question-mark
Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Implement a shaded in-proc Pravega cluster

See original GitHub issue

Problem Description The work of #573 provides an in-process Pravega cluster (LocalPravegaEmulator) for use in client-side integration tests. Dependency conflicts are likely to happen in such an environment, for example when writing Flink connector tests that use Flink’s in-process cluster.

A solution to the conflict is to build a shaded JAR of the Pravega emulator that isolates most of Pravega’s dependencies. Ideally the library would leave only io.pravega.* unshaded.

Suggested Fix I suggest the overall dependency tree be (as seen from the app perspective, with shaded deps elided):

flink-job
├─ flink-streaming-java (provided)
├─ pravega-connector (compile)
│  └─ pravega-client (compile)
└─ pravega-standalone (intTest)
   └─ pravega-client (compile)

To achieve the above, most of the shading work that is presently done in the connector should move to the client library.

Issue Analytics

  • State:closed
  • Created 6 years ago
  • Comments:5 (5 by maintainers)

github_iconTop GitHub Comments

1reaction
EronWrightcommented, Nov 22, 2017

Some exposition on the rationale for my suggestion of moving shading (of Netty, etc) to the client library.

When we apply shading to a library, our goal is to prevent a conflict with the application that uses the library by hiding its dependencies. I’ll use Netty as an example of a dependency that we wish to hide.

Hiding Netty has two aspects: 1. relocating the Netty classes to io.pravega.shaded.netty. 2. removing Netty as a listed dependency in the library’s POM. If we did (1) but not (2), shading would be ineffective because the application would still see Netty as a transitive dependency.

Now consider the pravega connector. Today it relocates all the dependencies and removes the client library from its effective POM. If the client were still listed as a dependency of the connector, the app would again see Netty and other client dependencies (which we want to avoid).

The ideal design would allow the connector to list the client as a dependency in the effective POM. Likewise pravega-standalone should list the client as a dependency. To achieve this, the client itself should take responsibility for shading its dependencies. No further shading would then be needed in the connector. Meanwhile the standalone library would need additional shading of the server dependencies.

0reactions
andreipaduroiucommented, Dec 13, 2018

I closed it for inactivity for more than 1 year and no assignees that are actively working on the project.

Read more comments on GitHub >

github_iconTop Results From Across the Web

Cluster Dependencies - Exploring Pravega
A Pravega Cluster consists of a set of services (Zookeeper, Bookkeeper, Segment Store, Controller and Long-Term Storage) that interact among each other.
Read more >
Pravega Kubernetes Operator - GitHub
The Pravega Operator manages Pravega clusters deployed to Kubernetes and automates tasks related to operating a Pravega cluster.The operator itself is built ...
Read more >
From Big Data to Fast Data: Efficient Stream Data Management
It addressed the volume challenge by distributing partitions of input data across the cluster. This allowed to deploy and scale a simple ...
Read more >
Argumentirano strojno uˇcenje - Artificial Intelligence Laboratory
not be in the same cluster [WCRS01; LTJ04]. It was shown that clustering becomes more robust with the use of prior knowledge. Combining...
Read more >
(PDF) Argument-Based Machine Learning | Ivan Bratko
We implemented the ABCN2 algorithm, an argument-based extension of the ... It was shown that clustering becomes more robust with the use of...
Read more >

github_iconTop Related Medium Post

No results found

github_iconTop Related StackOverflow Question

No results found

github_iconTroubleshoot Live Code

Lightrun enables developers to add logs, metrics and snapshots to live code - no restarts or redeploys required.
Start Free

github_iconTop Related Reddit Thread

No results found

github_iconTop Related Hackernoon Post

No results found

github_iconTop Related Tweet

No results found

github_iconTop Related Dev.to Post

No results found

github_iconTop Related Hashnode Post

No results found