Skip to main content
abstractalgorithms.devAbstract Algorithms
← All Problemsarchitecturemedium~30 min

Prometheus High-Cardinality Metric Explosion Exhausting TSDB RAM

Microservice instrumenting an HTTP request metric with user_id or transaction_id as a Prometheus metric label.

Scope, Requirements & Constraints

Microservice instrumenting an HTTP request metric with user_id or transaction_id as a Prometheus metric label. Failure cascade dynamic: Object Churn → Unbounded Off-Heap → GC Freezes → 💥.

Core Functional Requirements

  • ✓Operational Invariant: "Metric labels must strictly contain low-cardinality bounded enumerations; never place user IDs or UUIDs in metrics."
  • ✓Production Hardening: Memory Never place high-cardinality values (user IDs, email addresses, UUIDs) in Prometheus metric labels. Restrict labels strictly to low-cardinality enumerations (status code, HT
  • ✓Prevent cascading failure by isolating downstream degradation at the ingress boundary.

Scale & Latency Constraints

  • —Incident Scenario Baseline: Microservice instrumenting an HTTP request metric with user_id or transaction_id as a Prometheus metric label.
  • —Failure Cascade Path: Object Churn → Unbounded Off-Heap → GC Freezes → 💥
  • —Sub-millisecond rejection / shed latency for overloaded requests

Reference Architecture & System Blueprint

Memory Never place high-cardinality values (user IDs, email addresses, UUIDs) in Prometheus metric labels. Restrict labels strictly to low-cardinality enumerations (status code, HTTP method, endpoint pattern). Send high-cardinality data to distributed tracing spans. Invariant: "Metric labels must strictly contain low-cardinality bounded enumerations; never place user IDs or UUIDs in metrics."

System Architecture Blueprint
Production Topology
Prometheus High-Cardinality Metric Explosion Exhausting TSDB RAM Architecture Blueprint
Architecture Flow NarrativeIngress → Bounded Admission Controller → Adaptive Concurrency / Queue Gate → Protected Worker Engine; Overload Shed Path → HTTP 429 / Circuit Fallback (<0.1ms).

Hardened Implementation Steps

  1. 1Ingress Boundary Defense: Memory Never place high-cardinality values (user IDs, email addresses, UUIDs) in Prometheus metric labels. Restrict labels strictly to low-cardinality
  2. 2Failure Containment: Eliminate unconstrained buffering and propagate physical backpressure.
  3. 3Invariant Enforcement: Metric labels must strictly contain low-cardinality bounded enumerations; never place user IDs or UUIDs in metrics.

Production Code Contrast & Implementation

java
// HARDENED: Adaptive tail-based sampling with non-blocking ring buffer drop
Span span = Tracer.startSpan("process");
// Retain 100% of error spans (5xx) and latency outliers (>P95); sample 0.1% of nominal 200s
spanCollector.offerWithDropPolicy(span, DropPolicy.DROP_OLDEST_NOMINAL);

Key Decisions & Trade-Off Analysis

Architectural comparison of naive unconstrained allocations versus production-hardened mechanical sympathy.

Architectural Decisions & Trade-Offs

  • ⚖Ingress Admission: Naive unconstrained buffering causes queue bloat and GC exhaustion. Mitigation: Bounded ingress gate with fast non-blocking shedding.
  • ⚖Resource Strategy: Static capacity allocations collapse under brownouts. Mitigation: Dynamic adaptive limits scaling to real-time physical bounds.
  • ⚖Outage Impact: Cascading lock convoys and memory saturation. Mitigation: Isolated failure domains keep P99 latency flat and protected.

Alternatives Evaluated

  • →Naive Approach: OOMKill (Exit 137) With 5 million active users, the time series database creates 5 million distinct time-series streams. Prometheus TSDB memory consum
  • →Production-Hardened Pattern: Memory Never place high-cardinality values (user IDs, email addresses, UUIDs) in Prometheus metric labels. Restrict labels strictly to low-cardinality

Scale Bottlenecks & Operational Hazards

Operational Hazards & Mitigations

  • ⚠Cascading Failure: OOMKill (Exit 137) With 5 million active users, the time series database creates 5 million distinct time-series streams. Prometheus TSDB memory consumption explodes past 64 GB; Prometheus crashes into OOM loops, blinding SREs during incidents.
  • ⚠Breakdown Chain: Object Churn → Unbounded Off-Heap → GC Freezes → 💥
  • ⚠Shedding Failure: Without fast rejection at the boundary, queued dead work triggers multi-second GC pauses, CPU thrashing, and cascading 504 Gateway Timeouts.

Changed Constraints & System Extension

How does the system enforce physical resource boundaries under sudden downstream latency brownouts?

Extension Implementation Steps

  1. 1Track real-time latency gradients and inflight concurrency.
  2. 2Reject excess traffic at the ingress boundary within <0.1ms.
  3. 3Monitor P99 latency and verify autonomous recovery when downstream pressure subsides.

Follow-Up Trade-Offs

  • ⚖Architectural Invariant: "Metric labels must strictly contain low-cardinality bounded enumerations; never place user IDs or UUIDs in metrics."
  • ⚖Physical Resource Boundary: Dynamic admission control ensures worker threads never buffer dead requests during downstream brownouts.

Interactive Assessment & Knowledge Check

Test your understanding of the architecture decisions, data patterns, and trade-offs used in Prometheus High-Cardinality Metric Explosion Exhausting TSDB RAM.

Interactive Practice Quiz•Question 1 of 3
intermediate

Distributed Tracing: fundamentals

When is Distributed Tracing the right default for a production team?