>>> On 25 Jan 2011, at 17:12, Manik Surtani wrote:
>>>> How's this for a solution:
>>>>
>>>> 1) Test suite remains the same
>>> I don't think performance is affected by using JBossTM for test runs,
I'll need to check by running it
>>
>> It should be quicker. There is no maintenance of a persistent tx log.
> Not necessarily: the log would only be written for tx that span more than an
participant (have more than 1 XAReource enlisted). AFAIK there are very few tests like
that in ISPN (I can only remember one). If only one participant in tx the TM would
doesn't have to persist the log to disk.
There are a few tests testing transactions that span different cache instances. Each one
would be considered as its own RM.
indeed, but very few. I'll measure and see
the difference, that should tell us beyond doubt :)
Interesting! To my excuse that was added recently while I was on vacation :-)
I've just looked at the code and it is broken:
- the readonly flag is set on prepare with the value returned by last enlisted XAResouce
(bad!!)
- so on commit, if last enlisted XAResource returned r/o status, no commit is called on
*any* other enlisted resource (bad! bad! bad!)