source: trunk/packages/invirt-base/python/invirt/config.py @ 2507

Last change on this file since 2507 was 1934, checked in by price, 16 years ago

invirt.common: give clearer error message on missing config variable

File size: 5.2 KB
Line 
1from __future__ import with_statement
2
3import json
4from invirt.common import *
5import os
6from os import rename
7from os.path import getmtime
8from contextlib import closing
9import yaml
10import re
11
12try:    loader = yaml.CSafeLoader
13except: loader = yaml.SafeLoader
14
15src_path    = '/etc/invirt/master.yaml'
16src_dirpath = '/etc/invirt/conf.d'
17cache_path  = '/var/lib/invirt/cache.json'
18lock_path   = '/var/lib/invirt/cache.lock'
19
20def augment(d1, d2):
21    """Splice dict-tree d2 into d1.  Return d1.
22
23    d2 may be None for an empty dict-tree, because yaml.load produces that.
24
25    Example:
26    >>> d = {'a': {'b': 1}, 'c': 2}
27    >>> augment(d, {'a': {'d': 3}})
28    {'a': {'b', 1, 'd': 3}, 'c': 2}
29    >>> d
30    {'a': {'b', 1, 'd': 3}, 'c': 2}
31    """
32    if d2 is None:
33        return d1
34    for k in d2:
35        if k in d1 and isinstance(d1[k], dict):
36            augment(d1[k], d2[k])
37        else:
38            d1[k] = d2[k]
39    return d1
40
41def run_parts_list(dirname):
42    """Reimplements Debian's run-parts --list.
43
44    One difference from run-parts's behavior: run-parts --list /foo/
45    will give output like /foo//bar, but run_parts_list('/foo/') gives
46    /foo/bar in deference to Python conventions.
47
48    Matches documented behavior of run-parts in debianutils v2.28.2, dated 2007.
49    """
50    # From run-parts(8).
51    lanana_re   = re.compile('^[a-z0-9]+$')
52    lsb_re      = re.compile('^_?([a-z0-9_.]+-)+[a-z0-9]+$')
53    deb_cron_re = re.compile('^[a-z0-9][a-z0-9-]*$')
54    for name in os.listdir(dirname):
55        if lanana_re.match(name) or lsb_re.match(name) or deb_cron_re.match(name):
56            yield os.path.join(dirname, name)
57
58def list_files():
59    yield src_path
60    for name in run_parts_list(src_dirpath):
61        yield name
62
63def load_master():
64    config = dict()
65    for filename in list_files():
66        with closing(file(filename)) as f:
67            augment(config, yaml.load(f, loader))
68    return config
69
70def get_src_mtime():
71    return max(max(getmtime(filename) for filename in list_files()),
72               getmtime(src_dirpath))
73
74def load(force_refresh = False):
75    """
76    Try loading the configuration from the faster-to-load JSON cache at
77    cache_path.  If it doesn't exist or is outdated, load the configuration
78    instead from the original YAML file at src_path and regenerate the cache.
79    I assume I have the permissions to write to the cache directory.
80    """
81
82    # Namespace container for state variables, so that they can be updated by
83    # closures.
84    ns = struct()
85
86    if force_refresh:
87        do_refresh = True
88    else:
89        src_mtime = get_src_mtime()
90        try:            cache_mtime = getmtime(cache_path)
91        except OSError: do_refresh  = True
92        else:           do_refresh  = src_mtime + 1 >= cache_mtime
93
94        # We chose not to simply say
95        #
96        #   do_refresh = src_mtime >= cache_time
97        #
98        # because between the getmtime(src_path) and the time the cache is
99        # rewritten, the master configuration may have been updated, so future
100        # checks here would find a cache with a newer mtime than the master
101        # (and thus treat the cache as containing the latest version of the
102        # master).  The +1 means that for at least a full second following the
103        # update to the master, this function will refresh the cache, giving us
104        # 1 second to write the cache.  Note that if it takes longer than 1
105        # second to write the cache, then this situation could still arise.
106        #
107        # The getmtime calls should logically be part of the same transaction
108        # as the rest of this function (cache read + conditional cache
109        # refresh), but to wrap everything in an flock would cause the
110        # following cache read to be less streamlined.
111
112    if not do_refresh:
113        # Try reading from the cache first.  This must be transactionally
114        # isolated from concurrent writes to prevent reading an incomplete
115        # (changing) version of the data (but the transaction can share the
116        # lock with other concurrent reads).  This isolation is accomplished
117        # using an atomic filesystem rename in the refreshing stage.
118        try: 
119            with closing(file(cache_path)) as f:
120                ns.cfg = json.read(f.read())
121        except: do_refresh = True
122
123    if do_refresh:
124        # Atomically reload the source and regenerate the cache.  The read and
125        # write must be a single transaction, or a stale version may be
126        # written (if another read/write of a more recent configuration
127        # is interleaved).  The final atomic rename is to keep this
128        # transactionally isolated from the above cache read.  If we fail to
129        # acquire the lock, just try to load the master configuration.
130        try:
131            with lock_file(lock_path):
132                ns.cfg = load_master()
133                try: 
134                    with closing(file(cache_path + '.tmp', 'w')) as f:
135                        f.write(json.write(ns.cfg))
136                except: pass # silent failure
137                else: rename(cache_path + '.tmp', cache_path)
138        except IOError:
139            ns.cfg = load_master()
140    return ns.cfg
141
142dicts = load()
143structs = dicts2struct(dicts, '')
144
145# vim:et:sw=4:ts=4
Note: See TracBrowser for help on using the repository browser.