Gitaly - FY26Q4 Kickoff and planning

Summary

This issue will be used to plan projects and objectives for the quarter

Tasks

1. Title this issue <team name> - FYXXQX kickoff and planning
2. Fill in the section below for Project Epics, create new epics if required using the New Project template.
   • Assign the epic to the DRI
   • In the majority of cases, epics should be scoped to take a single quarter or less
   • Set a rough estimate for start and end date
   • If there is a corresponding Data Access Roadmap item for the project epic, link it in the spreadsheet
   • Set the label ~workflow-infra::Ready or workflow::ready for development
3. Close epics for the previous quarter
   • Ensure that all epics are closed with a final closing status in the description
   • Move over any work from the previous quarter's "Keep the Lights On" epic into the new epic

Project Work

New Project Epics

Topic	Start Date	Target End Date	DRI	Participants
Gitaly on k8s for SM	2025-11-03	2026-01-31	@oli.campeau	@echui-gitlab
RAFT Admin CLI	2025-11-03	2026-01-31	@jamesliu-gitlab	@mustafabayar

Project Epics Carried Over

These epics are currently in progress

Topic	Start Date	Target End Date	DRI	Participants
Raft: Support adding read-only replicas to a partition @qmnguyen0711 (+0 participants)	2025-04-12	2025-12-12		@divya_gitlab, @sdhanak, @jamesliu-gitlab , @mustafabayar , @echui-gitlab
Gitaly Performance Benchmarking @echui-gitlab @tkuah (+10 participants) groupgitaly	2025-09-13	2025-12-12	@jamesliu-gitlab
Raft: Routing Logic (+0 participants) groupgitaly	2025-10-11	2025-11-14	@qmnguyen0711	@oli.campeau, @sdhanak

Important Milestones

End of FY26Q4 - Gitaly Transaction Go/No-Go

By end of FY26Q4, we will know what the limits of Gitaly Transactions in its current implementation are. With OverlayFS as an optimization, we will be able to determine the minimum latency overhead that Gitaly Transactions will add, as well as the workload limits before we start to see pathological performance. This information will help us decide whether or not we can continue with the current implementation, or if we need to re-design the system.

Mid FY26Q4 - Gitaly RAFT Demo

By Mid FY26Q4, we will have enough logic in RAFT to do an end to end demo with RAFT replication and read request routing to show:

1. Adding a replica node to a RAFT cluster
2. Data gets replicated from primary to replicas
3. Read requests get routed to up to date replicas

We can even benchmark Gitaly RAFT with the benchmarking tool so we can see the impact of both slower write latency due to replication, but better read throughput due to read requests spread across the cluster.