CSTAR-630: Support case where client and server users are not the same

[csplinter]

Sometimes the cstar_perf tool host is not using the same linux user as the server under test host. This causes the DSE bootstrap process to fail because it tries to create the cstar_perf tool host home on the server under test host.

[automaton@10.200.175.89] out: [10.200.241.2] run: rm -rf /home/automaton/fab/dse_builds/dse-5.0.0
[automaton@10.200.175.89] out: [10.200.241.2] run: mkdir -p /home/automaton/fab/dse_builds
[automaton@10.200.175.89] out: [10.200.241.2] out: mkdir: cannot create directory ‘/home/automaton’: Permission denied
[automaton@10.200.175.89] out: [10.200.241.2] out: 
[automaton@10.200.175.89] out: 
[automaton@10.200.175.89] out: 
[automaton@10.200.175.89] out: Fatal error: run() received nonzero return code 1 while executing!
[automaton@10.200.175.89] out: 
[automaton@10.200.175.89] out: Requested: mkdir -p /home/automaton/fab/dse_builds
[automaton@10.200.175.89] out: Executed: /bin/bash -l -c "mkdir -p /home/automaton/fab/dse_builds"

In order to solve this, we only use os.path.expanduser to get the dse_build_local location. This PR uses os.path.join('~','fab','dse') as the dse_build_remote location. This allows the path to properly expand when the given commands are run on the server under test host.

Tested:

• cstar_docker setup
• jenkins@blade server under test host + automaton@openstack cstar_perf tool host

[nastra]

There are certain cases where os.path.join('~','fab','dse') would't work. One case that I can think of is if such a path is passed to the JVM as a parameter. What if we instead use the $HOME variable?

[csplinter]

Using $HOME fails here

[testcluster-01] run: ln -sfn ~/fab/dse_builds/dse-5.0.0 $HOME/fab/dse
[testcluster-01] download: <file obj> <- /home/cstar/$HOME/fab/dse/resources/cassandra/conf/cassandra.yaml

Fatal error: get() encountered an exception while downloading '/home/cstar/$HOME/fab/dse/resources/cassandra/conf/cassandra.yaml'

Underlying exception:
    No such file

Aborting.

This issue with using environment variables in the path appears to be specific to using the fab.get and fab.put commands

[csplinter]

I chose to add a local option to our get_dse_path and get_cassandra_path

fab_cassandra.py

def get_cassandra_path(local=True):
    return os.path.expanduser('~/fab/cassandra') if local else os.path.join(fab.run('pwd'), 'fab', 'cassandra')

def get_bin_path(local=True):
    return os.path.join(get_cassandra_path(local=local), 'bin')

fab_dse.py

def get_dse_path(local=True):
    return os.path.expanduser("~/fab/dse") if local else os.path.join(fab.run('pwd'), 'fab', 'dse')

def get_dse_conf_path(local=True):
    return os.path.join(get_dse_path(local=local), 'resources', 'dse', 'conf')

def get_cassandra_path(local=True):
    return os.path.join(get_dse_path(local=local), 'resources', 'cassandra')

[csplinter]

The behavior here with the fabric.operations.get function

download: <file obj> <- /home/cstar/$HOME/fab/dse/resources/cassandra/conf/cassandra.yaml

Is caused by the fact that they are joining home = ftp.normalize('.') with the remote_path. In our case, the home is /home/cstar and the remote path is $HOME/fab/dse/resources/cassandra/conf/cassandra.yaml. See the fabric source, we hit that case because the fabric code is first checking for os.path.isabs(remote_path) and on Unix, that means it begins with a slash. I do not see a way to work around this using $HOME in get_dse_path without modifying this path in each call to fab.get or fab.put.

Regarding the first solution of using os.path.join('~','fab','dse') and the case of using such a path as a JVM parameter, could we do something similar to what we do here? At this point I am having trouble finding an easier solution

[nastra]

Regarding the first solution of using os.path.join('~','fab','dse') and the case of using such a path as a JVM parameter, could we do something similar to what we do here?

The approach from https://github.com/datastax/cstar_perf/blob/master/tool/cstar_perf/tool/fab_dse.py/#L190-L191 sounds like a reasonable workaround

[csplinter]

@nastra this is ready for another review

• get_dse_path changed to use os.path.join('$HOME','fab','dse')
• fab.get and fab.put calls changed to use .replace('$HOME','~') if using get_dse_path. This includes calls to product.get_cassandra_path which calls get_dse_path when the product = dse
• See this comment for explanation of why we have to .replace('$HOME','~') for the fab.get and fab.put calls
• I did not change fab_cassandra.get_cassandra_path as I do not anticipate the Tech Cert team will test open source Cassandra and this fix is specific to our scenario. If it is requested later, we can always add it to the Cassandra code as well.
• I tested w/cstar_docker and w/client + server that have different users

[nastra]

+1