[ZODB-Dev] [Problem] "_v_" variables too volatile

Toby Dickenson tdickenson at geminidataloggers.com
Wed Dec 10 08:27:31 EST 2003


On Wednesday 10 December 2003 11:32, Dieter Maurer wrote:

> Unless I do not here strong objections, I will extend
> the ZODB such that:

+1 

>   *  setting a "_v_" attribute of a "Persistent" object
>      sets a "_p_volatile" attribute.

Im not keen on the name, because objects in this state are _less_ volatile 
than others. I think the name should say "lives until the end of a full 
transaction", but I dont have a good suggestion.

>   *  "incrGC" gets an argument "flushVolatile", defaulting to "1".

I think the default should be 0. I guess Im not the only one to have some code 
that calls incrGC to nudge the cache mid transaction. Lets make the full 
transaction boundary special - it can pass the non-default value.



I would like you to sketch out your cache purging implementation strategy. The 
most obvious approach may problems relating have efficiency and cache 
fairness.


-- 
Toby Dickenson




More information about the ZODB-Dev mailing list