Changeset 807


Ignore:
Timestamp:
Aug 1, 2008, 2:34:31 AM (16 years ago)
Author:
y_z
Message:

(partially) fixed the mtime comparison logic + comments

File:
1 edited

Legend:

Unmodified
Added
Removed
  • trunk/packages/sipb-xen-base/files/usr/share/python-support/sipb-xen-base/invirt/config.py

    r806 r807  
    1717    """
    1818
    19     # Namespace container for various state variables, so that they can be
    20     # updated by closures.
     19    # Namespace container for state variables, so that they can be updated by
     20    # closures.
    2121    ns = struct()
    2222
     
    2525    else:
    2626        src_mtime = getmtime(src_path)
    27         try:            cache_mtime   = getmtime(cache_path)
    28         except OSError: do_refresh = True
    29         else:           do_refresh = src_mtime > cache_mtime
     27        try:            cache_mtime = getmtime(cache_path)
     28        except OSError: do_refresh  = True
     29        else:           do_refresh  = src_mtime + 1 >= cache_mtime
     30
     31        # We chose not to simply say
     32        #
     33        #   do_refresh = src_mtime >= cache_time
     34        #
     35        # because between the getmtime(src_path) and the time the cache is
     36        # rewritten, the master configuration may have been updated, so future
     37        # checks here would find a cache with a newer mtime than the master
     38        # (and thus treat the cache as containing the latest version of the
     39        # master).  The +1 means that for at least a full second following the
     40        # update to the master, this function will refresh the cache, giving us
     41        # 1 second to write the cache.  Note that if it takes longer than 1
     42        # second to write the cache, then this situation could still arise.
     43        #
     44        # The getmtime calls should logically be part of the same transaction
     45        # as the rest of this function (cache read + conditional cache
     46        # refresh), but to wrap everything in an flock would cause the
     47        # following cache read to be less streamlined.
    3048
    3149    if not do_refresh:
Note: See TracChangeset for help on using the changeset viewer.