[Feature]: combine all graph views into one - LFX Mentorship

[yurishkuro]

For the application process please refer to this issue.

• Mentor: @yurishkuro
• Mentee: @prathamesh-mutkure

Requirement

As a user of Jaeger UI I want to have a consistent experience with service graph views.

Problem

Jaeger currently uses three distinct renderings for service graphs

• the DAG view, good for small architectures (uses Cytoscape.js)
• the System view good for large architectures (uses react-viz-force-graph)
• the internal Plexus lib used for deep dependency graphs (uses D3 and graphviz rendering engine)

Proposal

To reduce the different code paths and provide a consistent user experience we should converge on a single graph visualization.

Plexus is the likely choice, but it needs to be benchmarked on large graphs (the System view can handle several thousand nodes). We can also consider server-side filtering (proposed in #784).

We may also have to address #1081 (deprecated graphing libs)

Some Evaluation Criteria

• is there support for stable / deterministic layouts? This is mandatory for graph views of a single trace, where nodes are arranged according to time and depth.
• how "future proof" is the library - actively supported, big contributor community, etc. (including its own dependencies). Are there big projects that use it?
• how many new dependencies it pulls
• can it handle large graphs (say up to 5k nodes)?
• does it allow customization of the graph? Including labels on edges, different viz for notes, interaction with nodes (left/right click)

[punithnayak]

Hey @yurishkuro I want to work on this for the upcoming LFX mentorship . Will go through good first Issues and suggest improvements if needed

[sivasathyaseeelan]

Hi @yurishkuro, I am very interested and definitely apply for this project in the upcoming LFX mentorship!!

[shashwatm1111]

Hey @yurishkuro , would love to contribute to this project under the LFX mentorship . Also going through the first good issues and will work upon them .

[yurishkuro]

@shashwatm1111 welcome

[Nupoor10]

Hello @yurishkuro, I would love to get involved with this project. I am currently getting familiar with Jaeger and Plexus. Looking forward to making an amazing contribution!

[ShivangRawat30]

Hello @yurishkuro, I would love to contribute to this project under the LFX mentorship.

[PranitPatil03]

Hello @yurishkuro👋

I'm Pranit Patil, a student pursuing a Bachelor's degree in Computer Science at Mumbai University. Recently, I completed my internship at hubx.ai, where I worked as a full-stack developer intern. Prior to that, I gained experience as a front-end developer intern at Tech Cryptors.

My skills include JavaScript, ReactJS, Node, Express, MongoDB, and CSS libraries, such as Bootstrap. Through various projects, including full-stack ones, I have honed my expertise in full-stack development

I'm keen on participating in the LFX mentorship program and becoming an active community member. I kindly request your assistance in guiding me through project discussions.

Thank you.

[yurishkuro]

@prathamesh-mutkure's application was selected for this project. Congratulations!

[anikdhabal]

congrats @prathamesh-mutkure

[yashrsharma44]

Plexus is the likely choice, but it needs to be benchmarked on large graphs (the System view can handle several thousand nodes). We can also consider server-side filtering (proposed in #784).

Do we have any sample(s) of data, which we can use to run the benchmarks against?

[yurishkuro]

not to my knowledge. We have a storage integration test that saves a trace with 10k spans, but it just generates that on the fly. I would do the same here - a fake trace is not difficult to generate. It's a bit tricky to decide on the good way to connect different nodes so that we don't test on a contrived cases like a graph which is a single very long branch.