Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
erp5
Project overview
Project overview
Details
Activity
Releases
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Labels
Merge Requests
138
Merge Requests
138
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Analytics
Analytics
CI / CD
Repository
Value Stream
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Jobs
Commits
Open sidebar
nexedi
erp5
Commits
199e7247
Commit
199e7247
authored
Dec 26, 2017
by
Julien Muchembled
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
testnode: simplify logging
parent
c39caeaa
Changes
9
Hide whitespace changes
Inline
Side-by-side
Showing
9 changed files
with
263 additions
and
279 deletions
+263
-279
erp5/tests/testERP5TestNode.py
erp5/tests/testERP5TestNode.py
+34
-28
erp5/util/testnode/ProcessManager.py
erp5/util/testnode/ProcessManager.py
+40
-37
erp5/util/testnode/ScalabilityTestRunner.py
erp5/util/testnode/ScalabilityTestRunner.py
+52
-47
erp5/util/testnode/SlapOSControler.py
erp5/util/testnode/SlapOSControler.py
+20
-22
erp5/util/testnode/SlapOSMasterCommunicator.py
erp5/util/testnode/SlapOSMasterCommunicator.py
+30
-31
erp5/util/testnode/UnitTestRunner.py
erp5/util/testnode/UnitTestRunner.py
+11
-11
erp5/util/testnode/Updater.py
erp5/util/testnode/Updater.py
+7
-8
erp5/util/testnode/__init__.py
erp5/util/testnode/__init__.py
+18
-21
erp5/util/testnode/testnode.py
erp5/util/testnode/testnode.py
+51
-74
No files found.
erp5/tests/testERP5TestNode.py
View file @
199e7247
import
unittest
import
unittest
from
unittest
import
TestCase
from
unittest
import
TestCase
from
contextlib
import
contextmanager
from
erp5.util.testnode
import
logger
from
erp5.util.testnode.testnode
import
TestNode
,
test_type_registry
from
erp5.util.testnode.testnode
import
TestNode
,
test_type_registry
from
erp5.util.testnode.NodeTestSuite
import
SlapOSInstance
,
NodeTestSuite
from
erp5.util.testnode.NodeTestSuite
import
SlapOSInstance
,
NodeTestSuite
from
erp5.util.testnode.ProcessManager
import
ProcessManager
,
SubprocessError
from
erp5.util.testnode.ProcessManager
import
ProcessManager
,
SubprocessError
...
@@ -13,17 +15,26 @@ from erp5.util.testnode.SlapOSControler import createFolder
...
@@ -13,17 +15,26 @@ from erp5.util.testnode.SlapOSControler import createFolder
from
erp5.util.taskdistribution
import
TaskDistributor
from
erp5.util.taskdistribution
import
TaskDistributor
from
erp5.util.taskdistribution
import
TestResultProxy
from
erp5.util.taskdistribution
import
TestResultProxy
import
argparse
import
argparse
import
logging
import
os
import
os
import
shutil
import
shutil
import
subprocess
import
subprocess
import
sys
import
tempfile
import
tempfile
import
json
import
json
import
time
import
time
import
types
import
types
import
re
import
re
@
contextmanager
def
dummySuiteLog
(
_
):
yield
class
ERP5TestNode
(
TestCase
):
class
ERP5TestNode
(
TestCase
):
_handler
=
logging
.
StreamHandler
(
sys
.
stdout
)
_handler
.
setFormatter
(
logging
.
Formatter
(
'TESTNODE LOG: %(message)s'
))
def
setUp
(
self
):
def
setUp
(
self
):
self
.
_temp_dir
=
tempfile
.
mkdtemp
()
self
.
_temp_dir
=
tempfile
.
mkdtemp
()
self
.
working_directory
=
os
.
path
.
join
(
self
.
_temp_dir
,
'testnode'
)
self
.
working_directory
=
os
.
path
.
join
(
self
.
_temp_dir
,
'testnode'
)
...
@@ -49,13 +60,11 @@ class ERP5TestNode(TestCase):
...
@@ -49,13 +60,11 @@ class ERP5TestNode(TestCase):
os
.
mkdir
(
self
.
remote_repository0
)
os
.
mkdir
(
self
.
remote_repository0
)
os
.
mkdir
(
self
.
remote_repository1
)
os
.
mkdir
(
self
.
remote_repository1
)
os
.
mkdir
(
self
.
remote_repository2
)
os
.
mkdir
(
self
.
remote_repository2
)
def
log
(
*
args
,
**
kw
):
logging
.
getLogger
().
addHandler
(
self
.
_handler
)
for
arg
in
args
:
print
"TESTNODE LOG : %r, %r"
%
(
arg
,
kw
)
self
.
log
=
log
def
tearDown
(
self
):
def
tearDown
(
self
):
shutil
.
rmtree
(
self
.
_temp_dir
,
True
)
shutil
.
rmtree
(
self
.
_temp_dir
,
True
)
logging
.
getLogger
().
removeHandler
(
self
.
_handler
)
def
getTestNode
(
self
):
def
getTestNode
(
self
):
# XXX how to get property the git path ?
# XXX how to get property the git path ?
...
@@ -80,7 +89,12 @@ class ERP5TestNode(TestCase):
...
@@ -80,7 +89,12 @@ class ERP5TestNode(TestCase):
config
[
"frontend_url"
]
=
"http://frontend/"
config
[
"frontend_url"
]
=
"http://frontend/"
config
[
"software_list"
]
=
[
"foo"
,
"bar"
]
config
[
"software_list"
]
=
[
"foo"
,
"bar"
]
return
TestNode
(
self
.
log
,
config
)
testnode
=
TestNode
(
config
)
# By default, keep suite logs to stdout for easier debugging
# (stdout/stderr are automatically reported to ERP5).
# This is unset by test methods that check normal suite logging.
testnode
.
suiteLog
=
dummySuiteLog
return
testnode
def
getTestSuiteData
(
self
,
add_third_repository
=
False
,
def
getTestSuiteData
(
self
,
add_third_repository
=
False
,
add_broken_repository
=
False
,
reference
=
"foo"
):
add_broken_repository
=
False
,
reference
=
"foo"
):
...
@@ -634,13 +648,10 @@ shared = true
...
@@ -634,13 +648,10 @@ shared = true
allow_restart
=
False
,
test_title
=
None
,
project_title
=
None
):
allow_restart
=
False
,
test_title
=
None
,
project_title
=
None
):
global
counter
global
counter
# return no test to check if run method will run the next test suite
# return no test to check if run method will run the next test suite
if
counter
==
3
and
project_title
!=
'qux'
:
if
counter
!=
3
or
project_title
==
'qux'
:
result
=
None
else
:
test_result_path
=
os
.
path
.
join
(
test_result_path_root
,
test_title
)
test_result_path
=
os
.
path
.
join
(
test_result_path_root
,
test_title
)
result
=
TestResultProxy
(
self
.
_proxy
,
self
.
_retry_time
,
return
TestResultProxy
(
self
.
_proxy
,
self
.
_retry_time
,
self
.
_logger
,
test_result_path
,
node_title
,
revision
)
logger
,
test_result_path
,
node_title
,
revision
)
return
result
def
patch_runTestSuite
(
self
,
*
argv
,
**
kw
):
def
patch_runTestSuite
(
self
,
*
argv
,
**
kw
):
return
{
'status_code'
:
0
}
return
{
'status_code'
:
0
}
original_sleep
=
time
.
sleep
original_sleep
=
time
.
sleep
...
@@ -707,7 +718,7 @@ shared = true
...
@@ -707,7 +718,7 @@ shared = true
def
_checkCorrectStatus
(
expected_status
,
*
args
):
def
_checkCorrectStatus
(
expected_status
,
*
args
):
result
=
process_manager
.
spawn
(
*
args
)
result
=
process_manager
.
spawn
(
*
args
)
self
.
assertEqual
(
result
[
'status_code'
],
expected_status
)
self
.
assertEqual
(
result
[
'status_code'
],
expected_status
)
process_manager
=
ProcessManager
(
log
=
self
.
log
,
max_timeout
=
1
)
process_manager
=
ProcessManager
(
max_timeout
=
1
)
_checkCorrectStatus
(
0
,
*
[
'sleep'
,
'0'
])
_checkCorrectStatus
(
0
,
*
[
'sleep'
,
'0'
])
# We must make sure that if the command is too long that
# We must make sure that if the command is too long that
# it will be automatically killed
# it will be automatically killed
...
@@ -715,8 +726,7 @@ shared = true
...
@@ -715,8 +726,7 @@ shared = true
def
test_13_SlaposControlerResetSoftware
(
self
):
def
test_13_SlaposControlerResetSoftware
(
self
):
test_node
=
self
.
getTestNode
()
test_node
=
self
.
getTestNode
()
controler
=
SlapOSControler
(
self
.
working_directory
,
controler
=
SlapOSControler
(
self
.
working_directory
,
test_node
.
config
)
test_node
.
config
,
self
.
log
)
os
.
mkdir
(
controler
.
software_root
)
os
.
mkdir
(
controler
.
software_root
)
file_name
=
'AC_Ra
\
xc3
\
xad
zertic
\
xc3
\
xa1
ma'
file_name
=
'AC_Ra
\
xc3
\
xad
zertic
\
xc3
\
xa1
ma'
non_ascii_file
=
open
(
os
.
path
.
join
(
controler
.
software_root
,
file_name
),
'w'
)
non_ascii_file
=
open
(
os
.
path
.
join
(
controler
.
software_root
,
file_name
),
'w'
)
...
@@ -778,30 +788,26 @@ shared = true
...
@@ -778,30 +788,26 @@ shared = true
def
patch_createTestResult
(
self
,
revision
,
test_name_list
,
node_title
,
def
patch_createTestResult
(
self
,
revision
,
test_name_list
,
node_title
,
allow_restart
=
False
,
test_title
=
None
,
project_title
=
None
):
allow_restart
=
False
,
test_title
=
None
,
project_title
=
None
):
test_result_path
=
os
.
path
.
join
(
test_result_path_root
,
test_title
)
test_result_path
=
os
.
path
.
join
(
test_result_path_root
,
test_title
)
result
=
TestResultProxy
(
self
.
_proxy
,
self
.
_retry_time
,
return
TestResultProxy
(
self
.
_proxy
,
self
.
_retry_time
,
self
.
_logger
,
test_result_path
,
node_title
,
revision
)
logger
,
test_result_path
,
node_title
,
revision
)
return
result
def
patch_runTestSuite
(
self
,
*
argv
,
**
kw
):
def
patch_runTestSuite
(
self
,
*
argv
,
**
kw
):
return
{
'status_code'
:
0
}
return
{
'status_code'
:
0
}
def
checkTestSuite
(
test_node
):
def
checkTestSuite
(
test_node
):
test_node
.
node_test_suite_dict
test_node
.
node_test_suite_dict
rand_part_set
=
set
()
rand_part_set
=
set
()
self
.
assertEquals
(
2
,
len
(
test_node
.
node_test_suite_dict
))
self
.
assertEquals
(
2
,
len
(
test_node
.
node_test_suite_dict
))
self
.
assertIsNot
(
test_node
.
suite_log
,
None
)
self
.
assertTrue
(
isinstance
(
test_node
.
suite_log
,
types
.
MethodType
))
for
ref
,
suite
in
test_node
.
node_test_suite_dict
.
items
():
for
ref
,
suite
in
test_node
.
node_test_suite_dict
.
items
():
self
.
assertTrue
(
'var/log/testnode/%s'
%
suite
.
reference
in
\
self
.
assertTrue
(
'var/log/testnode/%s'
%
suite
.
reference
in
\
suite
.
suite_log_path
,
suite
.
suite_log_path
,
"Incorrect suite log path : %r"
%
suite
.
suite_log_path
)
"Incorrect suite log path : %r"
%
suite
.
suite_log_path
)
self
.
assertTrue
(
suite
.
suite_log_path
.
endswith
(
'suite.log'
))
m
=
re
.
search
(
'-(.*)/suite.log$'
,
suite
.
suite_log_path
)
m
=
re
.
match
(
'.*
\
-(.*)
\
/suite.log'
,
suite
.
suite_log_path
)
rand_part
=
m
.
groups
()[
0
]
rand_part
=
m
.
groups
()[
0
]
self
.
assertEqual
(
len
(
rand_part
),
10
)
self
.
assertEqual
(
len
(
rand_part
),
10
)
self
.
assertNotIn
(
rand_part
,
rand_part_set
)
self
.
assertNotIn
(
rand_part
,
rand_part_set
)
rand_part_set
.
add
(
rand_part
)
rand_part_set
.
add
(
rand_part
)
suite_log
=
open
(
suite
.
suite_log_path
,
'r'
)
with
open
(
suite
.
suite_log_path
)
as
suite_log
:
self
.
assertEquals
(
1
,
len
([
x
for
x
in
suite_log
.
readlines
()
\
self
.
assertIn
(
"Getting configuration from test suite"
,
if
x
.
find
(
"Activated logfile"
)
>=
0
]
))
suite_log
.
readline
(
))
RunnerClass
=
test_type_registry
[
my_test_type
]
RunnerClass
=
test_type_registry
[
my_test_type
]
original_sleep
=
time
.
sleep
original_sleep
=
time
.
sleep
...
@@ -837,6 +843,7 @@ shared = true
...
@@ -837,6 +843,7 @@ shared = true
original_createTestResult
=
TaskDistributor
.
createTestResult
original_createTestResult
=
TaskDistributor
.
createTestResult
TaskDistributor
.
createTestResult
=
patch_createTestResult
TaskDistributor
.
createTestResult
=
patch_createTestResult
test_node
=
self
.
getTestNode
()
test_node
=
self
.
getTestNode
()
del
test_node
.
suiteLog
# Change UnitTestRunner class methods
# Change UnitTestRunner class methods
original_prepareSlapOS
=
RunnerClass
.
_prepareSlapOS
original_prepareSlapOS
=
RunnerClass
.
_prepareSlapOS
...
@@ -934,7 +941,7 @@ shared = true
...
@@ -934,7 +941,7 @@ shared = true
SlapOSControler
.
runSoftwareRelease
=
runSoftwareRelease
SlapOSControler
.
runSoftwareRelease
=
runSoftwareRelease
def
callPrepareSlapOS
():
def
callPrepareSlapOS
():
runner
.
_prepareSlapOS
(
self
.
working_directory
,
node_test_suite
,
runner
.
_prepareSlapOS
(
self
.
working_directory
,
node_test_suite
,
test_node
.
log
,
create_partition
=
0
)
create_partition
=
0
)
def
callRaisingPrepareSlapos
():
def
callRaisingPrepareSlapos
():
self
.
assertRaises
(
SubprocessError
,
callPrepareSlapOS
)
self
.
assertRaises
(
SubprocessError
,
callPrepareSlapOS
)
...
@@ -989,9 +996,8 @@ shared = true
...
@@ -989,9 +996,8 @@ shared = true
def
patch_createTestResult
(
self
,
revision
,
test_name_list
,
node_title
,
def
patch_createTestResult
(
self
,
revision
,
test_name_list
,
node_title
,
allow_restart
=
False
,
test_title
=
None
,
project_title
=
None
):
allow_restart
=
False
,
test_title
=
None
,
project_title
=
None
):
test_result_path
=
os
.
path
.
join
(
test_result_path_root
,
test_title
)
test_result_path
=
os
.
path
.
join
(
test_result_path_root
,
test_title
)
result
=
TestResultProxy
(
self
.
_proxy
,
self
.
_retry_time
,
return
TestResultProxy
(
self
.
_proxy
,
self
.
_retry_time
,
self
.
_logger
,
test_result_path
,
node_title
,
revision
)
logger
,
test_result_path
,
node_title
,
revision
)
return
result
global
startTestSuiteDone
global
startTestSuiteDone
startTestSuiteDone
=
False
startTestSuiteDone
=
False
def
patch_startTestSuite
(
self
,
node_title
,
computer_guid
=
'unknown'
):
def
patch_startTestSuite
(
self
,
node_title
,
computer_guid
=
'unknown'
):
...
...
erp5/util/testnode/ProcessManager.py
View file @
199e7247
...
@@ -32,6 +32,7 @@ import threading
...
@@ -32,6 +32,7 @@ import threading
import
signal
import
signal
import
sys
import
sys
import
time
import
time
from
.
import
logger
MAX_TIMEOUT
=
3600
*
4
MAX_TIMEOUT
=
3600
*
4
...
@@ -67,28 +68,28 @@ def format_command(*args, **kw):
...
@@ -67,28 +68,28 @@ def format_command(*args, **kw):
cmdline
.
append
(
v
)
cmdline
.
append
(
v
)
return
' '
.
join
(
cmdline
)
return
' '
.
join
(
cmdline
)
def
subprocess_capture
(
p
,
log
,
log_prefix
,
get_output
=
True
):
def
subprocess_capture
(
p
,
log_prefix
,
get_output
=
True
):
def
readerthread
(
input
,
output
,
buffer
):
log
=
logger
.
info
if
log_prefix
:
log_prefix
+=
': '
def
readerthread
(
input
,
buffer
):
while
True
:
while
True
:
data
=
input
.
readline
()
data
=
input
.
readline
()
if
not
data
:
if
not
data
:
break
break
if
get_output
:
if
get_output
:
buffer
.
append
(
data
)
buffer
.
append
(
data
)
if
log_prefix
:
log
(
log_prefix
+
data
.
rstrip
(
'
\
n
'
))
data
=
"%s : "
%
log_prefix
+
data
data
=
data
.
rstrip
(
'
\
n
'
)
output
(
data
)
if
p
.
stdout
:
if
p
.
stdout
:
stdout
=
[]
stdout
=
[]
stdout_thread
=
threading
.
Thread
(
target
=
readerthread
,
stdout_thread
=
threading
.
Thread
(
target
=
readerthread
,
args
=
(
p
.
stdout
,
log
,
stdout
))
args
=
(
p
.
stdout
,
stdout
))
stdout_thread
.
daemon
=
True
stdout_thread
.
daemon
=
True
stdout_thread
.
start
()
stdout_thread
.
start
()
if
p
.
stderr
:
if
p
.
stderr
:
stderr
=
[]
stderr
=
[]
stderr_thread
=
threading
.
Thread
(
target
=
readerthread
,
stderr_thread
=
threading
.
Thread
(
target
=
readerthread
,
args
=
(
p
.
stderr
,
log
,
stderr
))
args
=
(
p
.
stderr
,
stderr
))
stderr_thread
.
daemon
=
True
stderr_thread
.
daemon
=
True
stderr_thread
.
start
()
stderr_thread
.
start
()
p
.
wait
()
p
.
wait
()
...
@@ -99,7 +100,7 @@ def subprocess_capture(p, log, log_prefix, get_output=True):
...
@@ -99,7 +100,7 @@ def subprocess_capture(p, log, log_prefix, get_output=True):
return
(
p
.
stdout
and
''
.
join
(
stdout
),
return
(
p
.
stdout
and
''
.
join
(
stdout
),
p
.
stderr
and
''
.
join
(
stderr
))
p
.
stderr
and
''
.
join
(
stderr
))
def
killCommand
(
pid
,
log
):
def
killCommand
(
pid
):
"""
"""
To prevent processes from reacting to the KILL of other processes,
To prevent processes from reacting to the KILL of other processes,
we STOP them all first, and we repeat until the list of children does not
we STOP them all first, and we repeat until the list of children does not
...
@@ -118,21 +119,20 @@ def killCommand(pid, log):
...
@@ -118,21 +119,20 @@ def killCommand(pid, log):
try
:
try
:
child
.
suspend
()
child
.
suspend
()
except
psutil
.
Error
,
e
:
except
psutil
.
Error
,
e
:
log
(
"killCommand/suspend: %s"
,
e
)
log
ger
.
debug
(
"killCommand/suspend: %s"
,
e
)
time
.
sleep
(
1
)
time
.
sleep
(
1
)
new_list
=
set
(
process
.
children
(
recursive
=
True
)).
difference
(
process_list
)
new_list
=
set
(
process
.
children
(
recursive
=
True
)).
difference
(
process_list
)
for
process
in
process_list
:
for
process
in
process_list
:
try
:
try
:
process
.
kill
()
process
.
kill
()
except
psutil
.
Error
,
e
:
except
psutil
.
Error
,
e
:
log
(
"killCommand/kill: %s"
,
e
)
log
ger
.
debug
(
"killCommand/kill: %s"
,
e
)
class
ProcessManager
(
object
):
class
ProcessManager
(
object
):
stdin
=
file
(
os
.
devnull
)
stdin
=
file
(
os
.
devnull
)
def
__init__
(
self
,
log
,
max_timeout
=
MAX_TIMEOUT
):
def
__init__
(
self
,
max_timeout
=
MAX_TIMEOUT
):
self
.
log
=
log
self
.
process_pid_set
=
set
()
self
.
process_pid_set
=
set
()
signal
.
signal
(
signal
.
SIGTERM
,
self
.
sigterm_handler
)
signal
.
signal
(
signal
.
SIGTERM
,
self
.
sigterm_handler
)
self
.
under_cancellation
=
False
self
.
under_cancellation
=
False
...
@@ -142,10 +142,10 @@ class ProcessManager(object):
...
@@ -142,10 +142,10 @@ class ProcessManager(object):
self
.
timer_set
=
set
()
self
.
timer_set
=
set
()
def
spawn
(
self
,
*
args
,
**
kw
):
def
spawn
(
self
,
*
args
,
**
kw
):
def
timeoutExpired
(
p
,
log
):
def
timeoutExpired
(
p
):
if
p
.
poll
()
is
None
:
if
p
.
poll
()
is
None
:
log
(
'PROCESS TOO LONG OR DEAD, GOING TO BE TERMINATED'
)
log
ger
.
warning
(
'PROCESS TOO LONG OR DEAD, GOING TO BE TERMINATED'
)
killCommand
(
p
.
pid
,
log
)
killCommand
(
p
.
pid
)
raise
SubprocessError
(
'Dead or too long process killed'
)
raise
SubprocessError
(
'Dead or too long process killed'
)
if
self
.
under_cancellation
:
if
self
.
under_cancellation
:
...
@@ -153,9 +153,6 @@ class ProcessManager(object):
...
@@ -153,9 +153,6 @@ class ProcessManager(object):
get_output
=
kw
.
pop
(
'get_output'
,
True
)
get_output
=
kw
.
pop
(
'get_output'
,
True
)
log_prefix
=
kw
.
pop
(
'log_prefix'
,
''
)
log_prefix
=
kw
.
pop
(
'log_prefix'
,
''
)
new_session
=
kw
.
pop
(
'new_session'
,
True
)
new_session
=
kw
.
pop
(
'new_session'
,
True
)
log
=
kw
.
pop
(
'log'
,
None
)
if
log
is
None
:
log
=
self
.
log
subprocess_kw
=
{}
subprocess_kw
=
{}
cwd
=
kw
.
pop
(
'cwd'
,
None
)
cwd
=
kw
.
pop
(
'cwd'
,
None
)
if
cwd
:
if
cwd
:
...
@@ -165,17 +162,16 @@ class ProcessManager(object):
...
@@ -165,17 +162,16 @@ class ProcessManager(object):
raise_error_if_fail
=
kw
.
pop
(
'raise_error_if_fail'
,
True
)
raise_error_if_fail
=
kw
.
pop
(
'raise_error_if_fail'
,
True
)
env
=
kw
and
dict
(
os
.
environ
,
**
kw
)
or
None
env
=
kw
and
dict
(
os
.
environ
,
**
kw
)
or
None
command
=
format_command
(
*
args
,
**
kw
)
command
=
format_command
(
*
args
,
**
kw
)
log
(
'subprocess_kw : %r'
%
(
subprocess_kw
,)
)
log
ger
.
info
(
'subprocess_kw : %r'
,
subprocess_kw
)
log
(
'$ '
+
command
)
log
ger
.
info
(
'$ %s'
,
command
)
sys
.
stdout
.
flush
()
sys
.
stdout
.
flush
()
p
=
subprocess
.
Popen
(
args
,
stdin
=
self
.
stdin
,
stdout
=
subprocess
.
PIPE
,
p
=
subprocess
.
Popen
(
args
,
stdin
=
self
.
stdin
,
stdout
=
subprocess
.
PIPE
,
stderr
=
subprocess
.
PIPE
,
env
=
env
,
**
subprocess_kw
)
stderr
=
subprocess
.
PIPE
,
env
=
env
,
**
subprocess_kw
)
self
.
process_pid_set
.
add
(
p
.
pid
)
self
.
process_pid_set
.
add
(
p
.
pid
)
timer
=
threading
.
Timer
(
self
.
max_timeout
,
timeoutExpired
,
args
=
(
p
,
log
))
timer
=
threading
.
Timer
(
self
.
max_timeout
,
timeoutExpired
,
args
=
(
p
,))
self
.
timer_set
.
add
(
timer
)
self
.
timer_set
.
add
(
timer
)
timer
.
start
()
timer
.
start
()
stdout
,
stderr
=
subprocess_capture
(
p
,
log
,
log_prefix
,
stdout
,
stderr
=
subprocess_capture
(
p
,
log_prefix
,
get_output
=
get_output
)
get_output
=
get_output
)
timer
.
cancel
()
timer
.
cancel
()
self
.
timer_set
.
discard
(
timer
)
self
.
timer_set
.
discard
(
timer
)
result
=
dict
(
status_code
=
p
.
returncode
,
command
=
command
,
result
=
dict
(
status_code
=
p
.
returncode
,
command
=
command
,
...
@@ -212,32 +208,39 @@ class ProcessManager(object):
...
@@ -212,32 +208,39 @@ class ProcessManager(object):
continue
continue
except (psutil.AccessDenied, psutil.NoSuchProcess):
except (psutil.AccessDenied, psutil.NoSuchProcess):
continue
continue
self.lo
g('ProcesssManager, killall on %s having pid %s',
logger.debu
g('ProcesssManager, killall on %s having pid %s',
name, process.pid)
name, process.pid)
to_kill_list.append(process.pid)
to_kill_list.append(process.pid)
for pid in to_kill_list:
for pid in to_kill_list:
killCommand(pid
, self.log
)
killCommand(pid)
def killPreviousRun(self, cancellation=False):
def killPreviousRun(self, cancellation=False):
self.lo
g('ProcessManager killPreviousRun, going to kill %r',
logger.debu
g('ProcessManager killPreviousRun, going to kill %r',
self.process_pid_set)
self.process_pid_set)
if cancellation:
if cancellation:
self.under_cancellation = True
self.under_cancellation = True
for timer in self.timer_set:
for timer in self.timer_set:
timer.cancel()
timer.cancel()
for pgpid in self.process_pid_set:
for pgpid in self.process_pid_set:
killCommand(pgpid
, self.log
)
killCommand(pgpid)
try:
try:
if os.path.exists(self.supervisord_pid_file):
pid_file = self.supervisord_pid_file
with open(self.supervisord_pid_file) as f:
except AttributeError:
supervisor_pid = int(f.read().strip())
pass
self.log('ProcessManager killPreviousRun, going to kill supervisor with pid %r',
else:
supervisor_pid)
del self.supervisord_pid_file
os.kill(supervisor_pid, signal.SIGTERM)
try:
except Exception:
if os.path.exists(pid_file):
self.log('ProcessManager killPreviousRun, exception when killing supervisor')
with open(pid_file) as f:
pid = int(f.read().strip())
logger.debug('ProcessManager killPreviousRun,'
' going to kill supervisor with pid %r', pid)
os.kill(pid, signal.SIGTERM)
except Exception:
logger.exception(
'ProcessManager killPreviousRun, exception when killing supervisor')
self.process_pid_set.clear()
self.process_pid_set.clear()
def sigterm_handler(self, signal, frame):
def sigterm_handler(self, signal, frame):
self.lo
g('SIGTERM_HANDLER')
logger.debu
g('SIGTERM_HANDLER')
sys.exit(1)
sys.exit(1)
erp5/util/testnode/ScalabilityTestRunner.py
View file @
199e7247
...
@@ -47,6 +47,8 @@ from erp5.util import taskdistribution
...
@@ -47,6 +47,8 @@ from erp5.util import taskdistribution
import
signal
import
signal
import
slapos.slap
import
slapos.slap
from
.
import
logger
# max time to instance changing state: 2 hour
# max time to instance changing state: 2 hour
MAX_INSTANCE_TIME
=
60
*
60
*
2
MAX_INSTANCE_TIME
=
60
*
60
*
2
# max time to register instance to slapOSMaster: 5 minutes
# max time to register instance to slapOSMaster: 5 minutes
...
@@ -59,12 +61,9 @@ MAX_PREPARE_TEST_SUITE = 3600*10*1.0 # 10 hours
...
@@ -59,12 +61,9 @@ MAX_PREPARE_TEST_SUITE = 3600*10*1.0 # 10 hours
class
ScalabilityTestRunner
():
class
ScalabilityTestRunner
():
def
__init__
(
self
,
testnode
):
def
__init__
(
self
,
testnode
):
self
.
testnode
=
testnode
self
.
testnode
=
testnode
self
.
log
=
self
.
testnode
.
log
self
.
slapos_controler
=
SlapOSControler
.
SlapOSControler
(
self
.
slapos_controler
=
SlapOSControler
.
SlapOSControler
(
self
.
testnode
.
working_directory
,
self
.
testnode
.
working_directory
,
self
.
testnode
.
config
,
self
.
testnode
.
config
)
self
.
log
)
# Create the slapos account configuration file and dir
# Create the slapos account configuration file and dir
key
=
self
.
testnode
.
test_suite_portal
.
getSlaposAccountKey
()
key
=
self
.
testnode
.
test_suite_portal
.
getSlaposAccountKey
()
certificate
=
self
.
testnode
.
test_suite_portal
.
getSlaposAccountCertificate
()
certificate
=
self
.
testnode
.
test_suite_portal
.
getSlaposAccountCertificate
()
...
@@ -80,8 +79,8 @@ class ScalabilityTestRunner():
...
@@ -80,8 +79,8 @@ class ScalabilityTestRunner():
# Get Slapos Master url used for api rest (using hateoas)
# Get Slapos Master url used for api rest (using hateoas)
self
.
slapos_api_rest_url
=
self
.
testnode
.
test_suite_portal
.
getSlaposHateoasUrl
()
self
.
slapos_api_rest_url
=
self
.
testnode
.
test_suite_portal
.
getSlaposHateoasUrl
()
self
.
log
(
"SlapOS Master url is: %s"
%
self
.
slapos_url
)
logger
.
info
(
"SlapOS Master url is: %s"
,
self
.
slapos_url
)
self
.
log
(
"SlapOS Master hateoas url is: %s"
%
self
.
slapos_api_rest_url
)
logger
.
info
(
"SlapOS Master hateoas url is: %s"
,
self
.
slapos_api_rest_url
)
self
.
key_path
,
self
.
cert_path
,
config_path
=
self
.
slapos_controler
.
createSlaposConfigurationFileAccount
(
self
.
key_path
,
self
.
cert_path
,
config_path
=
self
.
slapos_controler
.
createSlaposConfigurationFileAccount
(
key
,
certificate
,
self
.
slapos_url
,
self
.
testnode
.
config
)
key
,
certificate
,
self
.
slapos_url
,
self
.
testnode
.
config
)
...
@@ -102,7 +101,7 @@ class ScalabilityTestRunner():
...
@@ -102,7 +101,7 @@ class ScalabilityTestRunner():
"""
"""
A proxy to supply : Install a software on a specific node
A proxy to supply : Install a software on a specific node
"""
"""
self
.
log
(
"testnode, supply : %s %s"
,
software_path
,
computer_guid
)
logger
.
info
(
"testnode, supply : %s %s"
,
software_path
,
computer_guid
)
if
self
.
authorize_supply
:
if
self
.
authorize_supply
:
self
.
remaining_software_installation_dict
[
computer_guid
]
=
software_path
self
.
remaining_software_installation_dict
[
computer_guid
]
=
software_path
self
.
slapos_communicator
.
supply
(
software_path
,
computer_guid
)
self
.
slapos_communicator
.
supply
(
software_path
,
computer_guid
)
...
@@ -145,7 +144,7 @@ class ScalabilityTestRunner():
...
@@ -145,7 +144,7 @@ class ScalabilityTestRunner():
Create scalability instance
Create scalability instance
"""
"""
if
self
.
authorize_request
:
if
self
.
authorize_request
:
self
.
log
(
"testnode, request : %s"
,
instance_title
)
logger
.
info
(
"testnode, request : %s"
,
instance_title
)
config
=
self
.
_generateInstanceXML
(
software_configuration
,
config
=
self
.
_generateInstanceXML
(
software_configuration
,
test_result
,
test_suite
)
test_result
,
test_suite
)
request_kw
=
{
"partition_parameter_kw"
:
{
"_"
:
json
.
dumps
(
config
)}
}
request_kw
=
{
"partition_parameter_kw"
:
{
"_"
:
json
.
dumps
(
config
)}
}
...
@@ -168,14 +167,15 @@ ces or already launched.")
...
@@ -168,14 +167,15 @@ ces or already launched.")
# Dummy slapos answering
# Dummy slapos answering
def
_getSignal
(
self
,
signal
,
frame
):
def
_getSignal
(
self
,
signal
,
frame
):
self
.
lo
g
(
"Dummy SlapOS Master answer received."
)
logger
.
debu
g
(
"Dummy SlapOS Master answer received."
)
self
.
last_slapos_answer
.
append
(
True
)
self
.
last_slapos_answer
.
append
(
True
)
def
_prepareDummySlapOSAnswer
(
self
):
def
_prepareDummySlapOSAnswer
(
self
):
self
.
log
(
"Dummy slapOS answer enabled, send signal to %s (kill -10 %s) to simu
\
pid
=
os
.
getpid
()
late a SlapOS (positive) answer."
%
(
str
(
os
.
getpid
()),
str
(
os
.
getpid
()),))
logger
.
info
(
"Dummy slapOS answer enabled, send signal to %s (kill -USR1 %s)"
" to simulate a SlapOS (positive) answer."
,
pid
,
pid
)
signal
.
signal
(
signal
.
SIGUSR1
,
self
.
_getSignal
)
signal
.
signal
(
signal
.
SIGUSR1
,
self
.
_getSignal
)
def
_comeBackFromDummySlapOS
(
self
):
def
_comeBackFromDummySlapOS
(
self
):
self
.
log
(
"Dummy slapOS answer disabled, please don't send more signals."
)
logger
.
info
(
"Dummy slapOS answer disabled, please don't send more signals."
)
# use SIG_USR (kill)
# use SIG_USR (kill)
signal
.
signal
(
signal
.
SIGUSR1
,
signal
.
SIG_DFL
)
signal
.
signal
(
signal
.
SIGUSR1
,
signal
.
SIG_DFL
)
def
simulateSlapOSAnswer
(
self
):
def
simulateSlapOSAnswer
(
self
):
...
@@ -190,7 +190,8 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -190,7 +190,8 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
Return true if the specified software on the specified node is installed.
Return true if the specified software on the specified node is installed.
This method should communicates with SlapOS Master.
This method should communicates with SlapOS Master.
"""
"""
self
.
log
(
"Current software state: "
+
str
(
self
.
slapos_communicator
.
_getSoftwareState
()))
logger
.
info
(
"Current software state: %s"
,
self
.
slapos_communicator
.
_getSoftwareState
())
return
self
.
slapos_communicator
.
_getSoftwareState
()
==
SlapOSMasterCommunicator
.
SOFTWARE_STATE_INSTALLED
return
self
.
slapos_communicator
.
_getSoftwareState
()
==
SlapOSMasterCommunicator
.
SOFTWARE_STATE_INSTALLED
def
remainSoftwareToInstall
(
self
):
def
remainSoftwareToInstall
(
self
):
...
@@ -206,7 +207,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -206,7 +207,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
def
_updateInstanceXML
(
self
,
software_configuration
,
instance_title
,
def
_updateInstanceXML
(
self
,
software_configuration
,
instance_title
,
test_result
,
test_suite
):
test_result
,
test_suite
):
self
.
log
(
"testnode, updateInstanceXML : %s"
,
instance_title
)
logger
.
info
(
"testnode, updateInstanceXML : %s"
,
instance_title
)
config
=
self
.
_generateInstanceXML
(
software_configuration
,
config
=
self
.
_generateInstanceXML
(
software_configuration
,
test_result
,
test_suite
)
test_result
,
test_suite
)
request_kw
=
{
"partition_parameter_kw"
:
{
"_"
:
json
.
dumps
(
config
)}
}
request_kw
=
{
"partition_parameter_kw"
:
{
"_"
:
json
.
dumps
(
config
)}
}
...
@@ -217,15 +218,15 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -217,15 +218,15 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
"""
"""
Wait for 'max_time' the instance creation
Wait for 'max_time' the instance creation
"""
"""
self
.
lo
g
(
"Waiting for instance creation..."
)
logger
.
debu
g
(
"Waiting for instance creation..."
)
start_time
=
time
.
time
()
start_time
=
time
.
time
()
while
(
not
self
.
slapos_communicator
.
isInstanceRequested
(
instance_title
)
\
while
(
not
self
.
slapos_communicator
.
isInstanceRequested
(
instance_title
)
\
and
(
max_time
>
(
time
.
time
()
-
start_time
))
):
and
(
max_time
>
(
time
.
time
()
-
start_time
))
):
self
.
lo
g
(
"Instance not ready yet. Sleeping 5 sec."
)
logger
.
debu
g
(
"Instance not ready yet. Sleeping 5 sec."
)
time
.
sleep
(
5
)
time
.
sleep
(
5
)
if
(
time
.
time
()
-
start_time
)
>
max_time
:
if
(
time
.
time
()
-
start_time
)
>
max_time
:
raise
ValueError
(
"Instance '%s' not found after %s seconds"
%
(
instance_title
,
max_time
))
raise
ValueError
(
"Instance '%s' not found after %s seconds"
%
(
instance_title
,
max_time
))
self
.
lo
g
(
"Instance found on slapOSMaster"
)
logger
.
debu
g
(
"Instance found on slapOSMaster"
)
def
_initializeSlapOSConnection
(
self
):
def
_initializeSlapOSConnection
(
self
):
"""
"""
...
@@ -244,7 +245,8 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -244,7 +245,8 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
slapgrid_rest_uri
=
self
.
slapos_api_rest_url
)
slapgrid_rest_uri
=
self
.
slapos_api_rest_url
)
if
getattr
(
slap
,
'_hateoas_navigator'
,
None
)
is
None
:
if
getattr
(
slap
,
'_hateoas_navigator'
,
None
)
is
None
:
retry
+=
1
retry
+=
1
self
.
log
(
"Fail to load _hateoas_navigator waiting a bit and retry."
)
logger
.
info
(
"Fail to load _hateoas_navigator waiting a bit and retry."
)
time
.
sleep
(
30
)
time
.
sleep
(
30
)
else
:
else
:
break
break
...
@@ -268,12 +270,13 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -268,12 +270,13 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
not
os
.
path
.
exists
(
self
.
obfuscated_link_path
)
)
:
not
os
.
path
.
exists
(
self
.
obfuscated_link_path
)
)
:
try
:
try
:
os
.
symlink
(
path_to_suite
,
self
.
obfuscated_link_path
)
os
.
symlink
(
path_to_suite
,
self
.
obfuscated_link_path
)
self
.
log
(
"testnode, Symbolic link (%s->%s) created."
logger
.
info
(
"testnode, Symbolic link (%s->%s) created."
,
%
(
self
.
obfuscated_link_path
,
path_to_suite
))
self
.
obfuscated_link_path
,
path_to_suite
)
except
:
except
Exception
:
self
.
log
(
"testnode, Unable to create symbolic link to the testsuite."
)
msg
=
"testnode, Unable to create symbolic link to the testsuite."
raise
ValueError
(
"testnode, Unable to create symbolic link to the testsuite."
)
logger
.
exception
(
msg
)
self
.
log
(
"Sym link : %s %s"
%
(
path_to_suite
,
self
.
obfuscated_link_path
))
raise
ValueError
(
msg
)
logger
.
info
(
"Sym link : %s %s"
,
path_to_suite
,
self
.
obfuscated_link_path
)
# Construct the ipv6 obfuscated url of the software profile reachable from outside
# Construct the ipv6 obfuscated url of the software profile reachable from outside
self
.
reachable_address
=
os
.
path
.
join
(
self
.
reachable_address
=
os
.
path
.
join
(
...
@@ -301,7 +304,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -301,7 +304,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
"""
"""
Install testsuite softwares
Install testsuite softwares
"""
"""
self
.
lo
g
(
'Preparing SlapOS for Test Suite...'
)
logger
.
debu
g
(
'Preparing SlapOS for Test Suite...'
)
max_time
=
MAX_PREPARE_TEST_SUITE
max_time
=
MAX_PREPARE_TEST_SUITE
interval_time
=
60
interval_time
=
60
start_time
=
time
.
time
()
start_time
=
time
.
time
()
...
@@ -320,9 +323,9 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -320,9 +323,9 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
self
.
error_message
=
test_configuration
[
'error_message'
]
self
.
error_message
=
test_configuration
[
'error_message'
]
self
.
randomized_path
=
test_configuration
[
'randomized_path'
]
self
.
randomized_path
=
test_configuration
[
'randomized_path'
]
if
not
self
.
launchable
:
if
not
self
.
launchable
:
self
.
log
(
"Test suite %s is not actually launchable with
\
logger
.
info
(
"Test suite %s is not actually launchable"
the current cluster configuration."
%
(
node_test_suite
.
test_suite_title
,)
)
" with the current cluster configuration."
,
node_test_suite
.
test_suite_title
)
self
.
log
(
"ERP5 Master indicates : %s"
%
(
self
.
error_message
,)
)
logger
.
info
(
"ERP5 Master indicates : %s"
,
self
.
error_message
)
return
{
'status_code'
:
1
}
return
{
'status_code'
:
1
}
configuration_list
=
test_configuration
[
'configuration_list'
]
configuration_list
=
test_configuration
[
'configuration_list'
]
...
@@ -331,12 +334,12 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -331,12 +334,12 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
self
.
instance_title
=
self
.
_generateInstanceTitle
(
node_test_suite
.
test_suite_title
)
self
.
instance_title
=
self
.
_generateInstanceTitle
(
node_test_suite
.
test_suite_title
)
self
.
createSoftwareReachableProfilePath
(
node_test_suite
)
self
.
createSoftwareReachableProfilePath
(
node_test_suite
)
self
.
log
(
"Software reachable profile path is : %s "
%
(
self
.
reachable_profile
))
logger
.
info
(
"Software reachable profile path is: %s"
,
self
.
reachable_profile
)
# Initialize SlapOS Master Communicator
# Initialize SlapOS Master Communicator
self
.
slapos_communicator
=
SlapOSMasterCommunicator
.
SlapOSTester
(
self
.
slapos_communicator
=
SlapOSMasterCommunicator
.
SlapOSTester
(
self
.
instance_title
,
self
.
instance_title
,
self
.
log
,
slap
,
slap
,
order
,
order
,
supply
,
supply
,
...
@@ -353,9 +356,9 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -353,9 +356,9 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
# Waiting until all softwares are installed
# Waiting until all softwares are installed
while
(
self
.
remainSoftwareToInstall
()
while
(
self
.
remainSoftwareToInstall
()
and
(
max_time
>
(
time
.
time
()
-
start_time
))):
and
(
max_time
>
(
time
.
time
()
-
start_time
))):
self
.
log
(
"Master testnode is waiting
\
logger
.
info
(
"Master testnode is waiting for the end of"
for the end of
all software installation (for %ss) PID=%s."
,
"
all software installation (for %ss) PID=%s."
,
str
(
int
(
time
.
time
()
-
start_time
)),
str
(
os
.
getpid
()
))
int
(
time
.
time
()
-
start_time
),
os
.
getpid
(
))
time
.
sleep
(
interval_time
)
time
.
sleep
(
interval_time
)
# TODO : remove the line below wich simulate an answer from slapos master
# TODO : remove the line below wich simulate an answer from slapos master
self
.
_comeBackFromDummySlapOS
()
self
.
_comeBackFromDummySlapOS
()
...
@@ -363,7 +366,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -363,7 +366,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
# All softwares are not installed, however maxtime is elapsed, that's a failure.
# All softwares are not installed, however maxtime is elapsed, that's a failure.
return
{
'status_code'
:
1
}
return
{
'status_code'
:
1
}
self
.
authorize_request
=
True
self
.
authorize_request
=
True
self
.
lo
g
(
"Softwares installed."
)
logger
.
debu
g
(
"Softwares installed."
)
# Launch instance
# Launch instance
try
:
try
:
self
.
_createInstance
(
self
.
reachable_profile
,
self
.
_createInstance
(
self
.
reachable_profile
,
...
@@ -371,10 +374,11 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -371,10 +374,11 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
self
.
instance_title
,
self
.
instance_title
,
node_test_suite
.
test_result
,
node_test_suite
.
test_result
,
node_test_suite
.
test_suite
)
node_test_suite
.
test_suite
)
self
.
log
(
"Scalability instance requested."
)
logger
.
debug
(
"Scalability instance requested."
)
except
:
except
Exception
:
self
.
log
(
"Unable to launch instance"
)
msg
=
"Unable to launch instance"
raise
ValueError
(
"Unable to launch instance"
)
logger
.
exception
(
msg
)
raise
ValueError
(
msg
)
self
.
_waitInstanceCreation
(
self
.
instance_title
)
self
.
_waitInstanceCreation
(
self
.
instance_title
)
return
{
'status_code'
:
0
}
return
{
'status_code'
:
0
}
...
@@ -383,7 +387,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -383,7 +387,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
def
runTestSuite
(
self
,
node_test_suite
,
portal_url
):
def
runTestSuite
(
self
,
node_test_suite
,
portal_url
):
if
not
self
.
launchable
:
if
not
self
.
launchable
:
self
.
log
(
"Current test_suite is not actually launchable."
)
logger
.
info
(
"Current test_suite is not actually launchable."
)
return
{
'status_code'
:
1
}
# Unable to continue due to not realizable configuration
return
{
'status_code'
:
1
}
# Unable to continue due to not realizable configuration
configuration_list
=
node_test_suite
.
configuration_list
configuration_list
=
node_test_suite
.
configuration_list
test_list
=
range
(
0
,
len
(
configuration_list
))
test_list
=
range
(
0
,
len
(
configuration_list
))
...
@@ -392,7 +396,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -392,7 +396,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
self
.
testnode
.
config
[
'test_node_title'
],
self
.
testnode
.
config
[
'test_node_title'
],
True
,
node_test_suite
.
test_suite_title
,
True
,
node_test_suite
.
test_suite_title
,
node_test_suite
.
project_title
)
node_test_suite
.
project_title
)
self
.
lo
g
(
"Test Result created."
)
logger
.
debu
g
(
"Test Result created."
)
count
=
0
count
=
0
error_message
=
None
error_message
=
None
...
@@ -410,11 +414,11 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -410,11 +414,11 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
self
.
slapos_communicator
.
requestInstanceStart
()
self
.
slapos_communicator
.
requestInstanceStart
()
self
.
slapos_communicator
.
waitInstanceStarted
(
self
.
instance_title
)
self
.
slapos_communicator
.
waitInstanceStarted
(
self
.
instance_title
)
self
.
log
(
"[DEBUG]
INSTANCE CORRECTLY STARTED"
)
logger
.
debug
(
"
INSTANCE CORRECTLY STARTED"
)
# ROQUE XXX : for debug
# ROQUE XXX : for debug
if
True
:
if
True
:
self
.
lo
g
(
"RETURN FOR DEBUG"
)
logger
.
debu
g
(
"RETURN FOR DEBUG"
)
return
{
'status_code'
:
0
}
return
{
'status_code'
:
0
}
# Start only the current test
# Start only the current test
...
@@ -426,7 +430,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -426,7 +430,7 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
error_message
=
"Test case already tested."
error_message
=
"Test case already tested."
break
break
self
.
log
(
"Test for count : %d is in a running state."
%
count
)
logger
.
info
(
"Test for count : %d is in a running state."
,
count
)
# Wait for test case ending
# Wait for test case ending
test_case_start_time
=
time
.
time
()
test_case_start_time
=
time
.
time
()
...
@@ -463,14 +467,14 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -463,14 +467,14 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
test_result_line_proxy
.
stop
(
error_count
=
1
,
failure_count
=
1
,
test_result_line_proxy
.
stop
(
error_count
=
1
,
failure_count
=
1
,
stdout
=
error_message
,
stderr
=
error_message
)
stdout
=
error_message
,
stderr
=
error_message
)
test_result_proxy
.
reportFailure
(
stdout
=
error_message
)
test_result_proxy
.
reportFailure
(
stdout
=
error_message
)
self
.
lo
g
(
"Test Failed."
)
logger
.
debu
g
(
"Test Failed."
)
return
{
'status_code'
:
1
,
'error_message'
:
error_message
}
return
{
'status_code'
:
1
,
'error_message'
:
error_message
}
# Test is finished.
# Test is finished.
self
.
lo
g
(
"Test finished."
)
logger
.
debu
g
(
"Test finished."
)
return
{
'status_code'
:
0
}
return
{
'status_code'
:
0
}
def
_cleanUpOldInstance
(
self
):
def
_cleanUpOldInstance
(
self
):
self
.
lo
g
(
"_cleanUpOldInstance"
)
logger
.
debu
g
(
"_cleanUpOldInstance"
)
# Get title and link list of all instances
# Get title and link list of all instances
instance_dict
=
self
.
slapos_communicator
.
getHostingSubscriptionDict
()
instance_dict
=
self
.
slapos_communicator
.
getHostingSubscriptionDict
()
...
@@ -502,7 +506,8 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
...
@@ -502,7 +506,8 @@ late a SlapOS (positive) answer." %(str(os.getpid()),str(os.getpid()),))
computer_guid
=
instance_information_dict
[
'computer_guid'
],
computer_guid
=
instance_information_dict
[
'computer_guid'
],
state
=
'destroyed'
state
=
'destroyed'
)
)
self
.
log
(
"Instance '%s' deleted."
%
instance_information_dict
[
'title'
])
logger
.
debug
(
"Instance '%s' deleted."
,
instance_information_dict
[
'title'
])
def
_cleanUpNodesInformation
(
self
):
def
_cleanUpNodesInformation
(
self
):
self
.
involved_nodes_computer_guid
=
[]
self
.
involved_nodes_computer_guid
=
[]
...
...
erp5/util/testnode/SlapOSControler.py
View file @
199e7247
...
@@ -30,9 +30,9 @@ import slapos.slap
...
@@ -30,9 +30,9 @@ import slapos.slap
import
subprocess
import
subprocess
import
time
import
time
import
xml_marshaller
import
xml_marshaller
import
sys
import
argparse
import
argparse
from
slapos
import
client
from
slapos
import
client
from
.
import
logger
from
.Utils
import
createFolder
from
.Utils
import
createFolder
MAX_PARTITIONS
=
10
MAX_PARTITIONS
=
10
...
@@ -40,12 +40,11 @@ MAX_SR_RETRIES = 3
...
@@ -40,12 +40,11 @@ MAX_SR_RETRIES = 3
class
SlapOSControler
(
object
):
class
SlapOSControler
(
object
):
def
__init__
(
self
,
working_directory
,
config
,
log
):
def
__init__
(
self
,
working_directory
,
config
):
self
.
config
=
config
self
.
config
=
config
self
.
software_root
=
os
.
path
.
join
(
working_directory
,
'soft'
)
self
.
software_root
=
os
.
path
.
join
(
working_directory
,
'soft'
)
self
.
instance_root
=
os
.
path
.
join
(
working_directory
,
'inst'
)
self
.
instance_root
=
os
.
path
.
join
(
working_directory
,
'inst'
)
self
.
slapos_config
=
os
.
path
.
join
(
working_directory
,
'slapos.cfg'
)
self
.
slapos_config
=
os
.
path
.
join
(
working_directory
,
'slapos.cfg'
)
self
.
log
=
log
self
.
proxy_database
=
os
.
path
.
join
(
working_directory
,
'proxy.db'
)
self
.
proxy_database
=
os
.
path
.
join
(
working_directory
,
'proxy.db'
)
self
.
instance_config
=
{}
self
.
instance_config
=
{}
...
@@ -80,7 +79,7 @@ class SlapOSControler(object):
...
@@ -80,7 +79,7 @@ class SlapOSControler(object):
Ex :
Ex :
my_controler.supply('kvm.cfg', 'COMP-726')
my_controler.supply('kvm.cfg', 'COMP-726')
"""
"""
self
.
lo
g
(
'SlapOSControler : supply'
)
logger
.
debu
g
(
'SlapOSControler : supply'
)
parser
=
argparse
.
ArgumentParser
()
parser
=
argparse
.
ArgumentParser
()
parser
.
add_argument
(
"configuration_file"
)
parser
.
add_argument
(
"configuration_file"
)
parser
.
add_argument
(
"software_url"
)
parser
.
add_argument
(
"software_url"
)
...
@@ -92,9 +91,9 @@ class SlapOSControler(object):
...
@@ -92,9 +91,9 @@ class SlapOSControler(object):
try
:
try
:
local
=
client
.
init
(
config
)
local
=
client
.
init
(
config
)
local
[
'supply'
](
software_url
,
computer_guid
=
computer_id
,
state
=
state
)
local
[
'supply'
](
software_url
,
computer_guid
=
computer_id
,
state
=
state
)
self
.
log
(
'SlapOSControler
: supply %s %s %s'
,
software_url
,
computer_id
,
state
)
logger
.
debug
(
'SlapOSControler
: supply %s %s %s'
,
software_url
,
computer_id
,
state
)
except
Exception
:
except
Exception
:
self
.
log
(
"SlapOSControler.supply"
,
exc_info
=
sys
.
exc_info
()
)
logger
.
exception
(
"SlapOSControler.supply"
)
raise
ValueError
(
"Unable to supply (or remove)"
)
raise
ValueError
(
"Unable to supply (or remove)"
)
else
:
else
:
raise
ValueError
(
"Configuration file not found."
)
raise
ValueError
(
"Configuration file not found."
)
...
@@ -113,7 +112,7 @@ class SlapOSControler(object):
...
@@ -113,7 +112,7 @@ class SlapOSControler(object):
'kvm.cfg', 'cluster', { "_" : "{'toto' : 'titi'}" } )
'kvm.cfg', 'cluster', { "_" : "{'toto' : 'titi'}" } )
"""
"""
self
.
lo
g
(
'SlapOSControler : request-->SlapOSMaster'
)
logger
.
debu
g
(
'SlapOSControler : request-->SlapOSMaster'
)
current_intance_config
=
{
'software_type'
:
software_type
,
current_intance_config
=
{
'software_type'
:
software_type
,
'software_configuration'
:
software_configuration
,
'software_configuration'
:
software_configuration
,
'computer_guid'
:
computer_guid
,
'computer_guid'
:
computer_guid
,
...
@@ -145,10 +144,10 @@ class SlapOSControler(object):
...
@@ -145,10 +144,10 @@ class SlapOSControler(object):
if
state
==
'destroyed'
:
if
state
==
'destroyed'
:
del
self
.
instance_config
[
reference
]
del
self
.
instance_config
[
reference
]
elif
state
==
'started'
:
elif
state
==
'started'
:
self
.
lo
g
(
'Instance started with configuration: %s'
,
logger
.
debu
g
(
'Instance started with configuration: %s'
,
software_configuration
)
software_configuration
)
except
Exception
:
except
Exception
:
self
.
log
(
"SlapOSControler.request"
,
exc_info
=
sys
.
exc_info
()
)
logger
.
exception
(
"SlapOSControler.request"
)
raise
ValueError
(
"Unable to do this request"
)
raise
ValueError
(
"Unable to do this request"
)
else
:
else
:
raise
ValueError
(
"Configuration file not found."
)
raise
ValueError
(
"Configuration file not found."
)
...
@@ -163,21 +162,21 @@ class SlapOSControler(object):
...
@@ -163,21 +162,21 @@ class SlapOSControler(object):
)
)
def
destroyInstance
(
self
,
reference
):
def
destroyInstance
(
self
,
reference
):
self
.
lo
g
(
'SlapOSControler : delete instance'
)
logger
.
debu
g
(
'SlapOSControler : delete instance'
)
try
:
try
:
self
.
_requestSpecificState
(
reference
,
'destroyed'
)
self
.
_requestSpecificState
(
reference
,
'destroyed'
)
except
Exception
:
except
Exception
:
raise
ValueError
(
"Can't delete instance %r (instance not created?)"
%
reference
)
raise
ValueError
(
"Can't delete instance %r (instance not created?)"
%
reference
)
def
stopInstance
(
self
,
reference
):
def
stopInstance
(
self
,
reference
):
self
.
lo
g
(
'SlapOSControler : stop instance'
)
logger
.
debu
g
(
'SlapOSControler : stop instance'
)
try
:
try
:
self
.
_requestSpecificState
(
reference
,
'stopped'
)
self
.
_requestSpecificState
(
reference
,
'stopped'
)
except
Exception
:
except
Exception
:
raise
ValueError
(
"Can't stop instance %r (instance not created?)"
%
reference
)
raise
ValueError
(
"Can't stop instance %r (instance not created?)"
%
reference
)
def
startInstance
(
self
,
reference
):
def
startInstance
(
self
,
reference
):
self
.
lo
g
(
'SlapOSControler : start instance'
)
logger
.
debu
g
(
'SlapOSControler : start instance'
)
try
:
try
:
self
.
_requestSpecificState
(
reference
,
'started'
)
self
.
_requestSpecificState
(
reference
,
'started'
)
except
Exception
:
except
Exception
:
...
@@ -188,8 +187,8 @@ class SlapOSControler(object):
...
@@ -188,8 +187,8 @@ class SlapOSControler(object):
Update the XML configuration of an instance
Update the XML configuration of an instance
# Request same instance with different parameters.
# Request same instance with different parameters.
"""
"""
self
.
lo
g
(
'SlapOSControler : updateInstanceXML will request same'
logger
.
debu
g
(
'SlapOSControler : updateInstanceXML will request same'
' instance with new XML configuration...'
)
' instance with new XML configuration...'
)
try
:
try
:
self
.
request
(
reference
,
self
.
request
(
reference
,
...
@@ -203,7 +202,7 @@ class SlapOSControler(object):
...
@@ -203,7 +202,7 @@ class SlapOSControler(object):
raise
ValueError
(
"Can't update instance '%s' (may not exist?)"
%
reference
)
raise
ValueError
(
"Can't update instance '%s' (may not exist?)"
%
reference
)
def
_resetSoftware
(
self
):
def
_resetSoftware
(
self
):
self
.
log
(
'SlapOSControler
: GOING TO RESET ALL SOFTWARE : %r'
,
logger
.
info
(
'SlapOSControler
: GOING TO RESET ALL SOFTWARE : %r'
,
self
.
software_root
)
self
.
software_root
)
createFolder
(
self
.
software_root
,
True
)
createFolder
(
self
.
software_root
,
True
)
...
@@ -211,7 +210,7 @@ class SlapOSControler(object):
...
@@ -211,7 +210,7 @@ class SlapOSControler(object):
reset_software
=
False
,
software_path_list
=
None
):
reset_software
=
False
,
software_path_list
=
None
):
self
.
process_manager
=
process_manager
self
.
process_manager
=
process_manager
self
.
software_path_list
=
software_path_list
self
.
software_path_list
=
software_path_list
self
.
lo
g
(
'SlapOSControler, initialize, reset_software: %r'
,
reset_software
)
logger
.
debu
g
(
'SlapOSControler, initialize, reset_software: %r'
,
reset_software
)
config
=
self
.
config
config
=
self
.
config
slapos_config_dict
=
config
.
copy
()
slapos_config_dict
=
config
.
copy
()
slapos_config_dict
.
update
(
software_root
=
self
.
software_root
,
slapos_config_dict
.
update
(
software_root
=
self
.
software_root
,
...
@@ -248,8 +247,7 @@ class SlapOSControler(object):
...
@@ -248,8 +247,7 @@ class SlapOSControler(object):
computer_guid
=
config
[
'computer_id'
])
computer_guid
=
config
[
'computer_id'
])
computer
=
slap
.
registerComputer
(
config
[
'computer_id'
])
computer
=
slap
.
registerComputer
(
config
[
'computer_id'
])
except
Exception
:
except
Exception
:
self
.
log
(
"SlapOSControler.initializeSlapOSControler"
,
logger
.
exception
(
"SlapOSControler.initializeSlapOSControler"
)
exc_info
=
sys
.
exc_info
())
raise
ValueError
(
"Unable to registerSupply"
)
raise
ValueError
(
"Unable to registerSupply"
)
# Reset all previously generated software if needed
# Reset all previously generated software if needed
if
reset_software
:
if
reset_software
:
...
@@ -289,7 +287,7 @@ class SlapOSControler(object):
...
@@ -289,7 +287,7 @@ class SlapOSControler(object):
return
self
.
process_manager
.
spawn
(
*
args
,
**
kw
)
return
self
.
process_manager
.
spawn
(
*
args
,
**
kw
)
def
runSoftwareRelease
(
self
,
config
,
environment
,
**
kw
):
def
runSoftwareRelease
(
self
,
config
,
environment
,
**
kw
):
self
.
lo
g
(
"SlapOSControler.runSoftwareRelease"
)
logger
.
debu
g
(
"SlapOSControler.runSoftwareRelease"
)
cpu_count
=
str
(
os
.
sysconf
(
"SC_NPROCESSORS_ONLN"
))
cpu_count
=
str
(
os
.
sysconf
(
"SC_NPROCESSORS_ONLN"
))
os
.
environ
[
'MAKEFLAGS'
]
=
'-j'
+
cpu_count
os
.
environ
[
'MAKEFLAGS'
]
=
'-j'
+
cpu_count
os
.
environ
[
'NPY_NUM_BUILD_JOBS'
]
=
cpu_count
os
.
environ
[
'NPY_NUM_BUILD_JOBS'
]
=
cpu_count
...
@@ -309,7 +307,7 @@ class SlapOSControler(object):
...
@@ -309,7 +307,7 @@ class SlapOSControler(object):
def
runComputerPartition
(
self
,
config
,
environment
,
def
runComputerPartition
(
self
,
config
,
environment
,
stdout
=
None
,
stderr
=
None
,
cluster_configuration
=
None
,
**
kw
):
stdout
=
None
,
stderr
=
None
,
cluster_configuration
=
None
,
**
kw
):
self
.
lo
g
(
"SlapOSControler.runComputerPartition with cluster_config: %r"
,
logger
.
debu
g
(
"SlapOSControler.runComputerPartition with cluster_config: %r"
,
cluster_configuration
)
cluster_configuration
)
for
path
in
self
.
software_path_list
:
for
path
in
self
.
software_path_list
:
try
:
try
:
...
@@ -318,7 +316,7 @@ class SlapOSControler(object):
...
@@ -318,7 +316,7 @@ class SlapOSControler(object):
self
.
software_path_list
.
index
(
path
),
self
.
software_path_list
.
index
(
path
),
partition_parameter_kw
=
cluster_configuration
)
partition_parameter_kw
=
cluster_configuration
)
except
Exception
:
except
Exception
:
self
.
log
(
"SlapOSControler.runComputerPartition"
,
exc_info
=
sys
.
exc_info
()
)
logger
.
exception
(
"SlapOSControler.runComputerPartition"
)
raise
ValueError
(
"Unable to registerOpenOrder"
)
raise
ValueError
(
"Unable to registerOpenOrder"
)
# try to run for all partitions as one partition may in theory request another one
# try to run for all partitions as one partition may in theory request another one
...
@@ -329,7 +327,7 @@ class SlapOSControler(object):
...
@@ -329,7 +327,7 @@ class SlapOSControler(object):
'--pidfile'
,
os
.
path
.
join
(
self
.
instance_root
,
'slapos-node.pid'
),
'--pidfile'
,
os
.
path
.
join
(
self
.
instance_root
,
'slapos-node.pid'
),
'--cfg'
,
self
.
slapos_config
,
raise_error_if_fail
=
False
,
'--cfg'
,
self
.
slapos_config
,
raise_error_if_fail
=
False
,
log_prefix
=
'slapgrid_cp'
,
get_output
=
False
)
log_prefix
=
'slapgrid_cp'
,
get_output
=
False
)
self
.
lo
g
(
'slapgrid_cp status_dict : %r'
,
status_dict
)
logger
.
debu
g
(
'slapgrid_cp status_dict : %r'
,
status_dict
)
if
not
status_dict
[
'status_code'
]:
if
not
status_dict
[
'status_code'
]:
break
break
else
:
else
:
...
...
erp5/util/testnode/SlapOSMasterCommunicator.py
View file @
199e7247
...
@@ -10,6 +10,7 @@ import slapos.slap
...
@@ -10,6 +10,7 @@ import slapos.slap
from
slapos.slap
import
SoftwareProductCollection
from
slapos.slap
import
SoftwareProductCollection
from
requests.exceptions
import
HTTPError
from
requests.exceptions
import
HTTPError
from
..taskdistribution
import
SAFE_RPC_EXCEPTION_LIST
from
..taskdistribution
import
SAFE_RPC_EXCEPTION_LIST
from
.
import
logger
# max time to instance changing state: 2 hour
# max time to instance changing state: 2 hour
MAX_INSTANCE_TIME
=
60
*
60
*
2
MAX_INSTANCE_TIME
=
60
*
60
*
2
...
@@ -61,8 +62,7 @@ def retryOnNetworkFailure(func,
...
@@ -61,8 +62,7 @@ def retryOnNetworkFailure(func,
class
SlapOSMasterCommunicator
(
object
):
class
SlapOSMasterCommunicator
(
object
):
latest_state
=
None
latest_state
=
None
def
__init__
(
self
,
slap
,
slap_supply
,
slap_order
,
url
,
logger
):
def
__init__
(
self
,
slap
,
slap_supply
,
slap_order
,
url
):
self
.
_logger
=
logger
self
.
slap
=
slap
self
.
slap
=
slap
self
.
slap_order
=
slap_order
self
.
slap_order
=
slap_order
self
.
slap_supply
=
slap_supply
self
.
slap_supply
=
slap_supply
...
@@ -72,20 +72,20 @@ class SlapOSMasterCommunicator(object):
...
@@ -72,20 +72,20 @@ class SlapOSMasterCommunicator(object):
if
url
is
not
None
and
\
if
url
is
not
None
and
\
url
.
startswith
(
SOFTWARE_PRODUCT_NAMESPACE
):
url
.
startswith
(
SOFTWARE_PRODUCT_NAMESPACE
):
product
=
SoftwareProductCollection
(
self
.
_
logger
,
self
.
slap
)
product
=
SoftwareProductCollection
(
logger
,
self
.
slap
)
try
:
try
:
url
=
product
.
__getattr__
(
url
[
len
(
SOFTWARE_PRODUCT_NAMESPACE
):])
url
=
product
.
__getattr__
(
url
[
len
(
SOFTWARE_PRODUCT_NAMESPACE
):])
except
AttributeError
as
e
:
except
AttributeError
as
e
:
self
.
_logger
.
warning
(
'Error on get software release : %s '
%
e
.
message
)
logger
.
warning
(
'Error on get software release: %s '
,
e
.
message
)
self
.
url
=
url
self
.
url
=
url
@
retryOnNetworkFailure
@
retryOnNetworkFailure
def
_supply
(
self
):
def
_supply
(
self
):
if
self
.
computer_guid
is
None
:
if
self
.
computer_guid
is
None
:
self
.
_logger
(
'Nothing to supply for %s.'
%
(
self
.
name
)
)
logger
.
info
(
'Nothing to supply for %s.'
,
self
.
name
)
return
None
return
None
self
.
_logger
(
'Supply %s@%s'
,
self
.
url
,
self
.
computer_guid
)
logger
.
info
(
'Supply %s@%s'
,
self
.
url
,
self
.
computer_guid
)
return
self
.
slap_supply
.
supply
(
self
.
url
,
self
.
computer_guid
)
return
self
.
slap_supply
.
supply
(
self
.
url
,
self
.
computer_guid
)
@
retryOnNetworkFailure
@
retryOnNetworkFailure
...
@@ -97,7 +97,7 @@ class SlapOSMasterCommunicator(object):
...
@@ -97,7 +97,7 @@ class SlapOSMasterCommunicator(object):
self
.
request_kw
=
json
.
loads
(
request_kw
)
self
.
request_kw
=
json
.
loads
(
request_kw
)
else
:
else
:
self
.
request_kw
=
request_kw
self
.
request_kw
=
request_kw
self
.
_logger
(
'Request %s@%s: %s'
,
self
.
url
,
self
.
name
,
state
)
logger
.
info
(
'Request %s@%s: %s'
,
self
.
url
,
self
.
name
,
state
)
self
.
latest_state
=
state
self
.
latest_state
=
state
return
self
.
slap_order
.
request
(
return
self
.
slap_order
.
request
(
software_release
=
self
.
url
,
software_release
=
self
.
url
,
...
@@ -252,7 +252,7 @@ class SlapOSMasterCommunicator(object):
...
@@ -252,7 +252,7 @@ class SlapOSMasterCommunicator(object):
@
retryOnNetworkFailure
@
retryOnNetworkFailure
def
_getInstanceState
(
self
):
def
_getInstanceState
(
self
):
latest_state
=
self
.
latest_state
latest_state
=
self
.
latest_state
self
.
_logger
(
'latest_state = %r'
,
latest_state
)
logger
.
info
(
'latest_state = %r'
,
latest_state
)
if
latest_state
is
None
:
if
latest_state
is
None
:
return
INSTANCE_STATE_UNKNOWN
return
INSTANCE_STATE_UNKNOWN
...
@@ -297,8 +297,8 @@ class SlapOSMasterCommunicator(object):
...
@@ -297,8 +297,8 @@ class SlapOSMasterCommunicator(object):
try
:
try
:
monitor_information_dict
=
self
.
getRSSEntryFromMonitoring
(
monitor_v6_url
)
monitor_information_dict
=
self
.
getRSSEntryFromMonitoring
(
monitor_v6_url
)
except
Exception
:
except
Exception
:
self
.
_logger
(
'Unable to download promises for: %s'
%
(
instance
[
"title"
]))
logger
.
exception
(
'Unable to download promises for: %s'
,
self
.
_logger
(
traceback
.
format_exc
()
)
instance
[
"title"
]
)
monitor_information_dict
=
{
"message"
:
"Unable to download"
}
monitor_information_dict
=
{
"message"
:
"Unable to download"
}
message_list
.
append
({
message_list
.
append
({
...
@@ -311,11 +311,10 @@ class SlapOSMasterCommunicator(object):
...
@@ -311,11 +311,10 @@ class SlapOSMasterCommunicator(object):
})
})
except
slapos
.
slap
.
ServerError
:
except
slapos
.
slap
.
ServerError
:
self
.
_logger
(
'Got an error requesting partition for '
logger
.
error
(
'Got an error requesting partition for its state'
)
'its state'
)
return
INSTANCE_STATE_UNKNOWN
return
INSTANCE_STATE_UNKNOWN
except
Exception
:
except
Exception
:
self
.
_logge
r
(
"ERROR getting instance state"
)
logger
.
erro
r
(
"ERROR getting instance state"
)
return
INSTANCE_STATE_UNKNOWN
return
INSTANCE_STATE_UNKNOWN
started
=
0
started
=
0
...
@@ -348,24 +347,24 @@ class SlapOSMasterCommunicator(object):
...
@@ -348,24 +347,24 @@ class SlapOSMasterCommunicator(object):
"""
"""
Wait for 'max_time' an instance specific state
Wait for 'max_time' an instance specific state
"""
"""
self
.
_logger
(
"Waiting for instance state: %s"
%
state
)
logger
.
info
(
"Waiting for instance state: %s"
,
state
)
start_time
=
time
.
time
()
start_time
=
time
.
time
()
while
(
not
self
.
_getInstanceState
()
==
state
while
(
not
self
.
_getInstanceState
()
==
state
and
(
max_time
>
(
time
.
time
()
-
start_time
))):
and
(
max_time
>
(
time
.
time
()
-
start_time
))):
self
.
_logger
(
"Instance(s) not in %s state yet."
%
state
)
logger
.
info
(
"Instance(s) not in %s state yet."
,
state
)
self
.
_logger
(
"Current state: %s"
%
self
.
_getInstanceState
())
logger
.
info
(
"Current state: %s"
,
self
.
_getInstanceState
())
time
.
sleep
(
15
)
time
.
sleep
(
15
)
if
(
time
.
time
()
-
start_time
)
>
max_time
:
if
(
time
.
time
()
-
start_time
)
>
max_time
:
error_message
=
"Instance '%s' not '%s' after %s seconds"
%
(
instance_title
,
state
,
str
(
time
.
time
()
-
start_time
))
error_message
=
"Instance '%s' not '%s' after %s seconds"
%
(
instance_title
,
state
,
str
(
time
.
time
()
-
start_time
))
return
{
'error_message'
:
error_message
}
return
{
'error_message'
:
error_message
}
self
.
_logger
(
"Instance correctly '%s' after %s seconds."
%
(
state
,
str
(
time
.
time
()
-
start_time
)))
logger
.
info
(
"Instance correctly '%s' after %s seconds."
,
state
,
time
.
time
()
-
start_time
)
return
{
'error_message'
:
None
}
return
{
'error_message'
:
None
}
class
SlapOSTester
(
SlapOSMasterCommunicator
):
class
SlapOSTester
(
SlapOSMasterCommunicator
):
def
__init__
(
self
,
def
__init__
(
self
,
name
,
name
,
logger
,
slap
,
slap
,
slap_order
,
slap_order
,
slap_supply
,
slap_supply
,
...
@@ -374,7 +373,7 @@ class SlapOSTester(SlapOSMasterCommunicator):
...
@@ -374,7 +373,7 @@ class SlapOSTester(SlapOSMasterCommunicator):
request_kw
=
None
request_kw
=
None
):
):
super
(
SlapOSTester
,
self
).
__init__
(
super
(
SlapOSTester
,
self
).
__init__
(
slap
,
slap_supply
,
slap_order
,
url
,
logger
)
slap
,
slap_supply
,
slap_order
,
url
)
self
.
name
=
name
self
.
name
=
name
self
.
computer_guid
=
computer_guid
self
.
computer_guid
=
computer_guid
...
@@ -413,9 +412,10 @@ class SlapOSTester(SlapOSMasterCommunicator):
...
@@ -413,9 +412,10 @@ class SlapOSTester(SlapOSMasterCommunicator):
def
waitInstanceStarted
(
self
,
instance_title
):
def
waitInstanceStarted
(
self
,
instance_title
):
error_message
=
self
.
_waitInstance
(
instance_title
,
INSTANCE_STATE_STARTED
)[
"error_message"
]
error_message
=
self
.
_waitInstance
(
instance_title
,
INSTANCE_STATE_STARTED
)[
"error_message"
]
if
error_message
is
not
None
:
if
error_message
is
not
None
:
self
.
_logger
(
error_message
)
logger
.
error
(
error_message
)
self
.
_logger
(
"Do you use instance state propagation in your project?"
)
logger
.
error
(
"Do you use instance state propagation in your project?"
)
self
.
_logger
(
"Instance '%s' will be stopped and test aborted."
%
instance_title
)
logger
.
error
(
"Instance '%s' will be stopped and test aborted."
,
instance_title
)
self
.
requestInstanceStop
()
self
.
requestInstanceStop
()
time
.
sleep
(
60
)
time
.
sleep
(
60
)
raise
ValueError
(
error_message
)
raise
ValueError
(
error_message
)
...
@@ -423,15 +423,15 @@ class SlapOSTester(SlapOSMasterCommunicator):
...
@@ -423,15 +423,15 @@ class SlapOSTester(SlapOSMasterCommunicator):
def
waitInstanceStopped
(
self
,
instance_title
):
def
waitInstanceStopped
(
self
,
instance_title
):
error_message
=
self
.
_waitInstance
(
instance_title
,
INSTANCE_STATE_STOPPED
)[
"error_message"
]
error_message
=
self
.
_waitInstance
(
instance_title
,
INSTANCE_STATE_STOPPED
)[
"error_message"
]
if
error_message
is
not
None
:
if
error_message
is
not
None
:
self
.
_logge
r
(
error_message
)
logger
.
erro
r
(
error_message
)
self
.
_logge
r
(
"Do you use instance state propagation in your project?"
)
logger
.
erro
r
(
"Do you use instance state propagation in your project?"
)
raise
ValueError
(
error_message
)
raise
ValueError
(
error_message
)
def
waitInstanceDestroyed
(
self
,
instance_title
):
def
waitInstanceDestroyed
(
self
,
instance_title
):
error_message
=
self
.
_waitInstance
(
instance_title
,
INSTANCE_STATE_DESTROYED
)[
"error_message"
]
error_message
=
self
.
_waitInstance
(
instance_title
,
INSTANCE_STATE_DESTROYED
)[
"error_message"
]
if
error_message
is
not
None
:
if
error_message
is
not
None
:
self
.
_logge
r
(
error_message
)
logger
.
erro
r
(
error_message
)
self
.
_logge
r
(
"Do you use instance state propagation in your project?"
)
logger
.
erro
r
(
"Do you use instance state propagation in your project?"
)
raise
ValueError
(
error_message
)
raise
ValueError
(
error_message
)
class
SoftwareReleaseTester
(
SlapOSTester
):
class
SoftwareReleaseTester
(
SlapOSTester
):
...
@@ -439,7 +439,6 @@ class SoftwareReleaseTester(SlapOSTester):
...
@@ -439,7 +439,6 @@ class SoftwareReleaseTester(SlapOSTester):
def
__init__
(
self
,
def
__init__
(
self
,
name
,
name
,
logger
,
slap
,
slap
,
slap_order
,
slap_order
,
slap_supply
,
slap_supply
,
...
@@ -450,7 +449,7 @@ class SoftwareReleaseTester(SlapOSTester):
...
@@ -450,7 +449,7 @@ class SoftwareReleaseTester(SlapOSTester):
instance_timeout
=
3600
,
instance_timeout
=
3600
,
):
):
super
(
SoftwareReleaseTester
,
self
).
__init__
(
super
(
SoftwareReleaseTester
,
self
).
__init__
(
name
,
logger
,
slap
,
slap_order
,
slap_supply
,
url
,
computer_guid
,
request_kw
)
name
,
slap
,
slap_order
,
slap_supply
,
url
,
computer_guid
,
request_kw
)
self
.
state
=
TESTER_STATE_INITIAL
self
.
state
=
TESTER_STATE_INITIAL
self
.
transition_dict
=
{
self
.
transition_dict
=
{
...
@@ -536,7 +535,7 @@ class SoftwareReleaseTester(SlapOSTester):
...
@@ -536,7 +535,7 @@ class SoftwareReleaseTester(SlapOSTester):
"""
"""
Interrupt a running test sequence, putting it in idle state.
Interrupt a running test sequence, putting it in idle state.
"""
"""
self
.
_logger
(
'Invoking TearDown for %s@%s'
%
(
self
.
url
,
self
.
name
)
)
logger
.
info
(
'Invoking TearDown for %s@%s'
,
self
.
url
,
self
.
name
)
if
self
.
request_kw
is
not
None
:
if
self
.
request_kw
is
not
None
:
self
.
_request
(
INSTANCE_STATE_DESTROYED
)
self
.
_request
(
INSTANCE_STATE_DESTROYED
)
if
self
.
computer_guid
is
not
None
:
if
self
.
computer_guid
is
not
None
:
...
@@ -548,7 +547,7 @@ class SoftwareReleaseTester(SlapOSTester):
...
@@ -548,7 +547,7 @@ class SoftwareReleaseTester(SlapOSTester):
Check for missed deadlines (-> test failure), conditions for moving to
Check for missed deadlines (-> test failure), conditions for moving to
next state, and actually moving to next state (executing its payload).
next state, and actually moving to next state (executing its payload).
"""
"""
self
.
_logger
(
'[DEBUG]
TIC'
)
logger
.
debug
(
'
TIC'
)
deadline
=
self
.
deadline
deadline
=
self
.
deadline
if
deadline
<
now
and
deadline
is
not
None
:
if
deadline
<
now
and
deadline
is
not
None
:
...
@@ -562,7 +561,7 @@ class SoftwareReleaseTester(SlapOSTester):
...
@@ -562,7 +561,7 @@ class SoftwareReleaseTester(SlapOSTester):
instance_state
is
None
or
instance_state
is
None
or
instance_state
==
self
.
_getInstanceState
()):
instance_state
==
self
.
_getInstanceState
()):
self
.
_logger
(
'[DEBUG]
Going to state %s (%r)'
,
next_state
,
instance_state
)
logger
.
debug
(
'
Going to state %s (%r)'
,
next_state
,
instance_state
)
if
next_state
is
None
:
if
next_state
is
None
:
return
None
return
None
...
...
erp5/util/testnode/UnitTestRunner.py
View file @
199e7247
...
@@ -27,6 +27,7 @@
...
@@ -27,6 +27,7 @@
import
os
import
os
import
glob
import
glob
import
json
import
json
from
.
import
logger
from
.ProcessManager
import
SubprocessError
from
.ProcessManager
import
SubprocessError
from
.SlapOSControler
import
SlapOSControler
from
.SlapOSControler
import
SlapOSControler
from
.Utils
import
createFolder
from
.Utils
import
createFolder
...
@@ -49,21 +50,20 @@ class UnitTestRunner(object):
...
@@ -49,21 +50,20 @@ class UnitTestRunner(object):
"""
"""
return
SlapOSControler
(
return
SlapOSControler
(
working_directory
,
working_directory
,
self
.
testnode
.
config
,
self
.
testnode
.
config
)
self
.
testnode
.
log
)
def
_prepareSlapOS
(
self
,
working_directory
,
slapos_instance
,
log
,
def
_prepareSlapOS
(
self
,
working_directory
,
slapos_instance
,
create_partition
=
1
,
software_path_list
=
None
,
**
kw
):
create_partition
=
1
,
software_path_list
=
None
,
**
kw
):
"""
"""
Launch slapos to build software and partitions
Launch slapos to build software and partitions
"""
"""
slapproxy_log
=
os
.
path
.
join
(
self
.
testnode
.
config
[
'log_directory'
],
slapproxy_log
=
os
.
path
.
join
(
self
.
testnode
.
config
[
'log_directory'
],
'slapproxy.log'
)
'slapproxy.log'
)
log
(
'Configured slapproxy log to %r'
,
slapproxy_log
)
log
ger
.
debug
(
'Configured slapproxy log to %r'
,
slapproxy_log
)
reset_software
=
slapos_instance
.
retry_software_count
>
10
reset_software
=
slapos_instance
.
retry_software_count
>
10
if
reset_software
:
if
reset_software
:
slapos_instance
.
retry_software_count
=
0
slapos_instance
.
retry_software_count
=
0
log
(
'testnode, retry_software_count
: %r'
,
log
ger
.
info
(
'testnode, retry_software_count
: %r'
,
slapos_instance
.
retry_software_count
)
slapos_instance
.
retry_software_count
)
# XXX Create a new controler because working_directory can be
# XXX Create a new controler because working_directory can be
...
@@ -80,12 +80,12 @@ class UnitTestRunner(object):
...
@@ -80,12 +80,12 @@ class UnitTestRunner(object):
method_list
.
append
(
"runComputerPartition"
)
method_list
.
append
(
"runComputerPartition"
)
for
method_name
in
method_list
:
for
method_name
in
method_list
:
slapos_method
=
getattr
(
slapos_controler
,
method_name
)
slapos_method
=
getattr
(
slapos_controler
,
method_name
)
log
(
"Before status_dict = slapos_method(...)"
)
log
ger
.
debug
(
"Before status_dict = slapos_method(...)"
)
status_dict
=
slapos_method
(
self
.
testnode
.
config
,
status_dict
=
slapos_method
(
self
.
testnode
.
config
,
environment
=
self
.
testnode
.
config
[
'environment'
],
environment
=
self
.
testnode
.
config
[
'environment'
],
**
kw
)
**
kw
)
log
(
status_dict
)
log
ger
.
info
(
status_dict
)
log
(
"After status_dict = slapos_method(...)"
)
log
ger
.
debug
(
"After status_dict = slapos_method(...)"
)
if
status_dict
[
'status_code'
]
!=
0
:
if
status_dict
[
'status_code'
]
!=
0
:
slapos_instance
.
retry
=
True
slapos_instance
.
retry
=
True
slapos_instance
.
retry_software_count
+=
1
slapos_instance
.
retry_software_count
+=
1
...
@@ -103,7 +103,7 @@ class UnitTestRunner(object):
...
@@ -103,7 +103,7 @@ class UnitTestRunner(object):
# instance. This is a hack which must be removed.
# instance. This is a hack which must be removed.
config
=
self
.
testnode
.
config
config
=
self
.
testnode
.
config
return
self
.
_prepareSlapOS
(
test_node_slapos
.
working_directory
,
return
self
.
_prepareSlapOS
(
test_node_slapos
.
working_directory
,
test_node_slapos
,
self
.
testnode
.
log
,
create_partition
=
0
,
test_node_slapos
,
create_partition
=
0
,
software_path_list
=
config
.
get
(
"software_list"
),
software_path_list
=
config
.
get
(
"software_list"
),
cluster_configuration
=
{
cluster_configuration
=
{
'report-url'
:
config
.
get
(
"report-url"
,
""
),
'report-url'
:
config
.
get
(
"report-url"
,
""
),
...
@@ -116,7 +116,7 @@ class UnitTestRunner(object):
...
@@ -116,7 +116,7 @@ class UnitTestRunner(object):
Build softwares needed by testsuites
Build softwares needed by testsuites
"""
"""
return
self
.
_prepareSlapOS
(
node_test_suite
.
working_directory
,
return
self
.
_prepareSlapOS
(
node_test_suite
.
working_directory
,
node_test_suite
,
self
.
testnode
.
log
,
node_test_suite
,
software_path_list
=
[
node_test_suite
.
custom_profile_path
],
software_path_list
=
[
node_test_suite
.
custom_profile_path
],
cluster_configuration
=
{
'_'
:
json
.
dumps
(
node_test_suite
.
cluster_configuration
)})
cluster_configuration
=
{
'_'
:
json
.
dumps
(
node_test_suite
.
cluster_configuration
)})
...
@@ -124,7 +124,7 @@ class UnitTestRunner(object):
...
@@ -124,7 +124,7 @@ class UnitTestRunner(object):
return
self
.
_getSlapOSControler
(
return
self
.
_getSlapOSControler
(
node_test_suite
.
working_directory
).
instance_root
node_test_suite
.
working_directory
).
instance_root
def
runTestSuite
(
self
,
node_test_suite
,
portal_url
,
log
=
None
):
def
runTestSuite
(
self
,
node_test_suite
,
portal_url
):
config
=
self
.
testnode
.
config
config
=
self
.
testnode
.
config
run_test_suite_path_list
=
glob
.
glob
(
run_test_suite_path_list
=
glob
.
glob
(
self
.
getInstanceRoot
(
node_test_suite
)
+
"/*/bin/runTestSuite"
)
self
.
getInstanceRoot
(
node_test_suite
)
+
"/*/bin/runTestSuite"
)
...
...
erp5/util/testnode/Updater.py
View file @
199e7247
...
@@ -28,7 +28,7 @@ import errno
...
@@ -28,7 +28,7 @@ import errno
import
os
import
os
import
re
import
re
import
shutil
import
shutil
import
sys
from
.
import
logger
from
.ProcessManager
import
SubprocessError
from
.ProcessManager
import
SubprocessError
SVN_UP_REV
=
re
.
compile
(
r'^(?:At|Updated to) revision (\
d+).$
')
SVN_UP_REV
=
re
.
compile
(
r'^(?:At|Updated to) revision (\
d+).$
')
...
@@ -42,10 +42,9 @@ class Updater(object):
...
@@ -42,10 +42,9 @@ class Updater(object):
_git_cache = {}
_git_cache = {}
def __init__(self, repository_path,
log,
revision=None, git_binary='
git
',
def __init__(self, repository_path, revision=None, git_binary='
git
',
branch=None, realtime_output=True, process_manager=None, url=None,
branch=None, realtime_output=True, process_manager=None, url=None,
working_directory=None):
working_directory=None):
self.log = log
self.revision = revision
self.revision = revision
self._path_list = []
self._path_list = []
self.branch = branch
self.branch = branch
...
@@ -111,16 +110,16 @@ class Updater(object):
...
@@ -111,16 +110,16 @@ class Updater(object):
git_repository_path = os.path.join(self.getRepositoryPath(), '
.
git
')
git_repository_path = os.path.join(self.getRepositoryPath(), '
.
git
')
name = os.path.basename(os.path.normpath(self.getRepositoryPath()))
name = os.path.basename(os.path.normpath(self.getRepositoryPath()))
git_repository_link_path = os.path.join(self.getRepositoryPath(), '
%
s
.
git
' %name)
git_repository_link_path = os.path.join(self.getRepositoryPath(), '
%
s
.
git
' %name)
self.lo
g("checking link %s -> %s..",
logger.debu
g("checking link %s -> %s..",
git_repository_link_path, git_repository_path)
git_repository_link_path, git_repository_path)
if ( not os.path.lexists(git_repository_link_path) and
\
if ( not os.path.lexists(git_repository_link_path) and
\
not os.path.exists(git_repository_link_path) ):
not os.path.exists(git_repository_link_path) ):
try:
try:
os.symlink(git_repository_path, git_repository_link_path)
os.symlink(git_repository_path, git_repository_link_path)
self.lo
g("link: %s -> %s created",
logger.debu
g("link: %s -> %s created",
git_repository_link_path, git_repository_path)
git_repository_link_path, git_repository_path)
except OSError:
except OSError:
self.log
("Cannot create link from %s -> %s",
logger.error
("Cannot create link from %s -> %s",
git_repository_link_path, git_repository_path)
git_repository_link_path, git_repository_path)
def _git_find_rev(self, ref):
def _git_find_rev(self, ref):
...
@@ -148,7 +147,7 @@ class Updater(object):
...
@@ -148,7 +147,7 @@ class Updater(object):
raise NotImplementedError
raise NotImplementedError
def deleteRepository(self):
def deleteRepository(self):
self.log
("Wrong repository or wrong url, deleting repos %s",
logger.info
("Wrong repository or wrong url, deleting repos %s",
self.repository_path)
self.repository_path)
shutil.rmtree(self.repository_path)
shutil.rmtree(self.repository_path)
...
@@ -162,7 +161,7 @@ class Updater(object):
...
@@ -162,7 +161,7 @@ class Updater(object):
if remote_url == self.url:
if remote_url == self.url:
correct_url = True
correct_url = True
except SubprocessError:
except SubprocessError:
self.log("SubprocessError", exc_info=sys.exc_info()
)
logger.exception(""
)
if not(correct_url):
if not(correct_url):
self.deleteRepository()
self.deleteRepository()
if not os.path.exists(self.repository_path):
if not os.path.exists(self.repository_path):
...
...
erp5/util/testnode/__init__.py
View file @
199e7247
...
@@ -30,9 +30,12 @@ import logging
...
@@ -30,9 +30,12 @@ import logging
import
logging.handlers
import
logging.handlers
import
os
import
os
from
.testnode
import
TestNode
log_formatter
=
logging
.
Formatter
(
'%(asctime)s %(levelname)-8s %(message)s'
)
logger
=
logging
.
getLogger
(
__name__
)
logger
.
setLevel
(
logging
.
DEBUG
)
def
main
(
*
args
):
def
main
(
*
args
):
from
.testnode
import
TestNode
parser
=
argparse
.
ArgumentParser
()
parser
=
argparse
.
ArgumentParser
()
parser
.
add_argument
(
"configuration_file"
,
nargs
=
1
,
type
=
argparse
.
FileType
(),
parser
.
add_argument
(
"configuration_file"
,
nargs
=
1
,
type
=
argparse
.
FileType
(),
help
=
"Configuration file."
)
help
=
"Configuration file."
)
...
@@ -43,29 +46,24 @@ def main(*args):
...
@@ -43,29 +46,24 @@ def main(*args):
parsed_argument
=
parser
.
parse_args
(
list
(
args
))
parsed_argument
=
parser
.
parse_args
(
list
(
args
))
else
:
else
:
parsed_argument
=
parser
.
parse_args
()
parsed_argument
=
parser
.
parse_args
()
logger_format
=
'%(asctime)s %(name)-13s: %(levelname)-8s %(message)s'
formatter
=
logging
.
Formatter
(
logger_format
)
logging
.
basicConfig
(
level
=
logging
.
INFO
,
format
=
logger_format
)
logger
=
logging
.
getLogger
(
'erp5testnode'
)
CONFIG
=
{
'logger'
:
logger
.
info
,
'partition_reference'
:
'test0'
,
}
if
parsed_argument
.
console
or
parsed_argument
.
logfile
:
if
parsed_argument
.
console
or
parsed_argument
.
logfile
:
root
=
logging
.
getLogger
()
def
addHandler
(
handler
):
handler
.
setFormatter
(
log_formatter
)
root
.
addHandler
(
handler
)
if
parsed_argument
.
console
:
if
parsed_argument
.
console
:
logger
.
addHandler
(
logging
.
StreamHandler
())
addHandler
(
logging
.
StreamHandler
())
logger
.
info
(
'Activated console output.'
)
if
parsed_argument
.
logfile
:
if
parsed_argument
.
logfile
:
file_handler
=
logging
.
handlers
.
RotatingFileHandler
(
addHandler
(
logging
.
handlers
.
RotatingFileHandler
(
filename
=
parsed_argument
.
logfile
,
filename
=
parsed_argument
.
logfile
,
maxBytes
=
20000000
,
backupCount
=
4
)
maxBytes
=
20000000
,
backupCount
=
4
))
file_handler
.
setFormatter
(
formatter
)
logger
.
addHandler
(
file_handler
)
logger
.
info
(
'Activated logfile %r output'
,
parsed_argument
.
logfile
)
CONFIG
[
'log_file'
]
=
parsed_argument
.
logfile
else
:
else
:
logger
.
addHandler
(
logging
.
NullHandler
())
logger
.
disable
(
logging
.
CRITICAL
)
CONFIG
=
{
'partition_reference'
:
'test0'
,
}
config
=
ConfigParser
.
SafeConfigParser
()
config
=
ConfigParser
.
SafeConfigParser
()
# do not change case of option keys
# do not change case of option keys
config
.
optionxform
=
str
config
.
optionxform
=
str
...
@@ -107,5 +105,4 @@ def main(*args):
...
@@ -107,5 +105,4 @@ def main(*args):
CONFIG
[
'software_list'
]
=
filter
(
None
,
CONFIG
[
'software_list'
]
=
filter
(
None
,
config
.
get
(
"software_list"
,
"path_list"
).
split
(
","
))
config
.
get
(
"software_list"
,
"path_list"
).
split
(
","
))
testnode
=
TestNode
(
logger
.
info
,
CONFIG
)
TestNode
(
CONFIG
).
run
()
testnode
.
run
()
erp5/util/testnode/testnode.py
View file @
199e7247
...
@@ -25,13 +25,13 @@
...
@@ -25,13 +25,13 @@
#
#
##############################################################################
##############################################################################
import
os
import
os
import
sys
import
json
import
json
import
time
import
time
import
shutil
import
shutil
import
logging
import
logging
from
contextlib
import
contextmanager
from
slapos.slap.slap
import
ConnectionError
from
slapos.slap.slap
import
ConnectionError
from
.
import
logger
,
log_formatter
from
.ProcessManager
import
SubprocessError
,
ProcessManager
,
CancellationError
from
.ProcessManager
import
SubprocessError
,
ProcessManager
,
CancellationError
from
subprocess
import
CalledProcessError
from
subprocess
import
CalledProcessError
from
.Updater
import
Updater
from
.Updater
import
Updater
...
@@ -47,12 +47,6 @@ MAX_TEMP_TIME = 0.01 # time in days we should keep temp files
...
@@ -47,12 +47,6 @@ MAX_TEMP_TIME = 0.01 # time in days we should keep temp files
PROFILE_PATH_KEY
=
'profile_path'
PROFILE_PATH_KEY
=
'profile_path'
class
DummyLogger
(
object
):
def
__init__
(
self
,
func
):
for
name
in
(
'trace'
,
'debug'
,
'info'
,
'warn'
,
'warning'
,
'error'
,
'critical'
,
'fatal'
):
setattr
(
self
,
name
,
func
)
test_type_registry
=
{
test_type_registry
=
{
'UnitTest'
:
UnitTestRunner
,
'UnitTest'
:
UnitTestRunner
,
'ScalabilityTest'
:
ScalabilityTestRunner
,
'ScalabilityTest'
:
ScalabilityTestRunner
,
...
@@ -60,12 +54,10 @@ test_type_registry = {
...
@@ -60,12 +54,10 @@ test_type_registry = {
class
TestNode
(
object
):
class
TestNode
(
object
):
def
__init__
(
self
,
log
,
config
,
max_log_time
=
MAX_LOG_TIME
,
def
__init__
(
self
,
config
,
max_log_time
=
MAX_LOG_TIME
,
max_temp_time
=
MAX_TEMP_TIME
):
max_temp_time
=
MAX_TEMP_TIME
):
self
.
testnode_log
=
log
self
.
log
=
log
self
.
config
=
config
or
{}
self
.
config
=
config
or
{}
self
.
process_manager
=
ProcessManager
(
log
)
self
.
process_manager
=
ProcessManager
()
self
.
working_directory
=
config
[
'working_directory'
]
self
.
working_directory
=
config
[
'working_directory'
]
self
.
node_test_suite_dict
=
{}
self
.
node_test_suite_dict
=
{}
self
.
file_handler
=
None
self
.
file_handler
=
None
...
@@ -80,7 +72,7 @@ class TestNode(object):
...
@@ -80,7 +72,7 @@ class TestNode(object):
for
reference
in
reference_set
:
for
reference
in
reference_set
:
fpath
=
os
.
path
.
join
(
self
.
working_directory
,
reference
)
fpath
=
os
.
path
.
join
(
self
.
working_directory
,
reference
)
self
.
node_test_suite_dict
.
pop
(
reference
,
None
)
self
.
node_test_suite_dict
.
pop
(
reference
,
None
)
self
.
log
(
"testnode.purgeOldTestSuite, DELETING : %r"
,
fpath
)
logger
.
info
(
"testnode.purgeOldTestSuite, DELETING : %r"
,
fpath
)
if
os
.
path
.
isdir
(
fpath
):
if
os
.
path
.
isdir
(
fpath
):
shutil
.
rmtree
(
fpath
)
shutil
.
rmtree
(
fpath
)
else
:
else
:
...
@@ -164,7 +156,6 @@ shared = true
...
@@ -164,7 +156,6 @@ shared = true
def
updateRevisionList
(
self
,
node_test_suite
):
def
updateRevisionList
(
self
,
node_test_suite
):
config
=
self
.
config
config
=
self
.
config
log
=
self
.
log
revision_list
=
[]
revision_list
=
[]
try
:
try
:
for
vcs_repository
in
node_test_suite
.
vcs_repository_list
:
for
vcs_repository
in
node_test_suite
.
vcs_repository_list
:
...
@@ -173,54 +164,38 @@ shared = true
...
@@ -173,54 +164,38 @@ shared = true
branch
=
vcs_repository
.
get
(
'branch'
)
branch
=
vcs_repository
.
get
(
'branch'
)
# Make sure we have local repository
# Make sure we have local repository
updater
=
Updater
(
repository_path
,
git_binary
=
config
[
'git_binary'
],
updater
=
Updater
(
repository_path
,
git_binary
=
config
[
'git_binary'
],
branch
=
branch
,
log
=
log
,
process_manager
=
self
.
process_manager
,
branch
=
branch
,
process_manager
=
self
.
process_manager
,
working_directory
=
node_test_suite
.
working_directory
,
working_directory
=
node_test_suite
.
working_directory
,
url
=
vcs_repository
[
"url"
])
url
=
vcs_repository
[
"url"
])
updater
.
checkout
()
updater
.
checkout
()
revision_list
.
append
((
repository_id
,
updater
.
getRevision
()))
revision_list
.
append
((
repository_id
,
updater
.
getRevision
()))
except
SubprocessError
:
except
SubprocessError
:
log
(
"Error while getting repository, ignoring this test suite"
,
log
ger
.
warning
(
"Error while getting repository, ignoring this test suite"
,
exc_info
=
1
)
exc_info
=
1
)
return
False
return
False
node_test_suite
.
revision_list
=
revision_list
node_test_suite
.
revision_list
=
revision_list
return
True
return
True
def
registerSuiteLog
(
self
,
test_result
,
node_test_suite
):
@
contextmanager
"""
def
suiteLog
(
self
,
node_test_suite
):
Create a log dedicated for the test suite,
and register the url to master node.
"""
suite_log_path
,
folder_id
=
node_test_suite
.
createSuiteLog
()
suite_log_path
,
folder_id
=
node_test_suite
.
createSuiteLog
()
self
.
_initializeSuiteLog
(
suite_log_path
)
handler
=
logging
.
FileHandler
(
filename
=
suite_log_path
)
# TODO make the path into url
handler
.
setFormatter
(
log_formatter
)
test_result
.
reportStatus
(
'LOG url'
,
"%s/%s"
%
(
self
.
config
.
get
(
'httpd_url'
),
logger
.
info
(
'Suite logfile: %s'
,
suite_log_path
)
folder_id
),
''
)
try
:
self
.
log
(
"going to switch to log %r"
,
suite_log_path
)
logger
.
propagate
=
False
self
.
process_manager
.
log
=
self
.
log
=
self
.
suite_log
logger
.
addHandler
(
handler
)
return
suite_log_path
yield
folder_id
finally
:
def
_initializeSuiteLog
(
self
,
suite_log_path
):
logger
.
propagate
=
True
# remove previous handlers
logger
.
removeHandler
(
handler
)
logger
=
logging
.
getLogger
(
'testsuite'
)
if
self
.
file_handler
is
not
None
:
logger
.
removeHandler
(
self
.
file_handler
)
# and replace it with new handler
logger_format
=
'%(asctime)s %(name)-13s: %(levelname)-8s %(message)s'
formatter
=
logging
.
Formatter
(
logger_format
)
logging
.
basicConfig
(
level
=
logging
.
INFO
,
format
=
logger_format
)
self
.
file_handler
=
logging
.
FileHandler
(
filename
=
suite_log_path
)
self
.
file_handler
.
setFormatter
(
formatter
)
logger
.
addHandler
(
self
.
file_handler
)
logger
.
info
(
'Activated logfile %r output'
,
suite_log_path
)
self
.
suite_log
=
logger
.
info
def
checkRevision
(
self
,
test_result
,
node_test_suite
):
def
checkRevision
(
self
,
test_result
,
node_test_suite
):
if
node_test_suite
.
revision
==
test_result
.
revision
:
if
node_test_suite
.
revision
==
test_result
.
revision
:
return
return
log
=
self
.
log
logger
.
info
(
'Disagreement on tested revision, checking out: %r != %r'
,
log
(
'Disagreement on tested revision, checking out: %r != %r'
,
node_test_suite
.
revision
,
test_result
.
revision
)
node_test_suite
.
revision
,
test_result
.
revision
)
updater_kw
=
dict
(
git_binary
=
self
.
config
[
'git_binary'
],
log
=
log
,
updater_kw
=
dict
(
git_binary
=
self
.
config
[
'git_binary'
],
process_manager
=
self
.
process_manager
)
process_manager
=
self
.
process_manager
)
revision_list
=
[]
revision_list
=
[]
for
i
,
revision
in
enumerate
(
test_result
.
revision
.
split
(
','
)):
for
i
,
revision
in
enumerate
(
test_result
.
revision
.
split
(
','
)):
...
@@ -244,7 +219,7 @@ shared = true
...
@@ -244,7 +219,7 @@ shared = true
folder_path
=
os
.
path
.
join
(
log_directory
,
log_folder
)
folder_path
=
os
.
path
.
join
(
log_directory
,
log_folder
)
if
os
.
path
.
isdir
(
folder_path
):
if
os
.
path
.
isdir
(
folder_path
):
if
os
.
stat
(
folder_path
).
st_mtime
<
prune_time
:
if
os
.
stat
(
folder_path
).
st_mtime
<
prune_time
:
self
.
lo
g
(
"deleting log directory %r"
,
folder_path
)
logger
.
debu
g
(
"deleting log directory %r"
,
folder_path
)
shutil
.
rmtree
(
folder_path
)
shutil
.
rmtree
(
folder_path
)
def
_cleanupTemporaryFiles
(
self
):
def
_cleanupTemporaryFiles
(
self
):
...
@@ -262,22 +237,21 @@ shared = true
...
@@ -262,22 +237,21 @@ shared = true
try
:
try
:
stat
=
os
.
stat
(
folder_path
)
stat
=
os
.
stat
(
folder_path
)
if
stat
.
st_uid
==
user_id
and
stat
.
st_mtime
<
prune_time
:
if
stat
.
st_uid
==
user_id
and
stat
.
st_mtime
<
prune_time
:
self
.
lo
g
(
"deleting temp directory %r"
,
folder_path
)
logger
.
debu
g
(
"deleting temp directory %r"
,
folder_path
)
if
os
.
path
.
isdir
(
folder_path
):
if
os
.
path
.
isdir
(
folder_path
):
shutil
.
rmtree
(
folder_path
)
shutil
.
rmtree
(
folder_path
)
else
:
else
:
os
.
remove
(
folder_path
)
os
.
remove
(
folder_path
)
except
OSError
:
except
OSError
:
self
.
lo
g
(
"_cleanupTemporaryFiles exception"
,
exc_info
=
1
)
logger
.
warnin
g
(
"_cleanupTemporaryFiles exception"
,
exc_info
=
1
)
def
cleanUp
(
self
):
def
cleanUp
(
self
):
self
.
lo
g
(
'Testnode.cleanUp'
)
logger
.
debu
g
(
'Testnode.cleanUp'
)
self
.
process_manager
.
killPreviousRun
()
self
.
process_manager
.
killPreviousRun
()
self
.
_cleanupLog
()
self
.
_cleanupLog
()
self
.
_cleanupTemporaryFiles
()
self
.
_cleanupTemporaryFiles
()
def
run
(
self
):
def
run
(
self
):
log
=
self
.
log
config
=
self
.
config
config
=
self
.
config
portal_url
=
config
[
'test_suite_master_url'
]
portal_url
=
config
[
'test_suite_master_url'
]
test_node_slapos
=
SlapOSInstance
(
config
[
'slapos_directory'
])
test_node_slapos
=
SlapOSInstance
(
config
[
'slapos_directory'
])
...
@@ -286,11 +260,10 @@ shared = true
...
@@ -286,11 +260,10 @@ shared = true
test_result
=
None
test_result
=
None
try
:
try
:
node_test_suite
=
None
node_test_suite
=
None
self
.
log
=
self
.
process_manager
.
log
=
self
.
testnode_log
self
.
cleanUp
()
self
.
cleanUp
()
begin
=
time
.
time
()
begin
=
time
.
time
()
taskdistributor
=
taskdistribution
.
TaskDistributor
(
taskdistributor
=
taskdistribution
.
TaskDistributor
(
portal_url
,
logger
=
DummyLogger
(
log
)
)
portal_url
,
logger
=
logger
)
self
.
test_suite_portal
=
taskdistributor
# XXX ScalabilityTest
self
.
test_suite_portal
=
taskdistributor
# XXX ScalabilityTest
node_configuration
=
taskdistributor
.
subscribeNode
(
node_configuration
=
taskdistributor
.
subscribeNode
(
node_title
=
config
[
'test_node_title'
],
node_title
=
config
[
'test_node_title'
],
...
@@ -302,7 +275,7 @@ shared = true
...
@@ -302,7 +275,7 @@ shared = true
'process_timeout'
in
node_configuration
\
'process_timeout'
in
node_configuration
\
and
node_configuration
[
'process_timeout'
]
is
not
None
:
and
node_configuration
[
'process_timeout'
]
is
not
None
:
process_timeout
=
node_configuration
[
'process_timeout'
]
process_timeout
=
node_configuration
[
'process_timeout'
]
log
(
'Received and using process timeout from master: %i'
,
log
ger
.
info
(
'Received and using process timeout from master: %i'
,
process_timeout
)
process_timeout
)
self
.
process_manager
.
max_timeout
=
process_timeout
self
.
process_manager
.
max_timeout
=
process_timeout
test_suite_data
=
taskdistributor
.
startTestSuite
(
test_suite_data
=
taskdistributor
.
startTestSuite
(
...
@@ -312,22 +285,23 @@ shared = true
...
@@ -312,22 +285,23 @@ shared = true
# Backward compatiblity
# Backward compatiblity
test_suite_data
=
json
.
loads
(
test_suite_data
)
test_suite_data
=
json
.
loads
(
test_suite_data
)
test_suite_data
=
deunicodeData
(
test_suite_data
)
test_suite_data
=
deunicodeData
(
test_suite_data
)
log
(
"Got following test suite data from master : %r"
,
log
ger
.
info
(
"Got following test suite data from master : %r"
,
test_suite_data
)
test_suite_data
)
try
:
try
:
my_test_type
=
taskdistributor
.
getTestType
()
my_test_type
=
taskdistributor
.
getTestType
()
except
Exception
:
except
Exception
:
log
(
"testnode, error during requesting getTestType() method
"
log
ger
.
warning
(
"testnode, error during requesting getTestType()
"
"
from the distributor."
)
" method
from the distributor."
)
raise
raise
# Select runner according to the test type
# Select runner according to the test type
try
:
try
:
runner_class
=
test_type_registry
[
my_test_type
]
runner_class
=
test_type_registry
[
my_test_type
]
except
KeyError
:
except
KeyError
:
log
(
"testnode, Runner type %s not implemented."
,
my_test_type
)
logger
.
warning
(
"testnode, Runner type %s not implemented."
,
my_test_type
)
raise
NotImplementedError
raise
NotImplementedError
runner
=
runner_class
(
self
)
runner
=
runner_class
(
self
)
log
(
"Type of current test is %s"
,
my_test_type
)
log
ger
.
info
(
"Type of current test is %s"
,
my_test_type
)
# master testnode gets test_suites, slaves get nothing
# master testnode gets test_suites, slaves get nothing
runner
.
prepareSlapOSForTestNode
(
test_node_slapos
)
runner
.
prepareSlapOSForTestNode
(
test_node_slapos
)
# Clean-up test suites
# Clean-up test suites
...
@@ -349,14 +323,18 @@ shared = true
...
@@ -349,14 +323,18 @@ shared = true
config
[
'test_node_title'
],
False
,
config
[
'test_node_title'
],
False
,
node_test_suite
.
test_suite_title
,
node_test_suite
.
test_suite_title
,
node_test_suite
.
project_title
)
node_test_suite
.
project_title
)
log
(
"testnode, test_result : %r"
,
test_result
)
logger
.
info
(
"testnode, test_result : %r"
,
test_result
)
if
test_result
is
not
None
:
if
test_result
is
None
:
self
.
registerSuiteLog
(
test_result
,
node_test_suite
)
self
.
cleanUp
()
# XXX not a good place to do that
continue
with
self
.
suiteLog
(
node_test_suite
)
as
suite_log_folder_name
:
test_result
.
reportStatus
(
'LOG url'
,
"%s/%s"
%
(
config
.
get
(
'httpd_url'
),
suite_log_folder_name
),
''
)
self
.
checkRevision
(
test_result
,
node_test_suite
)
self
.
checkRevision
(
test_result
,
node_test_suite
)
node_test_suite
.
edit
(
test_result
=
test_result
)
node_test_suite
.
edit
(
test_result
=
test_result
)
# get cluster configuration for this test suite, this is needed to
# get cluster configuration for this test suite, this is needed to
# know slapos parameters to user for creating instances
# know slapos parameters to user for creating instances
log
(
"Getting configuration from test suite %s"
,
log
ger
.
info
(
"Getting configuration from test suite %s"
,
node_test_suite
.
test_suite_title
)
node_test_suite
.
test_suite_title
)
generated_config
=
taskdistributor
.
generateConfiguration
(
generated_config
=
taskdistributor
.
generateConfiguration
(
node_test_suite
.
test_suite_title
)
node_test_suite
.
test_suite_title
)
...
@@ -391,15 +369,14 @@ shared = true
...
@@ -391,15 +369,14 @@ shared = true
test_result
.
reportFailure
(
test_result
.
reportFailure
(
stdout
=
error_message
stdout
=
error_message
)
)
self
.
log
(
error_message
)
logger
.
error
(
error_message
)
raise
ValueError
(
error_message
)
raise
ValueError
(
error_message
)
else
:
else
:
raise
NotImplementedError
raise
NotImplementedError
# break the loop to get latest priorities from master
# break the loop to get latest priorities from master
break
break
self
.
cleanUp
()
except
(
SubprocessError
,
CalledProcessError
,
ConnectionError
)
as
e
:
except
(
SubprocessError
,
CalledProcessError
,
ConnectionError
)
as
e
:
log
(
""
,
exc_info
=
1
)
log
ger
.
exception
(
""
)
if
test_result
is
not
None
:
if
test_result
is
not
None
:
status_dict
=
getattr
(
e
,
"status_dict"
,
None
)
or
{
status_dict
=
getattr
(
e
,
"status_dict"
,
None
)
or
{
'stderr'
:
"%s: %s"
%
(
e
.
__class__
.
__name__
,
e
)}
'stderr'
:
"%s: %s"
%
(
e
.
__class__
.
__name__
,
e
)}
...
@@ -411,7 +388,7 @@ shared = true
...
@@ -411,7 +388,7 @@ shared = true
continue
continue
except
ValueError
as
e
:
except
ValueError
as
e
:
# This could at least happens if runTestSuite is not found
# This could at least happens if runTestSuite is not found
log
(
""
,
exc_info
=
1
)
log
ger
.
exception
(
""
)
if
node_test_suite
is
not
None
:
if
node_test_suite
is
not
None
:
node_test_suite
.
retry_software_count
+=
1
node_test_suite
.
retry_software_count
+=
1
if
test_result
is
not
None
:
if
test_result
is
not
None
:
...
@@ -420,24 +397,24 @@ shared = true
...
@@ -420,24 +397,24 @@ shared = true
stderr
=
"ValueError was raised : %s"
%
(
e
,),
stderr
=
"ValueError was raised : %s"
%
(
e
,),
)
)
except
CancellationError
:
except
CancellationError
:
log
(
""
,
exc_info
=
1
)
log
ger
.
exception
(
""
)
self
.
process_manager
.
under_cancellation
=
False
self
.
process_manager
.
under_cancellation
=
False
node_test_suite
.
retry
=
True
node_test_suite
.
retry
=
True
continue
continue
self
.
cleanUp
()
self
.
cleanUp
()
sleep_time
=
120
-
(
time
.
time
()
-
begin
)
sleep_time
=
120
-
(
time
.
time
()
-
begin
)
if
sleep_time
>
0
:
if
sleep_time
>
0
:
log
(
"End of processing, going to sleep %s"
,
sleep_time
)
log
ger
.
info
(
"End of processing, going to sleep %s"
,
sleep_time
)
time
.
sleep
(
sleep_time
)
time
.
sleep
(
sleep_time
)
except
Exception
:
except
Exception
:
log
(
""
,
exc_info
=
1
)
log
ger
.
exception
(
""
)
except
:
except
:
log
(
""
,
exc_info
=
1
)
log
ger
.
exception
(
""
)
raise
raise
finally
:
finally
:
# Nice way to kill *everything* generated by run process -- process
# Nice way to kill *everything* generated by run process -- process
# groups working only in POSIX compilant systems
# groups working only in POSIX compilant systems
# Exceptions are swallowed during cleanup phase
# Exceptions are swallowed during cleanup phase
log
(
"GENERAL EXCEPTION, QUITING"
)
log
ger
.
info
(
"GENERAL EXCEPTION, QUITING"
)
self
.
cleanUp
()
self
.
cleanUp
()
log
(
"GENERAL EXCEPTION, QUITING, cleanup finished"
)
log
ger
.
info
(
"GENERAL EXCEPTION, QUITING, cleanup finished"
)
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment