Locales and refresh()

[Jeff Reichenberg <jr...@hotwire.com>]

Hi,

We are using Tiles 2.1.4 in a Servlet environment and using the ResolvingLocaleUrlDefinitionDAO to load Tiles definition XML files and cache them in memory.  We have the need to hot deploy / refresh Tile definition XML changes, and therefore call to refresh() on the URlDefinitionsFactory after pushing out Tile definition XML changes.  This in turn calls refresh() on our definitionDao.  It all works fine, except in the case described below where refresh() /sometimes/ does not pick up the definition changes.  Thoughts on whether we are misconfigured or if this a bug in the framework are appreciated...


My understanding of how request locales are handled:

Tiles stores an in-memory cache of Tile definition sets keyed by locale.  These definitions are built lazily as requests come in, based on the locale of the request.  If the locale on the request doesn't yet have definitions associated with it, definitions are built for that locale (and parent locales) on-the-fly and cached.


My understanding of how refresh() is managed:

A global data structure caches last modified dates for the Tile definition XML files.  Whenever the definitions are read from files, the last modified dates are updated.  Upon refresh, these cached last modified dates are compared to the actual last modified date of each XML file on the file system to determine if a refresh is needed.


The problem:

The lazy loading of definition sets per request locale contaminates the global cache of last modified dates in the following situation:
    * Updated Tile definition XML files are pushed onto servers.  The servers continue to field requests.
    * A request with a previously unseen locale comes in, causing a set of definitions to be built for that locale and parent locales.  The updated XMLs are processed from the file system for the new locale only, which updates the global cache of XML file last modified dates.
    * We call refresh(), which compares file system last modified dates to the cached ones and only rebuilds if they differ.  Since the global cache of last modified dates was updated in the previous step, the framework doesn't detect that anything has changed.  It still holds out-dated definitions for all locales other than new ones that come in after the XML file push.  Some locales are now out of sync, and refresh does not bring them current since it is a no-op based on the updated last mod date cache.

The notion of refresh() relying on a global cache of last mod dates that can be updated when only a single locale is processed appears to be an issue.  Bug or misuse?

Thanks!

- Jeff Reichenberg

RE: Locales and refresh()

[Jeff Reichenberg <jr...@hotwire.com>]

Thank you.  Just filed https://issues.apache.org/jira/browse/TILES-503

I don't have a test case or patch unfortunately.  At this point we don't need localization of tiles, so our workaround for now is to force everything into the default locale by configuring our own LocaleResolver class.


-----Original Message-----
From: Antonio Petrelli [mailto:antonio.petrelli@gmail.com] 
Sent: Thursday, March 11, 2010 12:49 AM
To: users@tiles.apache.org
Subject: Re: Locales and refresh()

2010/3/10 Jeff Reichenberg <jr...@hotwire.com>:
> It all works fine, except in the case described below where refresh() /sometimes/ does not pick up the definition changes.  Thoughts on whether we are misconfigured or if this a bug in the framework are appreciated...
>...
> The notion of refresh() relying on a global cache of last mod dates that can be updated when only a single locale is processed appears to be an issue.  Bug or misuse?

Bug indeed. IIRC, when a top definition file is changed (in the sense
of locale hierarchy), lower definition maps are not invalidated.
Please open a JIRA issue:
https://issues.apache.org/jira/browse/TILES
Attach a test case if possible, possibly with a Maven war structure.
If you post a patch, your issue will be processed faster.

Thanks for the report!
Antonio

Re: Locales and refresh()

[Antonio Petrelli <an...@gmail.com>]

2010/3/10 Jeff Reichenberg <jr...@hotwire.com>:
> It all works fine, except in the case described below where refresh() /sometimes/ does not pick up the definition changes.  Thoughts on whether we are misconfigured or if this a bug in the framework are appreciated...
>...
> The notion of refresh() relying on a global cache of last mod dates that can be updated when only a single locale is processed appears to be an issue.  Bug or misuse?

Bug indeed. IIRC, when a top definition file is changed (in the sense
of locale hierarchy), lower definition maps are not invalidated.
Please open a JIRA issue:
https://issues.apache.org/jira/browse/TILES
Attach a test case if possible, possibly with a Maven war structure.
If you post a patch, your issue will be processed faster.

Thanks for the report!
Antonio