What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
GitHub is splitting the place where repositories are durably stored from the workers that answer Git requests. In the design its engineering team describes in an October 2026 post, authoritative repository data would live in Azure Blob Storage, while lightweight compute workers cache that data to serve reads and handle pushes. The stated goal is to keep large repositories responsive as CI systems and coding agents generate far more concurrent reads and writes than the current Spokes architecture was built for. The post also says the aim is improved reliability, scaling and recovery. It does not describe a reliability incident or claim that reliability was lost.
What Spokes does today
According to GitHub’s announcement, the existing Spokes system keeps a full copy of each repository on the local disks of several fileservers, five by default. Fast local disks keep Git operations low-latency, and the extra copies provide redundancy and spread read traffic across machines.
As an Amazon Associate I earn from qualifying purchases.
Reference updates, the moves that change which commit a branch or tag points to, go through a three-phase commit protocol that uses a quorum. That agreement is what lets CI, the web interface and API clients all see a consistent repository state.
The trade-off sits in the same replicas doing two jobs. They are the durable copies and they are the read capacity. GitHub says every replica takes part in every write, so a push finishes only as fast as the slowest replica in its set. Adding replicas to absorb more reads therefore adds write overhead. At the highest activity levels, losing quorum stops writes altogether.
The design GitHub describes
The redesign changes the coupling. Durable storage and request serving become separate layers, and several of the heaviest jobs move away from the machines answering live traffic.
Authoritative storage in Azure Blob Storage
In the proposed design, Azure Blob Storage holds authoritative repository data. Compute workers do not need to be full durable copies. Read capacity can grow by adding workers that cache what they need, without adding another durable replica to every push. This is the change that separates the cost of durability from the cost of serving reads.
Lightweight workers that repopulate from durable storage
Workers can be added for activity bursts and removed afterward. If one fails, its replacement can begin serving requests and fill its cache from durable storage, rather than first rebuilding a full repository copy. GitHub presents this as the recovery advantage of the model.
Recommended Free Tools
A shorter coordinated part of each push
GitHub’s central point is that only part of a push needs agreement. In the announcement, Brian Celenza, a principal software engineer working on GitHub storage and core services, writes: “The part of a push that truly needs agreement is the reference update itself.” Object storage, object-connectivity validation and secret scanning can mostly run in parallel with other writes. The stated aim is to keep the correctness coordination Git requires while shortening the critical path of the push.
Rank #3
Compaction and garbage collection off the live path
Under the current arrangement, heavy maintenance runs on the same hosts that serve live Git requests. The proposal assigns compaction and garbage collection to separate workers that operate against durable storage. Live traffic then competes less with housekeeping on the same machines.
Old and proposed designs compared
| Question | Spokes (current, per GitHub) | Proposed design (per GitHub) |
|---|---|---|
| Where authoritative data lives | Full copies on local disks of several fileservers (five by default) | Azure Blob Storage |
| Do extra reads need extra durable copies? | Yes; replicas double as read capacity | No; compute workers cache data and serve reads |
| How writes coordinate | Every replica joins every write; three-phase commit with quorum | Reference update still needs agreement; object work and scanning largely run in parallel |
| Behavior when a serving node fails | Not described in the announcement beyond the quorum trade-off | Replacement serves requests and repopulates its cache from durable storage |
| Where compaction and garbage collection run | On the hosts serving live Git requests | On separate workers against durable storage |
| How capacity responds to bursts | Adding replicas also adds write overhead | Workers added for bursts and removed afterward |
Why GitHub says the change is needed now
The announcement ties the redesign to agentic development. Agents may commit or checkpoint after many individual actions, which produces high write concurrency alongside heavy CI and code-scanning read fan-out. The figures GitHub gives are below, with the units and periods it uses.
Rank #4
| Measure | Figure in GitHub’s announcement |
|---|---|
| Pushes per month | From 0.69 billion to 3.35 billion, a 4.9× increase year over year |
| Pull request merges | Nearly four times the volume of a year earlier |
| GitHub Actions runs | 3.26 billion in September, more than 4× the year-earlier level |
| Commits | 7.38 billion in September, more than 5× the level a year earlier |
| Total Git activity per month | 218.2 billion events in September 2025; 473.3 billion in August 2026 |
| Busiest repository | Roughly one billion requests in August 2026 |
| Write throughput | Up to 35× higher in GitHub’s internal benchmarks |
The push figure is a monthly count. It should not be read as an annual or daily rate. The September figures for Actions runs and commits are stated as September without a year in the relevant sentence, so readers should take them as GitHub’s own reporting period in the announcement.
What stays the same for developers
- Branching, review, merge and history behave as they do now, according to GitHub.
- Branch protections, required reviews, audit logs and repository visibility are intended to be preserved as the infrastructure changes.
- GitHub says the service keeps operating during the rebuild, with no maintenance window that stops code movement and no required changes to customer development workflows.
What is established and what is not
The announcement describes an architecture and an active rebuild. It does not state a completion date, a customer rollout schedule or region-by-region availability, so this is not a finished migration that readers can check against a date.
Best Value
The 35× figure is GitHub’s internal benchmark result. The announcement does not describe the workload or the methodology, and it offers no independent or production-wide verification. “Up to” is the operative phrase. It should not be read as a typical or guaranteed gain for any customer’s workload.
For the full technical description, GitHub’s original post is at Building Git infrastructure for agent-scale development, published October 6, 2026 and updated October 7, 2026.
Quick Recap
“
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
