<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/github-s-postmortem-gets-dissected-vercel-tightens-the-free-tier-screws-g1beibdq5" -->

---
title: GitHub's postmortem gets dissected, Vercel tightens the...
description: GitHub's September 13 outage postmortem gets a proper autopsy today, exposing how a replica-lag-only safeguard missed a saturating primary. Vercel is cutting...
canonical: https://daily.dev/posts/github-s-postmortem-gets-dissected-vercel-tightens-the-free-tier-screws-g1beibdq5
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: GitHub's postmortem gets dissected, Vercel tightens the free-tier screws | daily.dev
og:description: GitHub's September 13 outage postmortem gets a proper autopsy today, exposing how a replica-lag-only safeguard missed a saturating primary. Vercel is cutting...
og:url: https://daily.dev/posts/github-s-postmortem-gets-dissected-vercel-tightens-the-free-tier-screws-g1beibdq5
og:image: https://api.daily.dev/og/posts/G1BeiBdQ5.png
og:image:alt: GitHub's postmortem gets dissected, Vercel tightens the free-tier screws
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# GitHub's postmortem gets dissected, Vercel tightens the free-tier screws

**[Backend Digest](https://daily.dev/sources/backend_digest)** · 4 min read · 0 upvotes · 0 comments

## Summary

GitHub's September 13 outage postmortem gets a proper autopsy today, exposing how a replica-lag-only safeguard missed a saturating primary. Vercel is cutting Hobby-tier deployment retention hard, from 10 protected production deployments down to 3, with immediate deletion instead of a 30-day grace period. AWS Lambda quietly 6x'd its execution timeout for a new instance type, opening the door to workloads that used to need Fargate or Batch. Elsewhere, a batch of inference-engineering talks make the case that orchestration, not model quality, is where the real infrastructure work is happening now.

## Content

**TLDR:** GitHub's September 13 outage postmortem gets a proper autopsy today, exposing how a replica-lag-only safeguard missed a saturating primary. Vercel is cutting Hobby-tier deployment retention hard, from 10 protected production deployments down to 3, with immediate deletion instead of a 30-day grace period. AWS Lambda quietly 6x'd its execution timeout for a new instance type, opening the door to workloads that used to need Fargate or Batch. Elsewhere, a batch of inference-engineering talks make the case that orchestration, not model quality, is where the real infrastructure work is happening now.

---

## GitHub's September 13 incident postmortem gets picked apart

An internal data-cleanup job saturated a shared database cluster's primary node, but the safeguard in place only watched replica lag, so it never caught the real problem. Long request-handler timeouts let blocked requests pile up on web servers, and a retry loop in token creation kept re-triggering failed writes, which dragged the outage out until responders shed load and killed the job. The write-up is worth reading for the broader point: every fix here (rate limits, new health signals, sharding) closes a known failure mode while adding a new layer of complexity you'll have to reason about during the next incident. [Read more](https://daily.dev/posts/2UYMWGOAD)

## Vercel guts Hobby-tier deployment retention

Free-tier projects now keep only the 3 most recent production deployments and 3 most recent deployments of any type, down from 10 protected production deployments plus separate preview protection. Worse, once you exceed the 10GB storage cap, unprotected deployments get deleted immediately instead of after a 30-day grace period. Vercel says deployment volume grew 6x year-over-year past 10 million/day, which is the real driver here, but if you're on Hobby and rely on old preview links for anything, go check your retention settings now. [Read more](https://daily.dev/posts/3VpDJhYJn)

## AWS Lambda's timeout jumps from 15 minutes to 90

Functions on Lambda Managed Instances now get a 90-minute execution ceiling, a 6x bump from the old 15-minute cap that's stuck around for years. It's aimed squarely at media processing, ETL, and AI inference jobs that previously needed a workaround like Step Functions, ECS, or Batch. AWS is upfront that Lambda still doesn't guarantee exactly-once processing, so longer-running functions need real idempotency work, not just a bigger timeout number. [Read more](https://daily.dev/posts/2zbRcpTYu)

## Inference talks converge on orchestration as the real bottleneck

A cluster of conference talks today all land on the same point from different angles. Meta engineers argue teams should optimize for cost per successful task instead of cost per token, framing an 'inference control plane' as the emerging Kubernetes-equivalent for GPU fleets. OpenAI describes splitting their load balancer into a control plane that computes globally optimal routing weights and a data plane that acts on cached weights fast, specifically to kill oscillations that were wrecking KV cache locality. Google's take is blunter: most LLM benchmarks are lying to you because of GIL-bottlenecked load generators and hidden settings like temperature=0, which is why they built inference-perf as a CNCF project to fix it. [Read more](https://daily.dev/feed-by-ids?id=bc6T4BC9w&id=LL852JSTC&id=ZGLKiUQZE)

---

## Also notable

- **Postgres row-level security has six quiet failure modes:** RLS moves tenant isolation into the database, but ENABLE without FORCE exempts the table owner, BYPASSRLS roles ignore RLS entirely, and permissive policies combine with OR so a later policy can silently widen access unless marked RESTRICTIVE. [Read more](https://daily.dev/posts/bM9k2DFoO)
- **Log indexing costs 87x more than ingestion on event-priced platforms:** A measured example shows 1 GiB of logs costs $0.10 to ingest on Datadog-style pricing but $8.75 to index, meaning shrinking log line size alone makes bills worse, not better, since it cuts bytes but not event count. [Read more](https://daily.dev/posts/WbarElP0b)
- **Node.js 26.9 turns on FFI by default at 37ns a call:** node:ffi now ships enabled (the old --experimental-ffi flag disables it instead), and while a trivial call costs 37-38ns versus 34-36ns for an N-API addon, a wrong buffer length passed to a pointer causes a segfault, so the Permission Model requires an explicit --allow-ffi flag. [Read more](https://daily.dev/posts/NN8Z3CglO)
- **Baseten's speculative decoding pipeline hits 3x on B200:** Dlash, a diffusion-based speculator succeeding Eagle 3, delivers over 3x improvement on a B200 running Qwen 38B, while Turboquant's polar-coordinate KV cache quantization proved impractical for data-center inference despite headlines, since it halves throughput. [Read more](https://daily.dev/posts/6p2TkrnsO)

## Similar posts on daily.dev

- [Saturation at GitHub: the saga continues](https://daily.dev/posts/saturation-at-github-the-saga-continues-2uymwgoad) · Surfing Complexity · 6 upvotes · 0 comments
- [Addressing GitHub’s recent availability issues](https://daily.dev/posts/addressing-github-s-recent-availability-issues-gzw2ygubb) · GitHub Blog · 3 upvotes · 0 comments

---

Tags: [#github](https://daily.dev/tags/github), [#aws-lambda](https://daily.dev/tags/aws-lambda), [#vercel](https://daily.dev/tags/vercel), [#ai-inference](https://daily.dev/tags/ai-inference)

[View this post on daily.dev](https://daily.dev/posts/github-s-postmortem-gets-dissected-vercel-tightens-the-free-tier-screws-g1beibdq5)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/github-s-postmortem-gets-dissected-vercel-tightens-the-free-tier-screws-g1beibdq5","headline":"GitHub's postmortem gets dissected, Vercel tightens the free-tier screws","text":"GitHub's September 13 outage postmortem gets a proper autopsy today, exposing how a replica-lag-only safeguard missed a saturating primary. Vercel is cutting Hobby-tier deployment retention hard, from 10 protected production deployments down to 3, with immediate deletion instead of a 30-day grace period. AWS Lambda quietly 6x'd its execution timeout for a new instance type, opening the door to workloads that used to need Fargate or Batch. Elsewhere, a batch of inference-engineering talks make the case that orchestration, not model quality, is where the real infrastructure work is happening now.","url":"https://daily.dev/posts/github-s-postmortem-gets-dissected-vercel-tightens-the-free-tier-screws-g1beibdq5","datePublished":"2026-09-20T04:17:37.470Z","dateModified":"2026-09-20T04:17:57.768Z","author":{"@type":"Organization","name":"Backend Digest","logo":"https://media.daily.dev/image/upload/s--9uCSrsAb--/f_auto,q_auto/v1773839398/logos/backend_digest?_a=BAMAMiiu0","url":"https://daily.dev/sources/backend_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/backend_digest","name":"Backend Digest"}}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Backend Digest","item":"https://daily.dev/sources/backend_digest"},{"@type":"ListItem","position":3,"name":"GitHub's postmortem gets dissected, Vercel tightens the free-tier screws"}]}
```

