Release Notes
Roadrunner SMB™ 1.3
Release date: August 24, 2026 Availability: AWS® Marketplace (same listing as 1.2)
Roadrunner SMB 1.3 hardens the platform for concentrated login storms while strengthening the administrative security model for both evaluation and production deployments. The release increases EFS parallelism, adds controlled admission at saturation, and addresses login-path and multi-channel failure modes exposed only at large scale.
Highlights
8 NFS mux channels per share. Each customer share uses 8 independent NFS I/O channels (efs-proxy sessions), optimized for login storms and high-throughput large-file writes. The supported setting is NfsChannels=8; do not reduce it.
Accept-path admission. When a node is at capacity, new connections receive SMB2 STATUS_SERVER_UNAVAILABLE after a short hold rather than a TCP reset, allowing Windows® clients to retry cleanly.
Characterized and scale-tested for FSLogix® login workloads. Real Windows Server® clients running FSLogix were used to characterize login-time SMB behavior. Those workload patterns were reproduced at scale with LoginSim and extended with CapacitySim to establish per-node capacity and resource-planning references.
Based on that characterization, a 4-node c6i.2xlarge Dual-AZ cluster supports planning for up to 20,000 users, while a 4-node c6i.4xlarge Dual-AZ cluster supports planning for up to 40,000 users. These results are intended to support capacity planning for WorkSpaces and other FSLogix-based environments.
Supported cluster size expanded to 1-8 nodes. The default remains 2 nodes. Clusters of 9-16 nodes remain Experimental (beta) and are not supported GA configurations.
Hardened Admin security posture. The Eval Kit now places the Admin UI and API behind CloudFront, ALB, and AWS WAF rather than exposing the Admin endpoint directly through a public NLB. Production Dual-AZ deployments keep the Admin UI and API private and require access through the customer's existing private-network controls, such as VPN, Direct Connect, a jump host, or equivalent mechanisms aligned with the customer's VPC and security policies. SMB service traffic remains private behind an NLB in both deployment tiers.
EFS defaults to Elastic throughput. EFS file systems created by the Eval Kit and Dual-AZ templates now default to Elastic throughput, allowing throughput to scale with workload rather than using the previous Bursting default. Bursting remains available as an override.
Stability improvements under login load
Concurrent directory creation is more reliable during login storms. First-time directory creation races on EFS that appeared under cold, highly concurrent login flights are addressed in 1.3.
ACL processing no longer depends on AWS calls in the login hot path. Under concentrated login load, 1.2 could generate additional AWS calls and latency during ACL generation. 1.3 removes those AWS calls from the critical login path.
File state remains consistent across mux channels. Rename, delete-pending, and post-write file-size reporting remain consistent across channels, preventing stale metadata from appearing during multi-channel operation.
SMB connection admission was hardened for saturation. A deeper accept queue, higher concurrency limits, and accept-path admission protect node stability under extreme connection demand. New connections receive STATUS_SERVER_UNAVAILABLE instead of failing with TCP resets when the node is saturated.
Administrative telemetry is more reliable under load. Session counts now use forked smbd children, and latency charts no longer freeze or disappear during sustained heavy load.
Updating from 1.2
A container rolling update applies 1.3 application changes. New stacks default to NfsChannels=8. The supported setting is NfsChannels=8; do not reduce it. Single-AZ is no longer a production Marketplace option; use Dual-AZ for production and the Eval Kit for evaluation.
Compatibility
- Cluster size 1-8 supported (default 2); 9-16 Experimental.
- Active Directory®, EFS, and Marketplace billing - unchanged from 1.2.
- 8 NFS mux channels per share.
nconnectand efs-proxyScaleUpConfigare not customer tunables.
Further reading
Roadrunner SMB 1.2
Release date: July 29, 2026
Availability: AWS Marketplace
Version 1.2 adds the Eval Kit - a self-contained, single-AZ evaluation deployment - first-class FSLogix profile-container support with a faster file-open path, multi-region deployments within a single AWS account, a redesigned Admin UI, and a roll-up of cold-start and Active Directory convergence fixes so nodes report ready only when they can actually serve traffic.
Highlights
Try Roadrunner SMB in 15 to 20 minutes with the Eval Kit. The new Eval Kit is a self-contained, single-AZ CloudFormation® deployment that stands up its own VPC and Active Directory, runs a headless First-Time Setup, and self-uninstalls on a timer after 30 days, removing the resources it created. Evaluators can reach a working SMB share without bringing their own Active Directory or designing a VPC first, making evaluations fast, easy and safe. See the Eval Kit Quick Start.
Built for FSLogix and VDI. Roadrunner SMB 1.2 supports FSLogix profile containers as a first-class workload, with the file-open path tuned specifically for the metadata-heavy pattern FSLogix logins generate - keeping the open path well ahead of login-storm demand and delivering responsive small-file performance for Windows Explorer.
What's in 1.2
- Eval Kit - self-contained single-AZ evaluation. A single-AZ CloudFormation deployment with its own VPC and Active Directory and headless First-Time Setup, typically ready in 15 to 20 minutes, that self-uninstalls automatically after 30 days. Evaluators can try Roadrunner SMB without bringing their own AD or VPC design. Teardown removes the compute, storage, load balancing, and directory services the kit created. See Errata below if you reinstall the Eval Kit in the same Region.
- First-class FSLogix support with an optimized open path. FSLogix profile containers are supported and tested for VDI and virtual-desktop deployments. The file-open path is tuned for the metadata-heavy behavior of profile logins: a measured FSLogix login issues only a modest number of opens across its login-and-settle window, and the optimized open path stays well ahead of that demand - so the open path is not the login-storm bottleneck. The same improvement delivers excellent small-file performance for Windows Explorer, with responsive profile-load and directory-listing behavior.
- Multi-region deployments per account (new deployments). CloudFormation templates now create IAM roles with a regional suffix (
rrsmb-<env>-<region>-*), so you can run one appliance per AWS Region in the same account. Existing 1.1 stacks that upgrade the container image only keep account-global role names until redeployed from a fresh 1.2 template. A manual stack update that adopts the new role names replaces IAM roles (brief disruption). - Redesigned Admin UI. A cluster-wide Dashboard overview as the landing page, a dedicated Metrics page for live throughput and latency charts, a Documentation sidebar link, and more reliable responsive navigation.
- Clearer live metrics under load. Cluster-aggregated pipeline throughput, DynamoDB® read/write and throttle visibility, cache hit rate, and percentile latency overlays, so operators see cluster-wide behavior instead of a single noisy node.
- More reliable cold-start and cluster convergence. Follower nodes restore Active Directory trust from shared state with real verification before serving, cluster readiness is gated so converging nodes aren't recycled prematurely, and boot-time EFS DNS and mount races are bounded rather than reported as "healthy" before a node can serve.
- More reliable domain administrator login. Domain admin authentication now relies solely on direct Active Directory authentication (Kerberos / machine-keytab LDAP), avoiding clustered winbind/idmap conditions that could surface as an incorrect password.
- More stable share materialization. Clearer in-flight versus failed states, safer reconciler behavior, and correct cache invalidation when share directories are created or renamed, so listings and permissions stay consistent across nodes.
- Fewer unnecessary service restarts. winbindd and smbd restart only when configuration or credentials actually change.
- Stress tested and performance tuned. Exercised under multi-client continuous stress and Windows login-scale stress testing, in addition to the existing smoke, smbtorture ACL, ACL-cache, and share-materialization gates.
Updating from 1.1
Container-only rolling updates (automatic or manual image change) apply 1.2 application fixes without changing IAM role names and without enabling multi-region in place.
To get multi-region IAM naming and the Eval Kit templates, deploy from the 1.2 CloudFormation launchers as a new stack, or plan a controlled stack update that adopts the new templates and role names. A stack update that adopts the new role names replaces IAM roles and causes a brief disruption.
Review release notes before production updates when your change-control process requires it.
Compatibility
- Cluster size 1-4 (default 2) - unchanged
- Active Directory, EFS, and Marketplace billing - unchanged
- New deployments: one appliance per AWS Region per account (regional IAM role names)
- Existing 1.1 stacks upgraded by image only: still one appliance per account until redeployed from 1.2 templates
- EFS-backed shares may differ from native NTFS® on some inheritance edge cases - see Known Limitations
- Host AMI may update via ASG independently of the container image
Errata
Eval Kit reinstall in the same Region. Uninstalling the Eval Kit (manual delete or the 30-day self-uninstall) is intended to remove the kit's CloudWatch log groups. In some cases two fixed-name groups can remain in the Region after the stack is gone:
/rrsmb/eval/<region>/cfn-lambda/rrsmb/eval/<region>/lambda
A later Eval Kit install in that same Region then fails at create time with a CloudFormation name-conflict validation error on those log groups. One Eval Kit per Region is supported; this only affects uninstall-then-reinstall in the same Region (not a first-time install).
Workaround: Before reinstalling the Eval Kit in a Region where a previous kit ran, delete any straggler groups, then launch again:
aws logs delete-log-group --region <region> \
--log-group-name "/rrsmb/eval/<region>/cfn-lambda"
aws logs delete-log-group --region <region> \
--log-group-name "/rrsmb/eval/<region>/lambda"
Ignore ResourceNotFoundException if a group is already gone. Then redeploy the Eval Kit as usual.
Admin UI Dashboard / Metrics charts may stop updating or show a partial series under load. During heavy data-plane traffic, Read/Write latency and related pipeline charts can briefly freeze, show "Waiting for series...", or jump to a shorter window. A page refresh restores the charts.
Workaround: Refresh the browser page if a chart stops updating or looks incomplete. A fix for the client-side poll/merge race is in progress for a subsequent update.