On 9/22/11 2:45 AM, Thomas Diesler wrote:
Yes, there are many config values that would require a restart of
respective subsystems or services. This in turn would cause an yet
undefined ripple to all the dependees.
Also, it is possible that a subsystem cannot decide alone whether a
write operation at runtime that would bounce a certain service is
permissible. For example: Service A has property foo. Changing foo,
requires A to restart. B depends on A. B cannot be restarted. Changing
foo would hence require a server restart, which cannot be decided by
looking at service A alone.
I would suggest to be very selective of what can be changed at runtime
and only allow write access through the management API to properties
that are known to be safe.
Yeah, I agree, if "changed at runtime" means "restart a service to
effect the change." See my recent response to Jaikiran on this thread;
it's very hard to reason about whether it's safe to restart a service.
We do however, need to support changing pretty much every attribute via
the management API. People should never be required to edit xml.
FYI, see "Applying Updates to Runtime Services" on [1] for more details
on how things are currently meant to work. It's the "resource-services"
case discussed there that IMO is most problematic. We currently only
have one such case though.
[
1]http://community.jboss.org/docs/DOC-16317
Changes that require server restart can be
collected and marked as such.
On 09/21/2011 05:43 PM, Brian Stansberry wrote:
> It's part of what I'm doing.
>
> I believe changes to these pool configs can't be applied directly to the
> runtime though, at least not without a header associated with the
> request indicating its ok to bounce the PoolConfigService. I presume
> bouncing that service will in turn bounce any EJB services that use that
> config. (If not, let me know, as it probably means we could just bounce
> it by default.)
>
> The StrictMaxPoolConfig stored in the PoolConfigService could be made
> mutable, in which case any EJB deployments deployed in the future would
> get the new config values, but any existing pools would remain as is.
> That seems like an unintuitive semantic for this particular use case
> though.
>
> On 9/21/11 9:01 AM, Jaikiran Pai wrote:
>> That looks like a separate issue. The bean instance pool attributes have
>> currently been modelled as read-only. More of an oversight rather than
>> intentional.
>>
>> I think Brian is already working on
>>
https://issues.jboss.org/browse/AS7-913 so this could be included as
>> part of that work. Else I'll take this up separately.
>>
>>
>> -Jaikiran
>> On Wednesday 21 September 2011 06:59 PM, Andrig Miller wrote:
>>> Why is it read-only? Seems like you should be able to set to a
>>> different value too. Another bug?
>>>
>>> Andy
>>>
>>> ------------------------------------------------------------------------
>>>
>>>
>>> *From: *"Jaikiran Pai"<jpai(a)redhat.com>
>>> *To: *jboss-as7-dev(a)lists.jboss.org
>>> *Sent: *Wednesday, September 21, 2011 4:39:34 AM
>>> *Subject: *Re: [jboss-as7-dev] [Detyped API Description] Example
>>> of multiple values for "allowed"?
>>>
>>> It's a bug. It should actually look like:
>>>
>>> "timeout-unit" => {
>>> "description" => "The instance acquisition
>>> timeout
>>> unit",
>>> "type" => STRING,
>>> "required" => false,
>>> "default" => "MINUTES",
>>> "allowed" => [
>>> "HOURS",
>>> "MINUTES",
>>> "SECONDS",
>>> "MILLISECONDS"
>>> ],
>>> "access-type" => "read-only",
>>> "storage" => "configuration"
>>> }
>>>
>>>
>>> I'll send a pull request with the fix.
>>>
>>> -Jaikiran
>>> On Wednesday 21 September 2011 04:02 PM, Jaikiran Pai wrote:
>>> > The model for EJB3 does allow more than one timeout unit type
>>> (SECONDS,
>>> > MINUTES and MILLISECONDS etc...). So that output doesn't look
>>> right. Let
>>> > me take a look why.
>>> >
>>> > -Jaikiran
>>> > On Wednesday 21 September 2011 03:53 PM, David Bosschaert wrote:
>>> >> Hi all,
>>> >>
>>> >> I'm looking at the EJB3 pools timeout-unit description, which
>>> is the
>>> >> following:
>>> >> "timeout-unit" => {
>>> >> "description" => "The instance acquisition timeout
unit",
>>> >> "type" => STRING,
>>> >> "required" => false,
>>> >> "default" => "MINUTES",
>>> >> "allowed" => "MILLISECONDS",
>>> >> "access-type" => "read-only",
>>> >> "storage" => "configuration"
>>> >> }
>>> >>
>>> >> Is there an example of an attribute somewhere where
'allowed'
>>> has more
>>> >> than 1 value? I'd like to test a console gui piece against it.
>>> >>
>>> >> Thanks,
>>> >>
>>> >> David
>>> >> _______________________________________________
>>> >> jboss-as7-dev mailing list
>>> >> jboss-as7-dev(a)lists.jboss.org
>>> >>
https://lists.jboss.org/mailman/listinfo/jboss-as7-dev
>>> >
>>> > _______________________________________________
>>> > jboss-as7-dev mailing list
>>> > jboss-as7-dev(a)lists.jboss.org
>>> >
https://lists.jboss.org/mailman/listinfo/jboss-as7-dev
>>>
>>> _______________________________________________
>>> jboss-as7-dev mailing list
>>> jboss-as7-dev(a)lists.jboss.org
>>>
https://lists.jboss.org/mailman/listinfo/jboss-as7-dev
>>>
>>>
>> _______________________________________________
>> jboss-as7-dev mailing list
>> jboss-as7-dev(a)lists.jboss.org
>>
https://lists.jboss.org/mailman/listinfo/jboss-as7-dev
>
--
Brian Stansberry
Principal Software Engineer
JBoss by Red Hat