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
Issues
0
Issues
0
List
Boards
Labels
Milestones
Merge Requests
0
Merge Requests
0
Analytics
Analytics
Repository
Value Stream
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Create a new issue
Commits
Issue Boards
Open sidebar
Papa Tamsir Kane
erp5
Commits
f75eb99c
Commit
f75eb99c
authored
Mar 14, 2013
by
Kazuhiko Shiozaki
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
initial copy of ZODB3-3.9.7's ZODB/scripts/analyze.py.
parent
c4b496d0
Changes
1
Show whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
143 additions
and
0 deletions
+143
-0
erp5/util/zodbanalyze/__init__.py
erp5/util/zodbanalyze/__init__.py
+143
-0
No files found.
erp5/util/zodbanalyze/__init__.py
0 → 100644
View file @
f75eb99c
#!/usr/bin/env python2.4
# Based on a transaction analyzer by Matt Kromer.
import
pickle
import
re
import
sys
import
types
from
ZODB.FileStorage
import
FileStorage
from
cStringIO
import
StringIO
class
FakeError
(
Exception
):
def
__init__
(
self
,
module
,
name
):
Exception
.
__init__
(
self
)
self
.
module
=
module
self
.
name
=
name
class
FakeUnpickler
(
pickle
.
Unpickler
):
def
find_class
(
self
,
module
,
name
):
raise
FakeError
(
module
,
name
)
class
Report
:
def
__init__
(
self
):
self
.
OIDMAP
=
{}
self
.
TYPEMAP
=
{}
self
.
TYPESIZE
=
{}
self
.
FREEMAP
=
{}
self
.
USEDMAP
=
{}
self
.
TIDS
=
0
self
.
OIDS
=
0
self
.
DBYTES
=
0
self
.
COIDS
=
0
self
.
CBYTES
=
0
self
.
FOIDS
=
0
self
.
FBYTES
=
0
def
shorten
(
s
,
n
):
l
=
len
(
s
)
if
l
<=
n
:
return
s
while
len
(
s
)
+
3
>
n
:
# account for ...
i
=
s
.
find
(
"."
)
if
i
==
-
1
:
# In the worst case, just return the rightmost n bytes
return
s
[
-
n
:]
else
:
s
=
s
[
i
+
1
:]
l
=
len
(
s
)
return
"..."
+
s
def
report
(
rep
):
print
"Processed %d records in %d transactions"
%
(
rep
.
OIDS
,
rep
.
TIDS
)
print
"Average record size is %7.2f bytes"
%
(
rep
.
DBYTES
*
1.0
/
rep
.
OIDS
)
print
(
"Average transaction size is %7.2f bytes"
%
(
rep
.
DBYTES
*
1.0
/
rep
.
TIDS
))
print
"Types used:"
fmt
=
"%-46s %7s %9s %6s %7s"
fmtp
=
"%-46s %7d %9d %5.1f%% %7.2f"
# per-class format
fmts
=
"%46s %7d %8dk %5.1f%% %7.2f"
# summary format
print
fmt
%
(
"Class Name"
,
"Count"
,
"TBytes"
,
"Pct"
,
"AvgSize"
)
print
fmt
%
(
'-'
*
46
,
'-'
*
7
,
'-'
*
9
,
'-'
*
5
,
'-'
*
7
)
typemap
=
rep
.
TYPEMAP
.
keys
()
typemap
.
sort
()
cumpct
=
0.0
for
t
in
typemap
:
pct
=
rep
.
TYPESIZE
[
t
]
*
100.0
/
rep
.
DBYTES
cumpct
+=
pct
print
fmtp
%
(
shorten
(
t
,
46
),
rep
.
TYPEMAP
[
t
],
rep
.
TYPESIZE
[
t
],
pct
,
rep
.
TYPESIZE
[
t
]
*
1.0
/
rep
.
TYPEMAP
[
t
])
print
fmt
%
(
'='
*
46
,
'='
*
7
,
'='
*
9
,
'='
*
5
,
'='
*
7
)
print
"%46s %7d %9s %6s %6.2fk"
%
(
'Total Transactions'
,
rep
.
TIDS
,
' '
,
' '
,
rep
.
DBYTES
*
1.0
/
rep
.
TIDS
/
1024.0
)
print
fmts
%
(
'Total Records'
,
rep
.
OIDS
,
rep
.
DBYTES
/
1024.0
,
cumpct
,
rep
.
DBYTES
*
1.0
/
rep
.
OIDS
)
print
fmts
%
(
'Current Objects'
,
rep
.
COIDS
,
rep
.
CBYTES
/
1024.0
,
rep
.
CBYTES
*
100.0
/
rep
.
DBYTES
,
rep
.
CBYTES
*
1.0
/
rep
.
COIDS
)
if
rep
.
FOIDS
:
print
fmts
%
(
'Old Objects'
,
rep
.
FOIDS
,
rep
.
FBYTES
/
1024.0
,
rep
.
FBYTES
*
100.0
/
rep
.
DBYTES
,
rep
.
FBYTES
*
1.0
/
rep
.
FOIDS
)
def
analyze
(
path
):
fs
=
FileStorage
(
path
,
read_only
=
1
)
fsi
=
fs
.
iterator
()
report
=
Report
()
for
txn
in
fsi
:
analyze_trans
(
report
,
txn
)
return
report
def
analyze_trans
(
report
,
txn
):
report
.
TIDS
+=
1
for
rec
in
txn
:
analyze_rec
(
report
,
rec
)
def
get_type
(
record
):
try
:
unpickled
=
FakeUnpickler
(
StringIO
(
record
.
data
)).
load
()
except
FakeError
,
err
:
return
"%s.%s"
%
(
err
.
module
,
err
.
name
)
except
:
raise
classinfo
=
unpickled
[
0
]
if
isinstance
(
classinfo
,
types
.
TupleType
):
mod
,
klass
=
classinfo
return
"%s.%s"
%
(
mod
,
klass
)
else
:
return
str
(
classinfo
)
def
analyze_rec
(
report
,
record
):
oid
=
record
.
oid
report
.
OIDS
+=
1
if
record
.
data
is
None
:
# No pickle -- aborted version or undo of object creation.
return
try
:
size
=
len
(
record
.
data
)
# Ignores various overhead
report
.
DBYTES
+=
size
if
oid
not
in
report
.
OIDMAP
:
type
=
get_type
(
record
)
report
.
OIDMAP
[
oid
]
=
type
report
.
USEDMAP
[
oid
]
=
size
report
.
COIDS
+=
1
report
.
CBYTES
+=
size
else
:
type
=
report
.
OIDMAP
[
oid
]
fsize
=
report
.
USEDMAP
[
oid
]
report
.
FREEMAP
[
oid
]
=
report
.
FREEMAP
.
get
(
oid
,
0
)
+
fsize
report
.
USEDMAP
[
oid
]
=
size
report
.
FOIDS
+=
1
report
.
FBYTES
+=
fsize
report
.
CBYTES
+=
size
-
fsize
report
.
TYPEMAP
[
type
]
=
report
.
TYPEMAP
.
get
(
type
,
0
)
+
1
report
.
TYPESIZE
[
type
]
=
report
.
TYPESIZE
.
get
(
type
,
0
)
+
size
except
Exception
,
err
:
print
err
if
__name__
==
"__main__"
:
path
=
sys
.
argv
[
1
]
report
(
analyze
(
path
))
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