GitHub says the pressure on its Git infrastructure is coming from sustained, concurrent activity: developers and coding agents making frequent commits, branches converging through merges, and CI systems repeatedly reading the same pushed code. Its announced redesign separates durable repository storage from the compute that serves Git requests, while narrowing the points where pushes must wait for agreement. GitHub reports up to 35 times higher write throughput in internal benchmarks, but has not published the methodology or detailed test conditions for that figure.
Why is GitHub rebuilding its Git infrastructure?
GitHub’s explanation is about the rate and concurrency of repository work, not simply the number of repositories. An agent can create frequent commits or checkpoints; parallel branches eventually converge on shared references through merges; and a single push can trigger many reads from CI, code scanning, the web interface, or API clients. Those overlapping operations put pressure on both writes and reads.
In an engineering post published October 6, 2026, and updated October 7, Brian Celenza, a principal software engineer working on GitHub storage and core services, described the goal as “rebuilding GitHub’s Git infrastructure while GitHub keeps running, creating a foundation for agent-scale software development.” GitHub’s published activity figures illustrate the scale it says it is addressing:
- Monthly Git events rose from 218.2 billion in September 2025 to 473.3 billion in August 2026.
- GitHub counted 7.38 billion commits in September 2026, more than five times its count a year earlier. Monthly pushes rose from 0.69 billion to 3.35 billion, which GitHub described as 4.9 times year over year.
- GitHub Actions ran 3.26 billion times in September 2026, more than four times the year-earlier volume. Pull request merges approached four times their year-earlier volume, though GitHub did not provide a precise merge count.
- The busiest repository received roughly one billion requests in August 2026.
These are figures reported by GitHub, not independently audited measures in the cited post. The post does not break down the activity into agent, human, and CI shares, so the totals show growth in GitHub’s overall workload rather than a measured contribution from coding agents alone.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute#1 Best Overall
How does GitHub’s current repository storage work?
GitHub says its Spokes system keeps a full repository copy on local disks across several fileservers—five by default. Those copies provide redundancy and distribute read traffic, while fast local disks serve Git operations. When a push updates a reference, a three-phase commit protocol uses a quorum so the web interface, CI, and API clients see a consistent repository state.
This design ties read capacity to write overhead. Adding a replica can help serve reads, but that replica also participates in writes; the push is constrained by the slowest replica in its set. If a replica is lost, available read capacity falls, and if the system loses quorum, writes stop. Faster clones can help with some read demand, but they do not solve the need to durably store writes and make them consistently visible to subsequent agents and CI jobs.
How is GitHub changing repository storage?
GitHub’s announced direction changes where durable data lives and which operations must coordinate before a push can complete. The post describes three related changes:
Coordinate only where Git semantics require agreement
GitHub aims to preserve agreement for the reference update while doing more object storage, connectivity validation, and secret scanning in parallel. The intended effect is a shorter push critical path: less work has to wait in sequence before the updated reference becomes visible.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Move compaction and garbage collection away from live requests
Separate workers will handle compaction and garbage collection against durable storage. In the design GitHub describes, those maintenance jobs no longer compete with live Git requests on the same serving hosts.
Separate durable storage from request-serving compute
GitHub names Azure Blob Storage as the authoritative durable layer and describes lightweight compute workers that cache repository data and serve requests. Because read scaling no longer requires adding another durable full copy that participates in every write, GitHub says read and write capacity can scale independently. Workers can be added for bursts in demand, and after a worker fails, a replacement can serve traffic while its cache fills.
What changes between the old and announced designs?
| Design question | Current system, as GitHub describes it | Announced direction |
|---|---|---|
| Where is repository data stored? | Full copies on local disks across several fileservers, five by default. | Azure Blob Storage is the authoritative durable layer; compute workers cache data and serve requests. |
| Does adding read capacity add write overhead? | Yes. A new replica participates in writes, and the slowest replica in the set can limit a push. | Read capacity can be added without adding another durable copy that participates in every write. |
| Where is coordination required for a push? | A three-phase commit protocol uses a quorum when a push updates a reference. | Agreement remains for the reference update, while more object storage, connectivity validation, and secret scanning work is performed in parallel. |
| Do maintenance jobs share the serving path? | The post describes fast local disks serving Git operations but does not separately specify the current placement of compaction and garbage collection. | Separate workers handle compaction and garbage collection against durable storage, away from live request-serving hosts. |
| How is a failed compute host recovered? | The post does not describe a replacement-worker recovery path for the current design. | A replacement worker can serve traffic while its cache fills after a worker failure. |
What does the “up to 35 times” throughput claim mean?
GitHub says the new architecture delivered up to 35 times higher write throughput in internal benchmarks. That is a company-reported maximum, not a general guarantee that every repository or push will be 35 times faster. The October 2026 post gives no benchmark methodology, baseline, workload mix, or comparison conditions, and it does not provide independent validation. Treat the figure as an early indicator of GitHub’s reported infrastructure results, not a user-level performance promise.
Will the infrastructure changes affect how developers use GitHub?
GitHub says the work is being done while the service remains online and is intended not to require customers to change how they build software. The company says it plans to preserve familiar workflows and governance controls, including branching, review, merges, repository history, branch protections, required reviews, audit logs, repository visibility, automation, and observability. The announcement describes an ongoing rebuild, not a completed migration, and gives no rollout completion date.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Quick Recap
Best Value
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.
