Mercurial > p > roundup > code
view roundup/backends/sessions_redis.py @ 7752:b2dbab2b34bc
fix(refactor): multiple fixups using ruff linter; more testing.
Converting to using the ruff linter and its rulesets. Fixed a number
of issues.
admin.py:
sort imports
use immutable tuples as default value markers for parameters where a
None value is valid.
reduced some loops to list comprehensions for performance
used ternary to simplify some if statements
named some variables to make them less magic
(e.g. _default_savepoint_setting = 1000)
fixed some tests for argument counts < 2 becomes != 2 so 3 is an
error.
moved exception handlers outside of loops for performance where
exception handler will abort loop anyway.
renamed variables called 'id' or 'dir' as they shadow builtin
commands.
fix translations of form _("string %s" % value) -> _("string %s") %
value so translation will be looked up with the key before
substitution.
end dicts, tuples with a trailing comma to reduce missing comma
errors if modified
simplified sorted(list(self.setting.keys())) to
sorted(self.setting.keys()) as sorted consumes whole list.
in if conditions put compared variable on left and threshold condition
on right. (no yoda conditions)
multiple noqa: suppression
removed unneeded noqa as lint rulesets are a bit different
do_get - refactor output printing logic: Use fast return if not
special formatting is requested; use isinstance with a tuple
rather than two isinstance calls; cleaned up flow and removed
comments on algorithm as it can be easily read from the code.
do_filter, do_find - refactor output printing logic. Reduce
duplicate code.
do_find - renamed variable 'value' that was set inside a loop. The
loop index variable was also named 'value'.
do_pragma - added hint to use list subcommand if setting was not
found. Replaced condition 'type(x) is bool' with 'isinstance(x,
bool)' for various types.
test_admin.py
added testing for do_list
better test coverage for do_get includes: -S and -d for multilinks,
error case for -d with non-link.
better testing for do_find including all output modes
better testing for do_filter including all output modes
fixed expected output for do_pragma that now includes hint to use
pragma list if setting not found.
| author | John Rouillard <rouilj@ieee.org> |
|---|---|
| date | Fri, 01 Mar 2024 14:53:18 -0500 |
| parents | 216662fbaaee |
| children |
line wrap: on
line source
"""This module defines a redis based store that's used by the CGI interface to store session and one-time-key information. Yes, it's called "sessions" - because originally it only defined a session class. It's now also used for One Time Key handling too. It uses simple strings rather than redis hash structure because the hash the values are always strings. We need to be able to represent the same data types available to rdbms and dbm session stores. session_dbm uses marshal.dumps and marshal.loads. This seems 4 or 18 times faster than the repr()/eval() used by session_rdbms. So use marshal even though it is impossible to read when viewing (using redis-cli). """ __docformat__ = 'restructuredtext' import marshal import redis import time from roundup.anypy.html import html_escape as escape from roundup.i18n import _ from roundup.backends.sessions_common import SessionCommon class BasicDatabase(SessionCommon): ''' Provide a nice encapsulation of a redis store. Keys are id strings, values are automatically marshalled data. ''' name = None default_lifetime = 60*60*24*7 # 1 week # FIXME: figure out how to allow admin to change this # to repr/eval using interfaces.py or other method. # marshalled data is not readable when debugging. tostr = marshal.dumps todict = marshal.loads def __init__(self, db): self.config = db.config url = self.config.SESSIONDB_REDIS_URL # Example at default port without auth. # redis://localhost:6379/0?health_check_interval=2 # # Do not allow decode_responses=True in url, data is # marshal'ed binary data that will get broken by decoding. # Enforce this in configuration. self.redis = redis.Redis.from_url(url=url, decode_responses=False) def makekey(self, key): '''method to namespace all keys using self.name:....''' return "%s:%s" % (self.name, key) def exists(self, infoid): return self.redis.exists(self.makekey(infoid)) def clear(self): '''Delete all keys from the database.''' self.redis.flushdb() _marker = [] def get(self, infoid, value, default=_marker): '''get a specific value from the data associated with a key''' infoid = self.makekey(infoid) v = self.redis.get(infoid) if not v: if default != self._marker: return default raise KeyError(_('Key %(key)s not found in %(name)s ' 'database.') % {"name": self.name, "key": escape(infoid)}) return self.todict(v)[value] def getall(self, infoid): '''return all values associated with a key''' try: d = self.redis.get(self.makekey(infoid)) if d is not None: d = self.todict(d) else: d = {} del d['__timestamp'] return d except KeyError: # It is possible for d to be malformed missing __timestamp. # If so, we get a misleading error, but anydbm does the # same so.... raise KeyError(_('Key %(key)s not found in %(name)s ' 'database.') % {"name": self.name, "key": escape(infoid)}) ''' def set_no_tranaction(self, infoid, **newvalues): """ this is missing transaction and may be affected by a race condition on update. This will work for redis-like embedded databases that don't support watch/multi/exec """ infoid = self.makekey(infoid) timestamp=None values = self.redis.get(infoid) if values is not None: values = self.todict(values) else: values={} try: timestamp = float(values['__timestamp']) except KeyError: pass # stay at None if '__timestamp' in newvalues: try: float(newvalues['__timestamp']) except ValueError: # keep original timestamp if present newvalues['__timestamp'] = timestamp or \ (time.time() + self.default_lifetime) else: newvalues['__timestamp'] = time.time() + self.default_lifetime values.update(newvalues) self.redis.set(infoid, self.tostr(values)) self.redis.expireat(infoid, int(values['__timestamp'])) ''' def set(self, infoid, **newvalues): """ Implement set using watch/multi/exec to get some protection against a change committing between getting the data and setting new fields and saving. """ infoid = self.makekey(infoid) timestamp = None with self.redis.pipeline() as transaction: # Give up and log after three tries. # Do not loop forever. for _retry in [1, 2, 3]: # I am ignoring transaction return values. # Assuming all errors will be via exceptions. # Not clear that return values that useful. transaction.watch(infoid) values = transaction.get(infoid) if values is not None: values = self.todict(values) else: values = {} try: timestamp = float(values['__timestamp']) except KeyError: pass # stay at None if '__timestamp' in newvalues: try: float(newvalues['__timestamp']) except ValueError: # keep original timestamp if present newvalues['__timestamp'] = timestamp or \ (time.time() + self.default_lifetime) else: newvalues['__timestamp'] = time.time() + \ self.default_lifetime values.update(newvalues) transaction.multi() transaction.set(infoid, self.tostr(values)) transaction.expireat(infoid, int(values['__timestamp'])) try: # assume this works or raises an WatchError # exception indicating I need to retry. # Since this is not a real transaction, an error # in one step doesn't roll back other changes. # so I again ignore the return codes as it is not # clear that I can do the rollback myself. # Format and other errors (e.g. expireat('d', 'd')) # raise exceptions that bubble up and result in mail # to admin. transaction.execute() break except redis.exceptions.WatchError: self.log_info( _('Key %(key)s changed in %(name)s db') % {"key": escape(infoid), "name": self.name} ) else: try: username = values['user'] except KeyError: username = "Not Set" raise Exception( _("Redis set failed after %(retries)d retries for " "user %(user)s with key %(key)s") % { "key": escape(infoid), "retries": _retry, "user": username}) def list(self): return list(self.redis.keys(self.makekey('*'))) def destroy(self, infoid=None): '''use unlink rather than delete as unlink is async and doesn't wait for memory to be freed server-side ''' self.redis.unlink(self.makekey(infoid)) def commit(self): ''' no-op ''' pass def lifetime(self, key_lifetime=None): """Return the proper timestamp to expire a key with key_lifetime specified in seconds. Default lifetime is self.default_lifetime. """ return time.time() + (key_lifetime or self.default_lifetime) def updateTimestamp(self, sessid): ''' Even Redis can be overwhelmed by multiple updates, so only update if old session key is > 60 seconds old. ''' sess = self.get(sessid, '__timestamp', None) now = time.time() # unlike other session backends, __timestamp is not set to now # but now + lifetime. if sess is None or now > sess + 60 - self.default_lifetime: lifetime = self.lifetime() # note set also updates the expireat on the key in redis self.set(sessid, __timestamp=lifetime) def clean(self): ''' redis handles key expiration, so nothing to do here. ''' pass def close(self): ''' redis uses a connection pool that self manages, so nothing to do on close.''' pass class Sessions(BasicDatabase): name = 'sessions' class OneTimeKeys(BasicDatabase): name = 'otks' # vim: set sts ts=4 sw=4 et si :
