Software and Hardware Remediations
summary
Fleet runs more than 5 million software remediations per day and about 5k hardware remediations per day. Detection uses a local agent plus on‑demand remote checks and a scheduler to generate alarms. Software remediations execute via FBAR (Facebook Auto-Remediation) with extracted customer remediation services, microservices for locking, config rollout and rate limiting (migrating from Python to C++). Hardware escalation uses a decision engine, RepairBrain, that sequences actions, queries detection after each step and augments rules with an imperfect ML model for prioritization and automation.
AI-written summary · about
chapters (9)
- 0:00Hardware and software remediations
- 1:10Daily remediation scale at Meta
- 2:56Remediation pipeline overview
- 4:42Detection challenges
- 8:35FBar software remediation
- 11:42Modernizing FBar architecture
- 13:28Reaper Brain hardware remediation
- 16:42Remediation pipeline summary
- 17:25Future detection and remediation
related talks
20mBridging the Safety Gap from Scripts to Full Auto-Remediation
51mRemediation-Aware Reachability: Patching Containers, Prioritizing with Agentic-CTI, and Scaling...
14mThe Challenges Of Remediation During A Cyber Incident
35mRemediate The Flag - Practical AppSec Training Platform
13mBeyond Manual Firefighting
52mAnatomy of a Vulnerability Response - A View from the Inside
46mRugged Software Engineering
49mWhen Hardware Gets Pwned Mechanics Of Detection, Response And Recovery