Hi Romain,
We are holding other work with this discussion.
Can we agree that this is good enough for a 1st version and move on with
a follow up PR?... It's not going to be worse than starting SE tasks
inside the container, like we have now.
Also, releasing Safegard 1.0.1 would be nice. There is unreleased code
in there that this work needs. We can live with the SNAPSHOT in the
meantime because there is no prediction of work for that SNAPSHOT.
Cheers.
Bruno Baptista
https://twitter.com/brunobat_
On 23/11/18 15:41, Romain Manni-Bucau wrote:
Le ven. 23 nov. 2018 à 16:34, Bruno Baptista <bruno...@gmail.com> a écrit :
Hi Romain,
About "The point is not the cdi bean but the executor. So high level you
deploy an
app not using safeguard but it being present and you ensure the container
has no executor resource instantiated (you will get one (the facade))."
Sorry Romain, I still don't understand how the code in the PR can
possible affect something not using the FT API or Safeguard in particular.
I think the code is ok but it uses assumptions which are likely not obvious
and it is not tested so next commit will break it - since this code must be
reworked anyway - and you will not see it.
So better to ensure the build guarantee all the outcome we want for end
users.
In relation to the Managed executor... What you say makes sense but I
wonder how likely it is to happen and if it's enough to hold the PR. Do
you have a custom executor example somewhere?
We have some in the core tests you can reuse. But long story short you run
your test, don't use safeguard and guarantee in @Test by looking up the
resource directly using internals (SystemInstance > ContainerSystem and so
on) that the instance is not yet instantiated. See for a test doing exactly
that:
https://github.com/apache/tomee/blob/master/container/openejb-core/src/test/java/org/apache/openejb/assembler/classic/LazyResourceTest.java#L41
To summarize:
1. CDI is lazy
2. we define the default executor as being lazy
3. we assume safeguard will not impact an app not using it
==> you must ensure that 3 didnt trigger an executor creation, it is fine
to rely on 1+2 (which means so "main" code)
Cheers
Bruno Baptista
https://twitter.com/brunobat_
On 23/11/18 15:14, Romain Manni-Bucau wrote:
Le ven. 23 nov. 2018 à 15:49, Bruno Baptista <bruno...@gmail.com> a
écrit :
Hi Romain,
Thanks for your comment.
The class doing the resource injection is lazy loaded, specifically
/FailsafeContainerExecutionManagerProvider/. I did verify it in
development but no test was produced... And to say the truth I wouldn't
know how to validate if a bean has already been loaded or not. Can you
please provide a test example?
The point is not the cdi bean but the executor. So high level you deploy
an
app not using safeguard but it being present and you ensure the container
has no executor resource instantiated (you will get one (the facade)).
Please explain what do you mean by "MP-fault-tolerance executor for that
case if noone exists". It will exist, that's the whole purpose of this
PR. Can you please provide an example where a
/ManagedScheduledExecutorService/ will not be present?
You can see it as "don't let it default to a random executor". This is
the
current behavior. So here is what can happen:
1. The user doesnt use any executor -> it defaults -> it is ok
2. The user uses one or more executors for his app -> it defaults to it
->
it messes up the app and does not have the expected setting
Case 2 is important cause it can really make it not functional and even
lead to locks in some cases so better to not let it happen and just
create
a safeguard executor if
the user didnt specify he wants safeguard to use the executor
"'mysafeguardexecutor".
This is why the config is important and I mentionned it early even if it
is
not the most sexy part to do, I agree.
Cheers
Bruno Baptista
https://twitter.com/brunobat_
On 23/11/18 14:39, Romain Manni-Bucau wrote:
It's lazily loaded, so no worries on that regard.
What is "it" here? :)
Conretely the bean instantiation yes cause it is normal scoped and the
resource too cause it is by default lazy in tomee (service-jar.xml) but
it
is worth a test that prevent regression on that behavior IMHO, I didn't
catch on in the PR.
Concretely in terms of container we can want to create a dedicated
MP-fault-tolerance executor for that case if noone exists and the user
didn't specify one cause this default behavior (cumulated with tomee
defaulting on @Resouce) will make this not reliable which is quite
ridiculous when you think about it for something about failt tolerance.
This is why it should be in before next release. Now if you do the PR
next
week it is fine, was not to do it today but to ensure it is not merged
and
the enthusiasm makes it forgotten.
Romain Manni-Bucau
@rmannibucau <https://twitter.com/rmannibucau> | Blog
<https://rmannibucau.metawerx.net/> | Old Blog
<http://rmannibucau.wordpress.com> | Github <
https://github.com/rmannibucau> |
LinkedIn <https://www.linkedin.com/in/rmannibucau> | Book
<
https://www.packtpub.com/application-development/java-ee-8-high-performance
Le ven. 23 nov. 2018 à 15:18, Jonathan Gallimore <
jonathan.gallim...@gmail.com> a écrit :
Maybe add those config options in a second PR? What do you think?
Jon
On Fri, Nov 23, 2018 at 2:01 PM Bruno Baptista <bruno...@gmail.com>
wrote:
Hi Romain,
In the end I decided to simply use the server default, for now.
It will only be used if annotations are called in the code. It's
lazily
loaded, so no worries on that regard.
Cheers.
Bruno Baptista
https://twitter.com/brunobat_
On 23/11/18 12:31, Romain Manni-Bucau wrote:
Didnt you want to make the pool configurable and not instantiated if
not
used?
Le ven. 23 nov. 2018 13:20, Daniel Cunha <daniels...@apache.org> a
écrit :
Hey Bruno,
awesome! It really sounds good! I just push my +1 :)
Em sex, 23 de nov de 2018 às 06:44, Bruno Baptista <
bruno...@gmail.com>
escreveu:
Thanks!
Additionally, I've requested a Safeguard 1.0.1 release. we
shouldn't
be
using snapshots.
Cheers
Bruno Baptista
https://twitter.com/brunobat_
On 22/11/18 19:30, Roberto Cortez wrote:
Cool! Thank you.
I’ll have a look.
On 22 Nov 2018, at 19:08, Bruno Baptista <bruno...@gmail.com>
wrote:
Hi!
I think the code is ready. Can some of you please review this
pull
request:
https://github.com/apache/tomee/pull/201
Related to:TOMEE-2278 <
https://issues.apache.org/jira/browse/TOMEE-2278>-
Use Managed Executor with Safeguard Fault Tolerance lib
Cheers.
--
Bruno Baptista
https://twitter.com/brunobat_
--
Daniel "soro" Cunha
https://twitter.com/dvlc_
