Xavier-Do
4444475ef4
This commit adds tooling to profile performance and save execution by
saving stack traces and queries to a file/database in specific format.
----------
Collectors
----------
For now, three different profiling modes (aka Collectors) are available
even if a last once should be introduced by @Gorash to profile qweb
execution.
- SQLCollector (or 'sql'): Saves the current stack trace and the query
every time Cursor.execute() is called. Any query executed on the thread
will be collected, no matter the cursor.
- PeriodicCollector (or 'traces_async'): Saves the stack trace every
'interval' seconds using a parallel thread to profile the caller thread.
The python implementation was optimized to minimize impact on
performance while remaining portable and easy to enable/disable
inside a odoo execution. Higher the frequency (lower the interval),
more impactful the profiling will become on the execution and increase
memory usage. From last experiments, 1ms looks to be a good minimum for
short executions.
- SyncCollector (or 'traces_sync'): Saves the stack trace every function
call/return. This collector is obviously quite impactful on performance
and can quickly overload the memory for long executions, but this is
quite useful to understand the precise path followed by some short
executions. Any time related information will be almost irrelevant with this
collector.
A base Collector defining minimal collectors features can easily be
extended to create custom collectors if needed.
----------------
Profiler & Usage
----------------
Collectors are not supposed to be used by themselves, but should be
given to a Profiler. The Profiler will synchronize collectors starts and
stop, and manage saving them to a file of in a ir_profile in the
database.
Exemple of usage:
```
with Profiler():
do_stuff()
```
This simple example will use the default collectors (sql and
traces_async) and save them to the database. The database is defined
automatically from current_thread 'dbname' if available.
Example of usage:
```
with Profiler(collectors=['sql'], db=False, path=/home/user/logs/do_stuff_profile/{time}):
do_stuff()
```
This more complex example disable the default behavior consisting
to save to the database, gives a path where the profile will be saved
and specify to only use the 'sql' collector. Note that
collectors=[SQLCollector()] would have the same behavior since
Collectors can be either a Collector instance or a string describing the
desired collector. This allows to define custom params for the
collectors and use custom collectors if needed.
Note that it is always possible to get results after execution without
saving it since they are available on the profiler.
```
with Profiler(collectors=['sql'], db=False) as p:
do_stuff()
print(len([None for entry in p.collectors[0].entries if ...]))
```
Profiler will also save the stack below the profiler start point, and
collectors will only collect the part of the stack over this stack.
This is a good way to reduce collectors CPU and memory usage.
Collected entries will be saved as follows:
```
[{
'start': 2.0,
'context': {},
'stack': [
['path_to_file', lno, 'func_name', 'line_content'],
...
],
},
...
]
```
SQLCollector will add three additional keys on each entry:
- query (query without parameters)
- full_query (mogrified query with parameters)
- time (the 'exact' execution time of the query)
----------------
ExecutionContext
----------------
A last tool, ExecutionContext, allows to define some context on some block of code:
Example of usage:
```
def process_modules(modules)
for module in modules:
with ExecutionContext(module=module): # note the 'not linter frienldy but still convenient' 2 spaces indentation
do_stuff(module):
```
This context will automatically be added in the stack as a virtual frame between
process_modules and do_stuff in order to split do_stuff from one single frame to
one frame per module.
----------
Speedscope
----------
The saved data are in a simple json format easy to analyze, but can't be visualized in
speedscope as they are. A utility class `Speedscope` can be used to generate a format
readable by speedscope. The used format is actually the format defined by speedscope,
meaning that all features should be available using it.
The output format is evented, meaning that we need to transform a list of samples
(a list of stack) to a list of event (going in/out a frame).
This is the main task of the Speedscope, as well as combining samples from different
sources, to display SQLCollector and PeriodicCollector results mixed together.
When stored on an ir_profile, the default speedscope generation can easily be generated
with the speedscope computed field.
This class can be used as it is but will mainly be useful for the next commit.
Special thanks to @rco-odoo for the in depth review and @Gorash for support.
759 lines
27 KiB
Python
759 lines
27 KiB
Python
# -*- coding: utf-8 -*-
|
|
# Part of Odoo. See LICENSE file for full copyright and licensing details.
|
|
|
|
|
|
"""
|
|
The PostgreSQL connector is a connectivity layer between the OpenERP code and
|
|
the database, *not* a database abstraction toolkit. Database abstraction is what
|
|
the ORM does, in fact.
|
|
"""
|
|
|
|
from contextlib import contextmanager
|
|
from functools import wraps
|
|
import itertools
|
|
import logging
|
|
import time
|
|
import uuid
|
|
import warnings
|
|
|
|
from decorator import decorator
|
|
import psycopg2
|
|
import psycopg2.extras
|
|
import psycopg2.extensions
|
|
from psycopg2.extensions import ISOLATION_LEVEL_AUTOCOMMIT, ISOLATION_LEVEL_READ_COMMITTED, ISOLATION_LEVEL_REPEATABLE_READ
|
|
from psycopg2.pool import PoolError
|
|
from werkzeug import urls
|
|
|
|
from odoo.api import Environment
|
|
|
|
psycopg2.extensions.register_type(psycopg2.extensions.UNICODE)
|
|
|
|
_logger = logging.getLogger(__name__)
|
|
|
|
def unbuffer(symb, cr):
|
|
if symb is None:
|
|
return None
|
|
return str(symb)
|
|
|
|
def undecimalize(symb, cr):
|
|
if symb is None:
|
|
return None
|
|
return float(symb)
|
|
|
|
psycopg2.extensions.register_type(psycopg2.extensions.new_type((700, 701, 1700,), 'float', undecimalize))
|
|
|
|
|
|
from . import tools
|
|
from .tools.func import frame_codeinfo
|
|
|
|
from .tools import parse_version as pv
|
|
if pv(psycopg2.__version__) < pv('2.7'):
|
|
from psycopg2._psycopg import QuotedString
|
|
def adapt_string(adapted):
|
|
"""Python implementation of psycopg/psycopg2#459 from v2.7"""
|
|
if '\x00' in adapted:
|
|
raise ValueError("A string literal cannot contain NUL (0x00) characters.")
|
|
return QuotedString(adapted)
|
|
|
|
psycopg2.extensions.register_adapter(str, adapt_string)
|
|
|
|
from datetime import timedelta
|
|
import threading
|
|
from inspect import currentframe
|
|
|
|
|
|
def flush_env(cr, *, clear=True):
|
|
""" Retrieve and flush an environment corresponding to the given cursor.
|
|
Also clear the environment if ``clear`` is true.
|
|
"""
|
|
env_to_flush = None
|
|
for env in list(Environment.envs):
|
|
# don't flush() on another cursor or with a RequestUID
|
|
if env.cr is cr and (isinstance(env.uid, int) or env.uid is None):
|
|
env_to_flush = env
|
|
if env.uid is not None:
|
|
break # prefer an environment with a real uid
|
|
|
|
if env_to_flush is not None:
|
|
env_to_flush['base'].flush()
|
|
if clear:
|
|
env_to_flush.clear() # clear remaining new records to compute
|
|
|
|
def clear_env(cr):
|
|
""" Retrieve and clear an environment corresponding to the given cursor """
|
|
for env in list(Environment.envs):
|
|
if env.cr is cr:
|
|
env.clear()
|
|
break
|
|
|
|
import re
|
|
re_from = re.compile('.* from "?([a-zA-Z_0-9]+)"? .*$')
|
|
re_into = re.compile('.* into "?([a-zA-Z_0-9]+)"? .*$')
|
|
|
|
sql_counter = 0
|
|
|
|
|
|
@decorator
|
|
def check(f, self, *args, **kwargs):
|
|
""" Wrap a cursor method that cannot be called when the cursor is closed. """
|
|
if self._closed:
|
|
raise psycopg2.OperationalError('Unable to use a closed cursor.')
|
|
return f(self, *args, **kwargs)
|
|
|
|
|
|
class BaseCursor:
|
|
""" Base class for cursors that manage pre/post commit hooks. """
|
|
|
|
def __init__(self):
|
|
self.precommit = tools.Callbacks()
|
|
self.postcommit = tools.Callbacks()
|
|
self.prerollback = tools.Callbacks()
|
|
self.postrollback = tools.Callbacks()
|
|
|
|
@contextmanager
|
|
@check
|
|
def savepoint(self, flush=True):
|
|
"""context manager entering in a new savepoint"""
|
|
name = uuid.uuid1().hex
|
|
if flush:
|
|
flush_env(self, clear=False)
|
|
self.precommit.run()
|
|
self.execute('SAVEPOINT "%s"' % name)
|
|
try:
|
|
yield
|
|
if flush:
|
|
flush_env(self, clear=False)
|
|
self.precommit.run()
|
|
except Exception:
|
|
if flush:
|
|
clear_env(self)
|
|
self.precommit.clear()
|
|
self.execute('ROLLBACK TO SAVEPOINT "%s"' % name)
|
|
raise
|
|
else:
|
|
self.execute('RELEASE SAVEPOINT "%s"' % name)
|
|
|
|
def __enter__(self):
|
|
""" Using the cursor as a contextmanager automatically commits and
|
|
closes it::
|
|
|
|
with cr:
|
|
cr.execute(...)
|
|
|
|
# cr is committed if no failure occurred
|
|
# cr is closed in any case
|
|
"""
|
|
return self
|
|
|
|
def __exit__(self, exc_type, exc_value, traceback):
|
|
if exc_type is None:
|
|
self.commit()
|
|
self.close()
|
|
|
|
|
|
class Cursor(BaseCursor):
|
|
"""Represents an open transaction to the PostgreSQL DB backend,
|
|
acting as a lightweight wrapper around psycopg2's
|
|
``cursor`` objects.
|
|
|
|
``Cursor`` is the object behind the ``cr`` variable used all
|
|
over the OpenERP code.
|
|
|
|
.. rubric:: Transaction Isolation
|
|
|
|
One very important property of database transactions is the
|
|
level of isolation between concurrent transactions.
|
|
The SQL standard defines four levels of transaction isolation,
|
|
ranging from the most strict *Serializable* level, to the least
|
|
strict *Read Uncommitted* level. These levels are defined in
|
|
terms of the phenomena that must not occur between concurrent
|
|
transactions, such as *dirty read*, etc.
|
|
In the context of a generic business data management software
|
|
such as OpenERP, we need the best guarantees that no data
|
|
corruption can ever be cause by simply running multiple
|
|
transactions in parallel. Therefore, the preferred level would
|
|
be the *serializable* level, which ensures that a set of
|
|
transactions is guaranteed to produce the same effect as
|
|
running them one at a time in some order.
|
|
|
|
However, most database management systems implement a limited
|
|
serializable isolation in the form of
|
|
`snapshot isolation <http://en.wikipedia.org/wiki/Snapshot_isolation>`_,
|
|
providing most of the same advantages as True Serializability,
|
|
with a fraction of the performance cost.
|
|
With PostgreSQL up to version 9.0, this snapshot isolation was
|
|
the implementation of both the ``REPEATABLE READ`` and
|
|
``SERIALIZABLE`` levels of the SQL standard.
|
|
As of PostgreSQL 9.1, the previous snapshot isolation implementation
|
|
was kept for ``REPEATABLE READ``, while a new ``SERIALIZABLE``
|
|
level was introduced, providing some additional heuristics to
|
|
detect a concurrent update by parallel transactions, and forcing
|
|
one of them to rollback.
|
|
|
|
OpenERP implements its own level of locking protection
|
|
for transactions that are highly likely to provoke concurrent
|
|
updates, such as stock reservations or document sequences updates.
|
|
Therefore we mostly care about the properties of snapshot isolation,
|
|
but we don't really need additional heuristics to trigger transaction
|
|
rollbacks, as we are taking care of triggering instant rollbacks
|
|
ourselves when it matters (and we can save the additional performance
|
|
hit of these heuristics).
|
|
|
|
As a result of the above, we have selected ``REPEATABLE READ`` as
|
|
the default transaction isolation level for OpenERP cursors, as
|
|
it will be mapped to the desired ``snapshot isolation`` level for
|
|
all supported PostgreSQL version (8.3 - 9.x).
|
|
|
|
Note: up to psycopg2 v.2.4.2, psycopg2 itself remapped the repeatable
|
|
read level to serializable before sending it to the database, so it would
|
|
actually select the new serializable mode on PostgreSQL 9.1. Make
|
|
sure you use psycopg2 v2.4.2 or newer if you use PostgreSQL 9.1 and
|
|
the performance hit is a concern for you.
|
|
|
|
.. attribute:: cache
|
|
|
|
Cache dictionary with a "request" (-ish) lifecycle, only lives as
|
|
long as the cursor itself does and proactively cleared when the
|
|
cursor is closed.
|
|
|
|
This cache should *only* be used to store repeatable reads as it
|
|
ignores rollbacks and savepoints, it should not be used to store
|
|
*any* data which may be modified during the life of the cursor.
|
|
|
|
"""
|
|
IN_MAX = 1000 # decent limit on size of IN queries - guideline = Oracle limit
|
|
|
|
def __init__(self, pool, dbname, dsn, serialized=True):
|
|
super().__init__()
|
|
|
|
self.sql_from_log = {}
|
|
self.sql_into_log = {}
|
|
|
|
# default log level determined at cursor creation, could be
|
|
# overridden later for debugging purposes
|
|
self.sql_log = _logger.isEnabledFor(logging.DEBUG)
|
|
|
|
self.sql_log_count = 0
|
|
|
|
# avoid the call of close() (by __del__) if an exception
|
|
# is raised by any of the following initialisations
|
|
self._closed = True
|
|
|
|
self.__pool = pool
|
|
self.dbname = dbname
|
|
# Whether to enable snapshot isolation level for this cursor.
|
|
# see also the docstring of Cursor.
|
|
self._serialized = serialized
|
|
|
|
self._cnx = pool.borrow(dsn)
|
|
self._obj = self._cnx.cursor()
|
|
if self.sql_log:
|
|
self.__caller = frame_codeinfo(currentframe(), 2)
|
|
else:
|
|
self.__caller = False
|
|
self._closed = False # real initialisation value
|
|
self.autocommit(False)
|
|
|
|
self._default_log_exceptions = True
|
|
|
|
self.cache = {}
|
|
self._now = None
|
|
|
|
def __build_dict(self, row):
|
|
return {d.name: row[i] for i, d in enumerate(self._obj.description)}
|
|
def dictfetchone(self):
|
|
row = self._obj.fetchone()
|
|
return row and self.__build_dict(row)
|
|
def dictfetchmany(self, size):
|
|
return [self.__build_dict(row) for row in self._obj.fetchmany(size)]
|
|
def dictfetchall(self):
|
|
return [self.__build_dict(row) for row in self._obj.fetchall()]
|
|
|
|
def __del__(self):
|
|
if not self._closed and not self._cnx.closed:
|
|
# Oops. 'self' has not been closed explicitly.
|
|
# The cursor will be deleted by the garbage collector,
|
|
# but the database connection is not put back into the connection
|
|
# pool, preventing some operation on the database like dropping it.
|
|
# This can also lead to a server overload.
|
|
msg = "Cursor not closed explicitly\n"
|
|
if self.__caller:
|
|
msg += "Cursor was created at %s:%s" % self.__caller
|
|
else:
|
|
msg += "Please enable sql debugging to trace the caller."
|
|
_logger.warning(msg)
|
|
self._close(True)
|
|
|
|
def _format(self, query, params=None):
|
|
encoding = psycopg2.extensions.encodings[self.connection.encoding]
|
|
return self._obj.mogrify(query, params).decode(encoding, 'replace')
|
|
|
|
@check
|
|
def execute(self, query, params=None, log_exceptions=None):
|
|
if params and not isinstance(params, (tuple, list, dict)):
|
|
# psycopg2's TypeError is not clear if you mess up the params
|
|
raise ValueError("SQL query parameters should be a tuple, list or dict; got %r" % (params,))
|
|
|
|
if self.sql_log:
|
|
_logger.debug("query: %s", self._format(query, params))
|
|
start = time.time()
|
|
try:
|
|
params = params or None
|
|
res = self._obj.execute(query, params)
|
|
except Exception as e:
|
|
if self._default_log_exceptions if log_exceptions is None else log_exceptions:
|
|
_logger.error("bad query: %s\nERROR: %s", tools.ustr(self._obj.query or query), e)
|
|
raise
|
|
|
|
# simple query count is always computed
|
|
self.sql_log_count += 1
|
|
delay = (time.time() - start)
|
|
current_thread = threading.current_thread()
|
|
if hasattr(current_thread, 'query_count'):
|
|
current_thread.query_count += 1
|
|
current_thread.query_time += delay
|
|
|
|
# optional hooks for performance and tracing analysis
|
|
for hook in getattr(current_thread, 'query_hooks', ()):
|
|
hook(self, query, params, start, delay)
|
|
|
|
# advanced stats only if sql_log is enabled
|
|
if self.sql_log:
|
|
delay *= 1E6
|
|
|
|
query_lower = self._obj.query.decode().lower()
|
|
res_from = re_from.match(query_lower)
|
|
if res_from:
|
|
self.sql_from_log.setdefault(res_from.group(1), [0, 0])
|
|
self.sql_from_log[res_from.group(1)][0] += 1
|
|
self.sql_from_log[res_from.group(1)][1] += delay
|
|
res_into = re_into.match(query_lower)
|
|
if res_into:
|
|
self.sql_into_log.setdefault(res_into.group(1), [0, 0])
|
|
self.sql_into_log[res_into.group(1)][0] += 1
|
|
self.sql_into_log[res_into.group(1)][1] += delay
|
|
return res
|
|
|
|
def split_for_in_conditions(self, ids, size=None):
|
|
"""Split a list of identifiers into one or more smaller tuples
|
|
safe for IN conditions, after uniquifying them."""
|
|
return tools.misc.split_every(size or self.IN_MAX, ids)
|
|
|
|
def print_log(self):
|
|
global sql_counter
|
|
|
|
if not self.sql_log:
|
|
return
|
|
def process(type):
|
|
sqllogs = {'from': self.sql_from_log, 'into': self.sql_into_log}
|
|
sum = 0
|
|
if sqllogs[type]:
|
|
sqllogitems = sqllogs[type].items()
|
|
_logger.debug("SQL LOG %s:", type)
|
|
for r in sorted(sqllogitems, key=lambda k: k[1]):
|
|
delay = timedelta(microseconds=r[1][1])
|
|
_logger.debug("table: %s: %s/%s", r[0], delay, r[1][0])
|
|
sum += r[1][1]
|
|
sqllogs[type].clear()
|
|
sum = timedelta(microseconds=sum)
|
|
_logger.debug("SUM %s:%s/%d [%d]", type, sum, self.sql_log_count, sql_counter)
|
|
sqllogs[type].clear()
|
|
process('from')
|
|
process('into')
|
|
self.sql_log_count = 0
|
|
self.sql_log = False
|
|
|
|
@check
|
|
def close(self):
|
|
return self._close(False)
|
|
|
|
def _close(self, leak=False):
|
|
global sql_counter
|
|
|
|
if not self._obj:
|
|
return
|
|
|
|
del self.cache
|
|
|
|
# simple query count is always computed
|
|
sql_counter += self.sql_log_count
|
|
|
|
# advanced stats only if sql_log is enabled
|
|
self.print_log()
|
|
|
|
self._obj.close()
|
|
|
|
# This force the cursor to be freed, and thus, available again. It is
|
|
# important because otherwise we can overload the server very easily
|
|
# because of a cursor shortage (because cursors are not garbage
|
|
# collected as fast as they should). The problem is probably due in
|
|
# part because browse records keep a reference to the cursor.
|
|
del self._obj
|
|
|
|
# Clean the underlying connection, and run rollback hooks.
|
|
self.rollback()
|
|
|
|
self._closed = True
|
|
|
|
if leak:
|
|
self._cnx.leaked = True
|
|
else:
|
|
chosen_template = tools.config['db_template']
|
|
templates_list = tuple(set(['template0', 'template1', 'postgres', chosen_template]))
|
|
keep_in_pool = self.dbname not in templates_list
|
|
self.__pool.give_back(self._cnx, keep_in_pool=keep_in_pool)
|
|
|
|
@check
|
|
def autocommit(self, on):
|
|
if on:
|
|
warnings.warn(
|
|
"Since Odoo 13.0, the ORM delays UPDATE queries for "
|
|
"performance reasons. Since then, using the ORM with "
|
|
" autocommit(True) is unsafe, as computed fields may not be "
|
|
"fully computed at commit.", DeprecationWarning, stacklevel=2)
|
|
isolation_level = ISOLATION_LEVEL_AUTOCOMMIT
|
|
else:
|
|
# If a serializable cursor was requested, we
|
|
# use the appropriate PotsgreSQL isolation level
|
|
# that maps to snapshot isolation.
|
|
# For all supported PostgreSQL versions (8.3-9.x),
|
|
# this is currently the ISOLATION_REPEATABLE_READ.
|
|
# See also the docstring of this class.
|
|
# NOTE: up to psycopg 2.4.2, repeatable read
|
|
# is remapped to serializable before being
|
|
# sent to the database, so it is in fact
|
|
# unavailable for use with pg 9.1.
|
|
isolation_level = \
|
|
ISOLATION_LEVEL_REPEATABLE_READ \
|
|
if self._serialized \
|
|
else ISOLATION_LEVEL_READ_COMMITTED
|
|
self._cnx.set_isolation_level(isolation_level)
|
|
|
|
@check
|
|
def after(self, event, func):
|
|
""" Register an event handler.
|
|
|
|
:param event: the event, either `'commit'` or `'rollback'`
|
|
:param func: a callable object, called with no argument after the
|
|
event occurs
|
|
|
|
Be careful when coding an event handler, since any operation on the
|
|
cursor that was just committed/rolled back will take place in the
|
|
next transaction that has already begun, and may still be rolled
|
|
back or committed independently. You may consider the use of a
|
|
dedicated temporary cursor to do some database operation.
|
|
"""
|
|
warnings.warn(
|
|
"Cursor.after() is deprecated, use Cursor.postcommit.add() instead.",
|
|
DeprecationWarning,
|
|
)
|
|
if event == 'commit':
|
|
self.postcommit.add(func)
|
|
elif event == 'rollback':
|
|
self.postrollback.add(func)
|
|
|
|
@check
|
|
def commit(self):
|
|
""" Perform an SQL `COMMIT` """
|
|
flush_env(self)
|
|
self.precommit.run()
|
|
result = self._cnx.commit()
|
|
self._now = None
|
|
self.prerollback.clear()
|
|
self.postrollback.clear()
|
|
self.postcommit.run()
|
|
return result
|
|
|
|
@check
|
|
def rollback(self):
|
|
""" Perform an SQL `ROLLBACK` """
|
|
clear_env(self)
|
|
self.precommit.clear()
|
|
self.postcommit.clear()
|
|
self.prerollback.run()
|
|
result = self._cnx.rollback()
|
|
self._now = None
|
|
self.postrollback.run()
|
|
return result
|
|
|
|
@check
|
|
def __getattr__(self, name):
|
|
return getattr(self._obj, name)
|
|
|
|
@property
|
|
def closed(self):
|
|
return self._closed
|
|
|
|
def now(self):
|
|
""" Return the transaction's timestamp ``NOW() AT TIME ZONE 'UTC'``. """
|
|
if self._now is None:
|
|
self.execute("SELECT (now() AT TIME ZONE 'UTC')")
|
|
self._now = self.fetchone()[0]
|
|
return self._now
|
|
|
|
|
|
class TestCursor(BaseCursor):
|
|
""" A pseudo-cursor to be used for tests, on top of a real cursor. It keeps
|
|
the transaction open across requests, and simulates committing, rolling
|
|
back, and closing:
|
|
|
|
test cursor | queries on actual cursor
|
|
------------------------+---------------------------------------
|
|
cr = TestCursor(...) | SAVEPOINT test_cursor_N
|
|
|
|
|
cr.execute(query) | query
|
|
|
|
|
cr.commit() | SAVEPOINT test_cursor_N
|
|
|
|
|
cr.rollback() | ROLLBACK TO SAVEPOINT test_cursor_N
|
|
|
|
|
cr.close() | ROLLBACK TO SAVEPOINT test_cursor_N
|
|
|
|
|
|
|
"""
|
|
_savepoint_seq = itertools.count()
|
|
|
|
def __init__(self, cursor, lock):
|
|
self._closed = False
|
|
self._cursor = cursor
|
|
# we use a lock to serialize concurrent requests
|
|
self._lock = lock
|
|
self._lock.acquire()
|
|
# in order to simulate commit and rollback, the cursor maintains a
|
|
# savepoint at its last commit
|
|
self._savepoint = "test_cursor_%s" % next(self._savepoint_seq)
|
|
self._cursor.execute('SAVEPOINT "%s"' % self._savepoint)
|
|
|
|
def close(self):
|
|
if not self._closed:
|
|
self.rollback()
|
|
self._closed = True
|
|
self._lock.release()
|
|
|
|
def autocommit(self, on):
|
|
_logger.debug("TestCursor.autocommit(%r) does nothing", on)
|
|
|
|
@check
|
|
def commit(self):
|
|
""" Perform an SQL `COMMIT` """
|
|
flush_env(self)
|
|
self.precommit.run()
|
|
self._cursor.execute('SAVEPOINT "%s"' % self._savepoint)
|
|
self.prerollback.clear()
|
|
self.postrollback.clear()
|
|
self.postcommit.clear() # TestCursor ignores post-commit hooks
|
|
|
|
@check
|
|
def rollback(self):
|
|
""" Perform an SQL `ROLLBACK` """
|
|
clear_env(self)
|
|
self.precommit.clear()
|
|
self.postcommit.clear()
|
|
self.prerollback.run()
|
|
self._cursor.execute('ROLLBACK TO SAVEPOINT "%s"' % self._savepoint)
|
|
self.postrollback.run()
|
|
|
|
def __getattr__(self, name):
|
|
value = getattr(self._cursor, name)
|
|
if callable(value) and self._closed:
|
|
raise psycopg2.OperationalError('Unable to use a closed cursor.')
|
|
return value
|
|
|
|
|
|
class PsycoConnection(psycopg2.extensions.connection):
|
|
pass
|
|
|
|
class ConnectionPool(object):
|
|
""" The pool of connections to database(s)
|
|
|
|
Keep a set of connections to pg databases open, and reuse them
|
|
to open cursors for all transactions.
|
|
|
|
The connections are *not* automatically closed. Only a close_db()
|
|
can trigger that.
|
|
"""
|
|
|
|
def locked(fun):
|
|
@wraps(fun)
|
|
def _locked(self, *args, **kwargs):
|
|
self._lock.acquire()
|
|
try:
|
|
return fun(self, *args, **kwargs)
|
|
finally:
|
|
self._lock.release()
|
|
return _locked
|
|
|
|
def __init__(self, maxconn=64):
|
|
self._connections = []
|
|
self._maxconn = max(maxconn, 1)
|
|
self._lock = threading.Lock()
|
|
|
|
def __repr__(self):
|
|
used = len([1 for c, u in self._connections[:] if u])
|
|
count = len(self._connections)
|
|
return "ConnectionPool(used=%d/count=%d/max=%d)" % (used, count, self._maxconn)
|
|
|
|
def _debug(self, msg, *args):
|
|
_logger.debug(('%r ' + msg), self, *args)
|
|
|
|
@locked
|
|
def borrow(self, connection_info):
|
|
"""
|
|
:param dict connection_info: dict of psql connection keywords
|
|
:rtype: PsycoConnection
|
|
"""
|
|
# free dead and leaked connections
|
|
for i, (cnx, _) in tools.reverse_enumerate(self._connections):
|
|
if cnx.closed:
|
|
self._connections.pop(i)
|
|
self._debug('Removing closed connection at index %d: %r', i, cnx.dsn)
|
|
continue
|
|
if getattr(cnx, 'leaked', False):
|
|
delattr(cnx, 'leaked')
|
|
self._connections.pop(i)
|
|
self._connections.append((cnx, False))
|
|
_logger.info('%r: Free leaked connection to %r', self, cnx.dsn)
|
|
|
|
for i, (cnx, used) in enumerate(self._connections):
|
|
if not used and cnx._original_dsn == connection_info:
|
|
try:
|
|
cnx.reset()
|
|
except psycopg2.OperationalError:
|
|
self._debug('Cannot reset connection at index %d: %r', i, cnx.dsn)
|
|
# psycopg2 2.4.4 and earlier do not allow closing a closed connection
|
|
if not cnx.closed:
|
|
cnx.close()
|
|
continue
|
|
self._connections.pop(i)
|
|
self._connections.append((cnx, True))
|
|
self._debug('Borrow existing connection to %r at index %d', cnx.dsn, i)
|
|
|
|
return cnx
|
|
|
|
if len(self._connections) >= self._maxconn:
|
|
# try to remove the oldest connection not used
|
|
for i, (cnx, used) in enumerate(self._connections):
|
|
if not used:
|
|
self._connections.pop(i)
|
|
if not cnx.closed:
|
|
cnx.close()
|
|
self._debug('Removing old connection at index %d: %r', i, cnx.dsn)
|
|
break
|
|
else:
|
|
# note: this code is called only if the for loop has completed (no break)
|
|
raise PoolError('The Connection Pool Is Full')
|
|
|
|
try:
|
|
result = psycopg2.connect(
|
|
connection_factory=PsycoConnection,
|
|
**connection_info)
|
|
except psycopg2.Error:
|
|
_logger.info('Connection to the database failed')
|
|
raise
|
|
result._original_dsn = connection_info
|
|
self._connections.append((result, True))
|
|
self._debug('Create new connection')
|
|
return result
|
|
|
|
@locked
|
|
def give_back(self, connection, keep_in_pool=True):
|
|
self._debug('Give back connection to %r', connection.dsn)
|
|
for i, (cnx, used) in enumerate(self._connections):
|
|
if cnx is connection:
|
|
self._connections.pop(i)
|
|
if keep_in_pool:
|
|
self._connections.append((cnx, False))
|
|
self._debug('Put connection to %r in pool', cnx.dsn)
|
|
else:
|
|
self._debug('Forgot connection to %r', cnx.dsn)
|
|
cnx.close()
|
|
break
|
|
else:
|
|
raise PoolError('This connection does not belong to the pool')
|
|
|
|
@locked
|
|
def close_all(self, dsn=None):
|
|
count = 0
|
|
last = None
|
|
for i, (cnx, used) in tools.reverse_enumerate(self._connections):
|
|
if dsn is None or cnx._original_dsn == dsn:
|
|
cnx.close()
|
|
last = self._connections.pop(i)[0]
|
|
count += 1
|
|
_logger.info('%r: Closed %d connections %s', self, count,
|
|
(dsn and last and 'to %r' % last.dsn) or '')
|
|
|
|
|
|
class Connection(object):
|
|
""" A lightweight instance of a connection to postgres
|
|
"""
|
|
def __init__(self, pool, dbname, dsn):
|
|
self.dbname = dbname
|
|
self.dsn = dsn
|
|
self.__pool = pool
|
|
|
|
def cursor(self, serialized=True):
|
|
cursor_type = serialized and 'serialized ' or ''
|
|
_logger.debug('create %scursor to %r', cursor_type, self.dsn)
|
|
return Cursor(self.__pool, self.dbname, self.dsn, serialized=serialized)
|
|
|
|
# serialized_cursor is deprecated - cursors are serialized by default
|
|
serialized_cursor = cursor
|
|
|
|
def __bool__(self):
|
|
raise NotImplementedError()
|
|
__nonzero__ = __bool__
|
|
|
|
def connection_info_for(db_or_uri):
|
|
""" parse the given `db_or_uri` and return a 2-tuple (dbname, connection_params)
|
|
|
|
Connection params are either a dictionary with a single key ``dsn``
|
|
containing a connection URI, or a dictionary containing connection
|
|
parameter keywords which psycopg2 can build a key/value connection string
|
|
(dsn) from
|
|
|
|
:param str db_or_uri: database name or postgres dsn
|
|
:rtype: (str, dict)
|
|
"""
|
|
if db_or_uri.startswith(('postgresql://', 'postgres://')):
|
|
# extract db from uri
|
|
us = urls.url_parse(db_or_uri)
|
|
if len(us.path) > 1:
|
|
db_name = us.path[1:]
|
|
elif us.username:
|
|
db_name = us.username
|
|
else:
|
|
db_name = us.hostname
|
|
return db_name, {'dsn': db_or_uri}
|
|
|
|
connection_info = {'database': db_or_uri}
|
|
for p in ('host', 'port', 'user', 'password', 'sslmode'):
|
|
cfg = tools.config['db_' + p]
|
|
if cfg:
|
|
connection_info[p] = cfg
|
|
|
|
return db_or_uri, connection_info
|
|
|
|
_Pool = None
|
|
|
|
def db_connect(to, allow_uri=False):
|
|
global _Pool
|
|
if _Pool is None:
|
|
_Pool = ConnectionPool(int(tools.config['db_maxconn']))
|
|
|
|
db, info = connection_info_for(to)
|
|
if not allow_uri and db != to:
|
|
raise ValueError('URI connections not allowed')
|
|
return Connection(_Pool, db, info)
|
|
|
|
def close_db(db_name):
|
|
""" You might want to call odoo.modules.registry.Registry.delete(db_name) along this function."""
|
|
global _Pool
|
|
if _Pool:
|
|
_Pool.close_all(connection_info_for(db_name)[1])
|
|
|
|
def close_all():
|
|
global _Pool
|
|
if _Pool:
|
|
_Pool.close_all()
|