Bugzilla – Bug 4744
SPEC - time-limit needs a unit of measure
Last modified: 2013-03-20 03:07:52 UTC
Within the chunk element, the time-limit needs to define the unit of time it's measuring. I would recommend milliseconds to be consistent with the rest of the way Java measures time.
I agree it needs to be specified. And I recognize the Java precedent of milliseconds. But do you really believe conforming to precedent is the right thing here? Batch times are usually seconds if not minutes. E.g.
E.g. a 3 minute chunk limit is 180000.
Using milliseconds will make the JSL less readable IMO. That would be fine if times less than a second where necessary. Such small times are typical for OLTP but not batch.
I think seconds as the unit of measure makes more sense. Opinions?
I think Michael's comment implies we need to add a new attribute, like time-limit-unit, and these 2 together can accurately define the time limit. We can default time-limit-unit to a common value such as minutes or seconds, while allowing use of other units like millis, nanos, or days. For example, ejb 3.1 stateful-timeout uses this structure with minutes as default unit:
I don't think we should make a change given the lateness of the date. There are other ways to achieve "quick checkpointing" (custom or item-based), and I think we should prioritize this low.
At the least, we do need a unit of measure specified in the spec.
It will have to come in a future revision. Anyone that needs to checkpoint more frequently than 1 second can do so with a checkpoint algorithm artifact.