Back to incidentsExtracted by NVIDIA Nemotron Generated by NVIDIA Nemotron
GitHub — August 17, 2026
Client-side retry behavior amplified traffic during recovery, increasing load on already degraded services and contributing to a cascading outage.
Duration7h 47m
ImpactMultiple services
Root causeRetry storm
Sourcegithub.blog
1
Incident2
Failure DNA3
Exposure4
Resilience Test5
Mitigation6
VerifiedFailure DNA
—
Trigger
—
Mechanism
—
Signals
—
Recommended test
—
Exposure
—
Match confidence
—
—
10async function callUpstream(attempt: number) {
11if (upstreamFailed && attempt < MAX_RETRIES) {
12—
13}
14}
Why this matters
—
Resilience test
Prove the vulnerability
Incident-specific test derived from the GitHub failure pattern.
Initial requests
—
Upstream requests
—
Request amplification
—
Mitigation
—
—
demo/retryStormPatched.ts
+const GLOBAL_RETRY_BUDGET = INITIAL_REQUESTS * 2;
+let retriesUsed = 0;
11if (upstreamFailed && retriesUsed < GLOBAL_RETRY_BUDGET) {
+retriesUsed++;
12callUpstream(attempt + 1);
13}
Verified
Same resilience test. Same failure scenario.
Mitigation verified
Before
—
Request amplification
Failed
→
After
—
Request amplification
Passed
Your codebase has learned from GitHub's outage.
The same failure mechanism that caused real production impact was reproduced, mitigated, and verified before it became yours.
Incident discovery via TavilyFailure analysis and mitigation via NVIDIA Nemotron