---
type: "raw_signal"
id: "taming-the-ml-firehose-scaling-feature-consistency-a8c02d"
slug: "taming-the-ml-firehose-scaling-feature-consistency-a8c02d"
title: "Taming the ML Firehose: Scaling Feature Consistency"
description: "Uber describes a feature-logging framework that records the exact feature values consumed during online inference and makes those records the canonical source for subsequent model training."
retrieval_nugget: "Uber describes a feature-logging framework that records the exact feature values consumed during online inference and makes those records the canonical source for subsequent model training."
observed_at: "2026-09-23"
record_date: "2026-09-23"
date_kind: "observed_at"
why_it_matters: "At reported peak prediction traffic of 8 million QPS, logging every prediction would create hundreds of billions of records per day, with an estimated 1.7 PB daily footprint. Uber therefore logs allow-listed model features, aliases verbose names to integer IDs, and joins prediction logs with client impression events in Flink so that training records retain predictions actually seen by users. The c"
novelty: "notable"
verification_level: "source-linked"
signal_type: "research"
source_platform: "uber.com"
topics: []
entities: []
related_patterns: []
source_url: "https://www.uber.com/us/en/blog/taming-ml-firehose"
source_urls: ["https://www.uber.com/us/en/blog/taming-ml-firehose"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "signal:taming-the-ml-firehose-scaling-feature-consistency-a8c02d"
status: "published"
visuals: []
routes: {"html":"https://newruntime.com/signals/taming-the-ml-firehose-scaling-feature-consistency-a8c02d/","markdown":"https://newruntime.com/signals/taming-the-ml-firehose-scaling-feature-consistency-a8c02d.md","json":"https://newruntime.com/signals/taming-the-ml-firehose-scaling-feature-consistency-a8c02d.json"}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
---

# Taming the ML Firehose: Scaling Feature Consistency

## Retrieval answer

Uber describes a feature-logging framework that records the exact feature values consumed during online inference and makes those records the canonical source for subsequent model training.

Uber describes a feature-logging framework that records the exact feature values consumed during online inference and makes those records the canonical source for subsequent model training. The aim is to reduce training-serving inconsistency caused by separate online and offline computation paths.

At reported peak prediction traffic of 8 million QPS, logging every prediction would create hundreds of billions of records per day, with an estimated 1.7 PB daily footprint. Uber therefore logs allow-listed model features, aliases verbose names to integer IDs, and joins prediction logs with client impression events in Flink so that training records retain predictions actually seen by users. The company says the allow list cuts payload size by 4–5×.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Raw Signals](https://newruntime.com/signals.json): Open every source-linked raw signal.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)