details:

Under ubuntu (upon fixing the start_erl) issue; file permissions of a few things in the lib directory are making it problematic to recuse copy (using FileUtils.cp_r) copy this directory as a regular user. (lib/app_epath.sh) - also (erts-5.9.1/bin/nodetool).

In the case of the erts dir; we can create a local symlink and should be fine; however we can't use that strategy for the lib dir since we have to rewrite env.sh. Without moving the lib dir; the test server can't rewrite the lib/env.sh file which is now needed to set variables for 1.4 to start.

I could not use FileUtils.cp_r and just move parts of lib individually; but my real question is - why are there .sh files in lib at all? If we have to have them; why are the permissions set to 700 instead of 755 like most of the other files in those directories.

[bkerley]

This is working for me on Mac OS X.

[seancribbs]

Why was this changed to the inverse? The point of "drop" was to wipe data but not modify the cluster, e.g. a development cluster for your specific app that occasionally needs to have its data wiped. You don't want to have to rejoin that cluster when you start it back up.

[randysecrist]

I wager that most users of test server are not setting up and retaining a cluster > size 1 as part of a test suite. That is what vagrant, chef, ansible, and other cluster provisioning tools are for. Interestingly; it seems as if the ring gets recreated w/ single node defaults if it is destroyed by drop & test suites I use haven't been negatively impacted by this.

I don't have a strong opinion on this matter; I'm more concerned about the need to support various source directory structures based upon frequently changing platform packaging. Willing to revert this change.

[bkerley]

So I did find a good reason to make test-client work on Ubuntu: travis.

[seancribbs]

Well... we can also refactor the suite so it uses random buckets/keys. But that doesn't fix the test server for other people.