Flag to purge cacheloader data upon restart

[ben.wang]

This is regarding to this Jira:
http://jira.jboss.com/jira/browse/JBCACHE-642?page=comments#action_12337423

I propose we should have a flag to purge loader data upon restart.
The rationale for this is to give user an option to purge dead, orphaned data once a while (e.g., if a cache instance crash and restart). We currently don't have the mechanism to purge store data older than a period of time (lie 1 week or month). Otherwise, there'd be no need for this flag.

I'd say we need this not just for FileCacheLoader but rather the generic cache loader store.

[manik]

And an added purge() method on the cache loader interface? Or a simple call to remove("/"), I suppose...

[belaban]

why doesn't remove("/") work here ?

[manik]

remove("/") would work. Just thinking long and hard about adding a purgeOnStartup option, which does beat the whole purpose of a cache loader. I guess it is needed in the described use case though.

[brian.stansberry]

If the use case is to "purge older than X on startup", then doesn't a new method require specifiying X? And cache loaders would have to timestamp their data?

[manik]

Is it that (which will be much more complex) or to bring down a node in the cluster, set the flag to true, and bring it back up again to wipe all old data? And then change the flag to true to restore normal behaviour?

[manik]

I have a preliminary (and very simplistic) impl of this in place - see if it works for you.

<!-- we can now have multiple cache loaders, which get chained -->
 <cacheloader>
 <class>org.jboss.cache.loader.FileCacheLoader</class>
 <!-- same as the old CacheLoaderConfig attribute -->
 <properties>
 location=/tmp/node1
 </properties>
 <!-- whether the cache loader writes are asynchronous -->
 <async>false</async>
 <!-- only one cache loader in the chain may set fetchPersistentState to true.
 An exception is thrown if more than one cache loader sets this to true. -->
 <fetchPersistentState>true</fetchPersistentState>
 <!-- determines whether this cache loader ignores writes - defaults to false. -->
 <ignoreModifications>false</ignoreModifications>
 <!-- if set to true, purges the contents of this cache loader when the cache starts up.
 Defaults to false. -->
 <purgeOnStartup>false</purgeOnStartup>
 </cacheloader>

[galder.zamarreno]

This would be quite useful as an JMX operation, but having it configurable
could be a bit risky. People could use it once and then forget to to switch it
back off and then reporting that no data is persisted.

Besides, they would have to stop Cache, execute it with purge on, stop it
again, and then, change configuration back again and restart it.

[brian.stansberry]

Yes, that's why I think that to be useful the value of this flag needs to be a maxAge rather than a boolean.

[ben.wang]

I will try it out!

Ideally, a maxAge setting is the best way to go. You only purge items that are older than X. But this adds complexity becuase of additional timestamp. So the next best thing is then to do it manually.

[manik]

I actually do like Galder's suggestion about making this a JMX operation though - being able to purge the cache loader

a) without the need to restart the app server
b) without risk of forgetting to switch this off
c) cleaner cache loader configuration

Would an approach like this work for the use case?

[ben.wang]

If it is easier to implement, yes, it will work.

[manik]

It is not necessarily easier to implement (given that the alternative is already implemented), but just a cleaner mechanism. Saves a server restart too.

Thoughts?

[manik]

Ok, HEAD now has an MBean operation to purge the cache loaders. The config file option still is there as well. Both methods for purging the cache loaders will be there in 1.4.0.BETA2.

We'll look at pulling one of these out before releasing a CR.

[belaban]

Maybe I'm a bit slow, but I don't get the use case. Why do you want to purge at startup ?
Anyway: +1 on making this a JMX op, -1 on adding yet another attribute.