I do not feel that we need to restrict the names of jar files that users may
deploy. GEODE-7436 does the most reasonable thing in the vast majority of
cases, but a user is always free to override the default logic, either by
manually un-deploying some existing jar before deploying a new one, or renaming
one of them if they truly require both of two lookalike jars deployed
simultaneously. This is consistent with the ideal of making simple things
easy, but hard things possible.
I fail to see a problem here that rises to the level of justifying "breaking
changes" with "no transition ability".
@Udo, the work you are citing in conjunction with this appears to be related to
classloader changes. Can you clarify whether your proposed restrictions on jar
names are essential to implement your classloader changes, or just an unrelated
thing you happened to notice in the course of that work?
On 10/7/20, 4:45 PM, "Udo Kohlmeyer" <u...@vmware.com> wrote:
@Owen, not sure if I'd use "harmless". I'd use "unlikely", rather than
"harmless", as it can still have harmful consequences.
I think the "intuitive" nature of the versioning means we have to have a
standard jar file format, so that the system can intuitively understand that
"some-jar-0.2" is the update to "some-jar-0.1". There HAS to be some form of
format for the auto-version-update to work, otherwise one could never compare.
Maybe I'm just trying to be more explicit. That way, any implementation of
the "deploy jar" functionality can rely on a single fact, that the jar files
will be in a known format.
The alternative is maven-like in behavior... then we don't have to deal
with explicit file formats. When a jar is deployed, the feature behaves more
maven-like, and "artifact name", "version" and file path is required. That way,
the jar file does not ever have to meet the required format. All information
required has been provided by the user. Removes the false-positives that could
be introduced.
As there is no public API here, the only change we would have to deal with
is the gfsh scripting, which I assume is not so horrible.
Anyway.. just thoughts..
--Udo
On 10/8/20, 9:57 AM, "Owen Nichols" <onich...@vmware.com> wrote:
The goal of the GEODE-7436 change is that when user deploys
some-jar-0.1, then later deploys some-jar-0.2, we will do the intuitive thing
and treat it as an update. In Geode 1.11 and earlier, we instead wound up with
*both* deployed, which is bad!
In the weird example below of spark-network-common_2.11-2.3.1.jar, it
is harmless that we stem it as spark-network-common_2 (*if* for some very crazy
reason a user NEEDS both spark-network-common_2.11-0.0.0.jar and
spark-network-common_2.12-0.0.0.jar deployed side-by-side, they can simply
rename the jar to something more sane)
On 10/7/20, 3:45 PM, "Anthony Baker" <bak...@vmware.com> wrote:
Given the wide variety of filenames possible do we even need a
classification scheme? IOW, why not just take what the user gives us and say
thank you :-). Is this restriction imposed by our *implementation* choices?
Anthony
> On Oct 7, 2020, at 3:24 PM, Jinmei Liao <jil...@vmware.com> wrote:
>
> Wait, that reason doesn't make much sense either. Dale/Darrel, do
you remember why we did what we did?
>
> On 10/7/20, 3:12 PM, "Jinmei Liao" <jil...@vmware.com> wrote:
>
> I believe we did this for a reason, can't remember exactly
what though. Most probably drive by user's existing filenames. I believe we are
probably concerned that user's jar name might contain "_" or "-" themselves,
like common-logging.jar etc. So we had to resort to finding the first "."
followed by a digit to determine where the version pattern begins.
>
> On 10/7/20, 1:44 PM, "Udo Kohlmeyer" <u...@vmware.com> wrote:
>
> Hi there Geode Dev List,
>
> Whilst doing work on
GEODE-8466<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fissues.apache.org%2Fjira%2Fbrowse%2FGEODE-8466&data=04%7C01%7Conichols%40vmware.com%7Cc112712a47c5466beb1c08d86b1b073f%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C637377111154606914%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=YoZbWLnP9p4WilvrQCA3X0b4lJ8%2FEkWRdqAlRc4uTq8%3D&reserved=0>
and looking at the functionality that the
ClassPathLoader.java<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fapache%2Fgeode%2Fblob%2Fdevelop%2Fgeode-core%2Fsrc%2Fmain%2Fjava%2Forg%2Fapache%2Fgeode%2Finternal%2FClassPathLoader.java&data=04%7C01%7Conichols%40vmware.com%7Cc112712a47c5466beb1c08d86b1b073f%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C637377111154606914%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=nGYdgpQpHHqFkQMPiHk7WpTc1wQHfeMx28oHFlWKOyQ%3D&reserved=0>,
JarDeployer.java<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fapache%2Fgeode%2Fblob%2Fdevelop%2Fgeode-core%2Fsrc%2Fmain%2Fjava%2Forg%2Fapache%2Fgeode%2Finternal%2FJarDeployer.java&data=04%7C01%7Conichols%40vmware.com%7Cc112712a47c5466beb1c08d86b1b073f%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C637377111154606914%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=%2B%2Fk2JB9eas%2FR50wQY1SqCN34PVWRFomiWD%2BQD0v9p40%3D&reserved=0>
and
DeployedJar.java<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fapache%2Fgeode%2Fblob%2Fdevelop%2Fgeode-core%2Fsrc%2Fmain%2Fjava%2Forg%2Fapache%2Fgeode%2Finternal%2FDeployedJar.java&data=04%7C01%7Conichols%40vmware.com%7Cc112712a47c5466beb1c08d86b1b073f%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C637377111154616914%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=4GMkWrRTi49NfojafHYrZiaylipAnNh14uVGkJUzlXg%3D&reserved=0>
provide around the “Deploy Jar” functionality, we came across some interesting
“supported” filename patterns.
>
> According to the
JarDeployerFileTest.java<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fapache%2Fgeode%2Fblob%2Fdevelop%2Fgeode-core%2Fsrc%2FintegrationTest%2Fjava%2Forg%2Fapache%2Fgeode%2Finternal%2FJarDeployerFileTest.java&data=04%7C01%7Conichols%40vmware.com%7Cc112712a47c5466beb1c08d86b1b073f%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C637377111154616914%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=WoHo5TtQdE87mp8AVtyEjLUBIWuqsHe9uLNgdYbC1mo%3D&reserved=0>
the “supported” formats are as follows:
>
>
assertThat(JarDeployer.getArtifactId("abc.jar")).isEqualTo("abc");
>
assertThat(JarDeployer.getArtifactId("abc-1.jar")).isEqualTo("abc");
>
assertThat(JarDeployer.getArtifactId("ab.c.1.jar")).isEqualTo("ab.c");
>
assertThat(JarDeployer.getArtifactId("abc.v1.jar")).isEqualTo("abc.v1");
>
assertThat(JarDeployer.getArtifactId("abc-1.0.snapshot.jar")).isEqualTo("abc");
>
assertThat(JarDeployer.getArtifactId("abc-1.0.v1.jar")).isEqualTo("abc");
>
assertThat(JarDeployer.getArtifactId("spark-network-common_2.11-2.3.1.jar"))
> .isEqualTo("spark-network-common_2");
> Which don’t make any sense. As the generally accepted norm
for a version jar file would be: “<artifact name>[ - <major> . <minor> .
<patch> - <Release Tag> ] .jar”. (note the syntax in red)
>
> I want to suggest that we DISCONTINUE supporting all jar
name formats other than the one mentioned above IMMEDIATELY. As the supported
name format is just “funky” but also wrong and can lead to misclassification of
the artifact name…. as some of you with a keen eye would have spotted already 😉
>
> For those who did not spot the mistake…
“spark-network-common_2.11-2.3.1.jar” is incorrectly classified and has the
WRONG artifact name. As “spark-network-common_2.11” is the correct artifact
name NOT “spark-network-common_2”!
>
> I would like to introduce this change with
GEODE-8466<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fissues.apache.org%2Fjira%2Fbrowse%2FGEODE-8466&data=04%7C01%7Conichols%40vmware.com%7Cc112712a47c5466beb1c08d86b1b073f%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C637377111154616914%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=VCbGclDoBMuXjwN6CbUD7PeAPWjX04Fi%2FsXuBFWgpfo%3D&reserved=0>.
This would be a “breaking” change, but we should change this sooner than
later. There is no transition ability here, as it would be too hard to have
Geode support both, as there is no simple way for the system to decide if the
name conforms to the “correct” format or not.
>
> DISCUSS!!!
>
> --Udo
>
>
>
