def check_all(interactive):
all_success = NO_ERROR
# 1: Install
checked_install = check_installation(interactive)
if (is_error(checked_install)):
return checked_install
else:
all_success = checked_install
print(
"================================================================================"
)
# 2: Connection
checked_connection = check_connection(interactive)
if (is_error(checked_connection)):
return checked_connection
else:
all_success = checked_connection
print(
"================================================================================"
)
# 3: Heartbeat
checked_hb = check_heartbeat(interactive)
if (is_error(checked_hb)):
return checked_hb
else:
all_success = checked_hb
print(
"================================================================================"
)
checked_highcpumem = check_high_cpu_memory(interactive)
if (is_error(checked_highcpumem)):
return checked_highcpumem
else:
all_success = checked_highcpumem
print(
"================================================================================"
)
checked_syslog = check_syslog(interactive)
if (is_error(checked_syslog)):
return checked_syslog
else:
all_success = checked_syslog
print(
"================================================================================"
)
checked_cl = check_custom_logs(interactive)
if (is_error(checked_cl)):
return checked_cl
else:
all_success = checked_cl
return all_success
def check_syslog(interactive, prev_success=NO_ERROR):
print("CHECKING FOR SYSLOG ISSUES...")
success = prev_success
# check if installed / connected / running correctly
print("Checking if omsagent installed and running...")
# check installation
if (get_oms_version() == None):
print_errors(ERR_OMS_INSTALL)
print(
"Running the installation part of the troubleshooter in order to find the issue..."
)
print(
"================================================================================"
)
return check_installation(interactive,
err_codes=False,
prev_success=ERR_FOUND)
# check connection
checked_la_endpts = check_log_analytics_endpts()
if (checked_la_endpts != NO_ERROR):
print_errors(checked_la_endpts)
print(
"Running the connection part of the troubleshooter in order to find the issue..."
)
print(
"================================================================================"
)
return check_connection(interactive,
err_codes=False,
prev_success=ERR_FOUND)
# check running
workspace_id = geninfo_lookup('WORKSPACE_ID')
if (workspace_id == None):
error_info.append(('Workspace ID', OMSADMIN_PATH))
return ERR_INFO_MISSING
checked_omsagent_running = check_omsagent_running(workspace_id)
if (checked_omsagent_running != NO_ERROR):
print_errors(checked_omsagent_running)
print(
"Running the general health part of the troubleshooter in order to find the issue..."
)
print(
"================================================================================"
)
return check_heartbeat(interactive, prev_success=ERR_FOUND)
# check for service controller
print("Checking if machine has a valid service controller...")
checked_sc = check_service_controller()
if (is_error(checked_sc)):
return checked_sc
else:
success = print_errors(checked_sc)
# check rsyslog / syslogng running
print("Checking if machine has rsyslog or syslog-ng running...")
checked_services = check_services()
if (is_error(checked_services)):
return print_errors(checked_services)
else:
success = print_errors(checked_services)
# check for syslog.conf and syslog destination file
print("Checking for syslog configuration files...")
checked_conf_files = check_conf_files()
if (is_error(checked_conf_files)):
if (checked_conf_files in [ERR_OMS_INSTALL, ERR_FILE_MISSING]):
print_errors(checked_conf_files)
print(
"Running the installation part of the troubleshooter in order to find the issue..."
)
print(
"================================================================================"
)
return check_installation(interactive,
err_codes=False,
prev_success=ERR_FOUND)
else:
return print_errors(checked_conf_files)
else:
success = print_errors(checked_conf_files)
return success
def check_custom_logs(interactive, prev_success=NO_ERROR):
if (interactive):
print(" To check if you are using custom logs, please go to https://ms.portal.azure.com\n"\
" and navigate to your workspace. Once there, please navigate to the 'Advanced\n"\
" settings' blade, and then go to 'Data' > 'Custom Logs'. There you should be\n"\
" to see any custom logs you may have.\n")
using_cl = get_input("Are you currently using custom logs? (y/n)",\
(lambda x : x.lower() in ['y','yes','n','no']),\
"Please type either 'y'/'yes' or 'n'/'no' to proceed.")
# not using custom logs
if (using_cl in ['n', 'no']):
print("Continuing on with the rest of the troubleshooter...")
print(
"================================================================================"
)
return prev_success
# using custom logs
else:
print("Continuing on with troubleshooter...")
print(
"--------------------------------------------------------------------------------"
)
print("CHECKING FOR CUSTOM LOG ISSUES...")
success = prev_success
# check if installed / connected / running correctly
print("Checking if omsagent installed and running...")
# check installation
if (get_oms_version() == None):
print_errors(ERR_OMS_INSTALL)
print(
"Running the installation part of the troubleshooter in order to find the issue..."
)
print(
"================================================================================"
)
return check_installation(interactive,
err_codes=False,
prev_success=ERR_FOUND)
# check connection
checked_la_endpts = check_log_analytics_endpts()
if (checked_la_endpts != NO_ERROR):
print_errors(checked_la_endpts)
print(
"Running the connection part of the troubleshooter in order to find the issue..."
)
print(
"================================================================================"
)
return check_connection(interactive,
err_codes=False,
prev_success=ERR_FOUND)
# check running
workspace_id = geninfo_lookup('WORKSPACE_ID')
if (workspace_id == None):
error_info.append(('Workspace ID', OMSADMIN_PATH))
return ERR_INFO_MISSING
checked_omsagent_running = check_omsagent_running(workspace_id)
if (checked_omsagent_running != NO_ERROR):
print_errors(checked_omsagent_running)
print(
"Running the general health part of the troubleshooter in order to find the issue..."
)
print(
"================================================================================"
)
return check_heartbeat(interactive, prev_success=ERR_FOUND)
# check customlog.conf
print("Checking for custom log configuration files...")
checked_clconf = check_customlog_conf(interactive)
if (is_error(checked_clconf)):
return print_errors(checked_clconf)
else:
success = print_errors(checked_clconf)
return success
def check_high_cpu_memory(interactive, prev_success=NO_ERROR):
print("CHECKING FOR HIGH CPU / MEMORY USAGE...")
success = prev_success
# check if installed / connected / running correctly
print("Checking if omsagent installed and running...")
# check installation
if (get_oms_version() == None):
print_errors(ERR_OMS_INSTALL)
print("Running the installation part of the troubleshooter in order to find the issue...")
print("================================================================================")
return check_installation(interactive, err_codes=False, prev_success=ERR_FOUND)
# check connection
checked_la_endpts = check_log_analytics_endpts()
if (checked_la_endpts != NO_ERROR):
print_errors(checked_la_endpts)
print("Running the connection part of the troubleshooter in order to find the issue...")
print("================================================================================")
return check_connection(interactive, err_codes=False, prev_success=ERR_FOUND)
# check running
workspace_id = geninfo_lookup('WORKSPACE_ID')
if (workspace_id == None):
error_info.append(('Workspace ID', OMSADMIN_PATH))
return ERR_INFO_MISSING
checked_omsagent_running = check_omsagent_running(workspace_id)
if (checked_omsagent_running != NO_ERROR):
print_errors(checked_omsagent_running)
print("Running the general health part of the troubleshooter in order to find the issue...")
print("================================================================================")
return check_heartbeat(interactive, prev_success=ERR_FOUND)
# TODO: decide if should keep this in or not
# check disk space
# print("Checking recent modifications to largest files...")
# checked_disk_space = check_disk_space()
# if (checked_disk_space != NO_ERROR):
# return print_errors(checked_disk_space)
# check log rotation
print("Checking if log rotation is working correctly...")
checked_logrot = check_log_rotation()
if (is_error(checked_logrot)):
return print_errors(checked_logrot)
else:
success = print_errors(checked_logrot)
# check CPU capacity
print("Checking if OMI is at 100% CPU (may take some time)...")
checked_highcpu = check_omi_cpu()
if (is_error(checked_highcpu)):
return print_errors(checked_highcpu)
else:
success = print_errors(checked_highcpu)
# check slab memory / dentry cache issue
print("Checking slab memory / dentry cache usage...")
checked_slabmem = check_slab_memory()
if (is_error(checked_slabmem)):
return print_errors(checked_slabmem)
else:
success = checked_slabmem
return success