source: misc/tools/acceptance_tests/test_ssh_to_compute_nodes.py @ 7656

#!/bin/env python

"""Testlet to check that can ssh to each subordinate node of the cluster."""

import os
import time
import test_utils as util

name = 'Test of compute-node accessibility via ssh'

def test(logfile):
    result = True

    (cluster, domain) = util.get_hostname()

    # get python to run
    python_env_var = os.getenv('PYTHON')

    # get cluster-specific information
    cluster_dict = util.get_cluster_info(cluster)
    if cluster_dict is None:
        util.log_print_nl(logfile, 'Sorry, you must be on one of these clusters:')
        for name in Cluster_Info:
            util.log_print_nl(logfile, '\t%s' % name)
        util.log_print_nl(logfile, 'You are on the %s machine.' % cluster)
        return False

    # get max width of the compute node names
    width = len(cluster_dict['node_stem']) + len(str(cluster_dict['num_nodes']))

    # get list of non-bad nodes
    test_nodes = util.get_node_numbers(cluster, strip_bad_nodes=True)

    # test each node
    for node_num in test_nodes:
        node_name = cluster_dict['node_stem'] % node_num
        util.log_print(logfile, "Testing: %s" % node_name.ljust(width+2))
        (_, fd) = os.popen4('ssh %s exit' % node_name)
        data = fd.read()
        status = fd.close()
        #if status is None:
        if len(data) == 0:
            util.log_print_nl(logfile, 'OK')
        else:
            util.log_print(logfile, data)
            result = False

    return result


if __name__ == '__main__':
    import sys

    logfile = 'test.log'
    if len(sys.argv) > 1:
        logfile = sys.argv[1]

    try:
        os.remove(logfile)
    except:
        pass

    if not test(logfile):
        sys.exit(10)

    sys.exit(0)