[ML] Explain Log Rate Spikes: Progress bar stuck at 0% then jumps to 80%

[walterra]

Kibana version: 8.4.0 BC

Describe the bug:

Esp. on cloud, once a user starts the analysis, the progress bar is stuck for a while at 0% —. It's not obvious that the analysis is running and there's no "loading state message" after the —.

Expected behavior:

The progress bar should have a "loading state message" and progress should happen earlier.

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[walterra]

This is most likely related to the proxy used in cloud. It looks like the proxy is buffering the first 4KB and only then starts streaming. It replaces the Transfer-Encoding: chunked header with Content-Length: ... for responses smaller than 4KB. While further investigation is ongoing, a short term fix for this could be to send 4KB of dummy data first to trigger proper streaming.

[walterra]

Experiments done in #139637 surfaced that every time the proxy buffers up to 4KB after a flush, not just on the first 4KB.

[vadimkibana]

Cloud issue: https://github.com/elastic/cloud/issues/106308

[walterra]

Closing this one, created a follow up issue to keep track of removing the workaround once proxies support smaller payloads: #141194.