[ML] File data visualizer: filter running inference endpoints

[jgowdyelastic]

When listing the inference endpoints available for the semantic text field, we should only list ones where the underlying model is deployed, otherwise the ingest can fail due to a timeout as the model is being deployed and an ML node spun up in the background.

This PR adds a check to the data_visualizer/inference_endpoints endpoint to ensure only sparse_embedding and text_embedding types are used and they have at least one allocation.
NOTE, the allocation check is currently commented out waiting on an es change. elastic/elasticsearch#115095

Also renames the endpoint from data_visualizer/inference_services -> data_visualizer/inference_endpoints
And renames variables which were incorrectly named "service" rather than "endpoint"

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[jeffvestal]

Another thing that would be nice to filter is only to show inference APIs for embedding services.
However, this may be tricky with external services.

[peteharverson]

Tested and LGTM.

One observation - in my local testing, uploading a PDF file (500 lines, 160KB) for the first time can take a couple of minutes - presumably as a new .elser-2 deployment is created. During this time, the progress page stays in step 4 (Uploading data) with no visible sign of anything happening. Is there anything we can do to improve the messaging here to provide more details on what is happening?

[jgowdyelastic]

@jeffvestal

Another thing that would be nice to filter is only to show inference APIs for embedding services.
However, this may be tricky with external services.

I'm now filtering for sparse_embedding and text_embedding types

[jeffvestal]

I'm now filtering for sparse_embedding and text_embedding types

Nice! Thanks

[jgowdyelastic]

One observation - in my local testing, uploading a PDF file (500 lines, 160KB) for the first time can take a couple of minutes - presumably as a new .elser-2 deployment is created. During this time, the progress page stays in step 4 (Uploading data) with no visible sign of anything happening. Is there anything we can do to improve the messaging here to provide more details on what is happening?

There will be a change going into es for 8.16 which fixes num_allocations. We can use that to ensure only deployed models are available for use. This will mean the pause you're seeing here for download/deployment will not happen.

In the future we could add a step to trigger this auto deployment by calling the infer endpoint during upload and report the status of the download and deployment as an addition step in the upload progress. #196696

[elasticmachine]

💛 Build succeeded, but was flaky

• Buildkite Build
• Commit: 9044218

Failed CI Steps

• FTR Configs #60

Test Failures

• [job] [logs] FTR Configs #60 / Rule execution logic API Detection Engine - Execution logic @ess @serverless Indicator match type rules, alert suppression Code execution path: events count is smaller than threats count should suppress an alert on real rule executions

Metrics [docs]

Async chunks

Total size of all lazy-loaded chunks that will be downloaded as the user navigates the app

id	before	after	diff
dataVisualizer	614.0KB	614.0KB	+2.0B

History

• 💚 Build #243572 succeeded f4dfb40

cc @jgowdyelastic