The Latest Features in CloudFS and the Work You No Longer Have to Do by Hand

CloudFS 8.7.1 Delivers Prewarm Jobs That Run Overnight, Snapshots on the Quarter Hour, Ring Health in a Single API Call, and Lock Aging That Cuts the Tail Off a Decommission

Key Takeaways:

• Prewarm jobs can now be scheduled to run overnight or over a weekend instead of watched, and the minimum snapshot interval drops from 60 minutes to 15 minutes, cutting the work exposed between recovery points by as much as 75% during peak periods.
• Ring-wide health is a single API call, Grafana Cloud Advanced connects through a secure tunnel with no inbound firewall rules, and multiple named audit consumers can now receive the full event stream concurrently, with alerts when delivery fails.
• Stale global read-write locks age out automatically after seven days, so decommissioning a node that holds millions of them fits inside a change window rather than running for weeks and unpairing an HA-Local configuration keeps SMB sessions connected, moving that job out of the maintenance window.

Running a file platform requires a steady stream of operational activities. Administrators need to make sure data is available where and when users need it, protect it on a regular schedule that meets the business objectives, monitor the health and capacity of the environment, and respond when something needs attention.

As the environment changes, they also need to add capacity, manage infrastructure changes, and eventually retire aging hardware. Each of these activities takes time and coordination, and some still require administrators to work late at night or over a weekend.

The features in Panzura CloudFS 8.7, released in March, handed over a set of those activities to customer IT teams as self-service controls including Prewarm Provision, File Lock Release, and Health Check Diagnostics. The features in CloudFS 8.7.1, released on August 19, take yet another step in the same direction and reduce the manual work those activities require. The prewarm jobs can be scheduled instead of watched. The health check became an API call. And when a node is no longer making changes, its locks can simply be returned to their owner, making node retirement much easier.

What follows starts with the changes that apply to any CloudFS deployment from day one, then moves to the ones that matter most to teams already running a ring in production.

Table 1. What changed from CloudFS 8.7 to CloudFS 8.7.1

Staging before anyone asks for it: Prewarm provisioning grows up

Prewarm Provision arrived in CloudFS 8.7 to stage data from the object store into local cache before users ask for it. In CloudFS 8.7.1, it now has the operational capability to make it usable at scale.

• Scheduling. Prewarm jobs, across directory paths and file lists, can now be scheduled in advance and execute automatically at the specified time, updating status as they go. Prewarming pulls a lot of data across the WAN and potentially competes with the people you are trying to help. Now, it can be run on off-hours or a weekend before a new office opens, without anyone staying up to start it. The endpoints are POST /pz/cfs/api/v1/prewarm/start/directorypaths, /prewarm/start/uploadfilelist, and /prewarm/generatefilelist.
• Search and filtering. Job history filters by status, node name, job ID, submitter, and CloudFS share, through the WebUI or /pz/cfs/api/v1/prewarm/jobdetails. On a deployment running dozens of jobs across many nodes, this is the difference between answering if a site completed a prewarm and simply scrolling until you find it.
• UNC and DFS paths in the file list. Uploaded CSVs now accept UNC and DFS paths alongside CloudFS-native ones, and the system converts them before prewarming starts. The file lists most teams already have exported from legacy shares and carry legacy paths. Hand-rewriting them was the step where jobs failed.
• File size in generated lists. Generate File List now returns file path and file size. Cache is finite, and a prewarm job larger than the cache it targets evicts the working set to make room. Having size in front of you before you submit turns that into a matter of arithmetic rather than discovery.
• Naming. Generated archives are now named for the job that produced them rather than the generic generated_file, and scheduled job messages carry the job ID. Small, and worth it the first time you have four batch files in a downloads folder.
• HPE VM Essentials, formerly Morpheus. CloudFS now supports it as a hypervisor. A large number of organizations are reassessing their virtualization platform on licensing grounds, and HPE’s KVM-based option, licensed per socket, is on most of those shortlists. Supporting it means the file platform is not a blocker for a hypervisor decision.
• Azure OS disk conversion. Standard SSD disks can now be converted to Premium SSDs, including the OS disk. This is important because Microsoft’s single-instance VM connectivity SLA requires premium storage across all operating system and data disks, so a Standard SSD OS disk left the VM outside the SLA regardless of what the data disks were. It is now possible to convert in place without rebuilding the node.
• AWS instance conversion. M4.4xlarge instances can be converted to M7i.4xlarge. M4 is a long-serving family while M7i runs on 4th Generation Intel Xeon Scalable processors with substantially better price performance, network bandwidth, and EBS throughput. An in-place conversion path means getting the benefit of the improvement without rebuilding the node and re-earning its cache.

Protecting work in progress: Snapshots on the quarter hour

The minimum scheduled snapshot interval drops from one hour to 15 minutes, giving administrators more control over how frequently active work is protected. An environment can maintain hourly snapshots throughout the day, for example, while increasing protection to 15-minute intervals during known peak periods when users are creating and changing the most data.

That shorter interval reduces the gap between the last good snapshot and an unwanted event, such as a bad overwrite, a script that traversed the wrong directory, or an encryption event caught in progress. During those high-activity periods, moving from 60- to 15-minute snapshots can reduce the amount of work at risk by as much as 75%, while allowing administrators to balance more frequent protection against the additional snapshot overhead.

Health data where your team already looks

• Grafana Cloud. CloudFS now supports Grafana Cloud Advanced, letting customers connect their Grafana Cloud instances to on-premises InfluxDB data hosted on CloudFS nodes. Most enterprises already run Grafana, and the point of the integration is that CloudFS health stops sitting in a separate console that someone needs to remember to open.
  
  The connection uses Grafana Private Data Source Connect, which establishes a secure tunnel from an agent inside your network without opening inbound firewall ports. Security teams tend to have a firm view about inbound rules for a monitoring integration, and this avoids the conversation. New Executive Overview and Traffic Light summary widgets, with drill-down, give real-time health across large node deployments, and the PDC agent can be started, stopped, and monitored from the CloudFS WebUI or the REST API rather than by someone with shell access to the host running it.

• Ring Health Check API. A single call now returns a consolidated health report for every node in the ring: GET /pz/cfs/api/v1/ring/healthcheck, with GET /pz/cfs/api/v1/node/healthcheck for one node. CloudFS 8.7 made health checks self-service in the interface, and version 8.7.1 makes them programmatic, which is what NOC dashboards, scheduled automation, and third-party ITSM and observability platforms need. Checking ring health goes from a task someone performs to a matter of monitoring.

Platform support now includes the following:

• HPE VM Essentials, formerly Morpheus. CloudFS now supports it as a hypervisor. A large number of organizations are reassessing their virtualization platform on licensing grounds, and HPE’s KVM-based option, licensed per socket, is on most of those shortlists. Supporting it means the file platform is not a blocker for a hypervisor decision.
• Azure OS disk conversion. Standard SSD disks can now be converted to Premium SSDs, including the OS disk. This is important because Microsoft’s single-instance VM connectivity SLA requires premium storage across all operating system and data disks, so a Standard SSD OS disk left the VM outside the SLA regardless of what the data disks were. It is now possible to convert in place without rebuilding the node.
• AWS instance conversion. M4.4xlarge instances can be converted to M7i.4xlarge. M4 is a long-serving family while M7i runs on 4th Generation Intel Xeon Scalable processors with substantially better price performance, network bandwidth, and EBS throughput. An in-place conversion path means getting the benefit of the improvement without rebuilding the node and re-earning its cache.

One event stream, every consumer that needs it

Until now, only one RabbitMQ consumer could be registered for third-party audit events at a time. That was a real constraint, because file audit events are the raw material for two different jobs. Data security platforms consume them to spot the behavioral signatures of ransomware, insider activity, and compromised credentials. Panzura Nexus consumes them as part of making file data available and governable for AI systems. One registration slot meant choosing.

CloudFS 8.7.1 supports multiple named consumers running concurrently over independent RabbitMQ connections, each receiving the full, unfiltered stream. They are isolated from one another. A failure in one vendor’s pipeline does not affect delivery to the rest, and registering or deregistering one does not disturb another. Add them under Configuration > Monitoring > Audit Settings.

Delivery failures now surface as alerts, which addresses another risk. An audit stream that stops is not obvious from the outside because the tooling keeps running, the dashboards keep rendering, and the gap gets discovered during an investigation or an audit, when it is crucial and there is little time to fix it.

In fact, detection windows are already long enough without a blind spot inside them. IBM’s 2026 Cost of a Data Breach Report puts the mean time to identify and contain a breach at 247 days, reversing several years of improvement, and incidents running past the 200-day mark cost an average of $5.65 million against $4.32 million for those resolved faster.

A security platform that silently stopped receiving file events is how a lifecycle reaches 247 days. SYS.AUDIT.TP1.1 fires after 10 consecutive delivery failures to a configured vendor, naming the vendor and the failure count, with a five-minute minimum between alerts. SYS.AUDIT.TP2.1 fires when a vendor is enabled for third-party audit but RabbitMQ is not configured, and reports how many events were dropped.

For teams already running CloudFS in production

The changes that follow apply once a deployment has been in service long enough to accumulate history: nodes to retire, locks that have piled up behind them, and HA pairs to reconfigure. A new deployment will meet them eventually. An existing one can use them now.

Decommissioning loses its multi-week tail

A CloudFS node that has been in production for a few years accumulates global read-write locks. On a large deployment, that could be millions of them. A large number of them reside on files nobody has opened since the project closed out, and the lock has no way of knowing that.

It matters at the exact moment the node has to go. Decommissioning, deactivating, or recovering a node means migrating every lock back to the lessor, and that migration represents a slow operation. That means several weeks in the worst cases, with an administrator checking progress throughout.

CloudFS 8.7.1 ages them out instead. When a file has gone seven days without being accessed or modified, its lock returns to the lessor automatically. The threshold is configurable, and the feature is turned on by default. Panzura Support can enable or disable it on request. Files under active collaboration are unaffected, because seven days of silence is the bar.

The payoff affects the project schedule. Hardware refreshes, office consolidations, lease expirations, and cloud migrations all queue behind a node finishing its exit. When the locks have already drained in the ordinary course of business, retirement becomes a maintenance task with an end time you can put in a change ticket.

Nodes that leave cleanly and stay gone

Three related changes make node removal more predictable.

Managed decommissioning and deactivation now perform a graceful shutdown, which gives connected clients the chance to flush their data first. That allows for a planned exit rather than an abrupt one, and the difference is important because it is the one that users notice.

A force-decommissioned node now boots into maintenance mode and cannot rejoin the ring. A retired node that reappears on the network, for instance after a reboot, when someone racks it somewhere else, or after a VM gets restored from a snapshot by mistake, is a challenge whoever has to work out why the ring is behaving strangely. That path is closed.

And a deactivated node can now be decommissioned directly. Previously, retiring one meant bringing it back up first, assuming that is still possible. The hardware may be dead, or the site could even be closed. The lease on the rack may have expired. Removing the reactivation step removes a dependency on conditions you no longer control.

An HA change you can make in an afternoon

Unpairing an HA-Local configuration used to disrupt SMB client connectivity. That meant it was an after-hours job because it disconnects users and requires them to rejoin Active Directory.

There is a reason those changes get pushed to midnight. ITIC’s 2024 downtime survey, which polled more than 1,000 firms worldwide, puts the cost of a single hour above $300,000 for more than 90% of mid-size and large enterprises, before any litigation or penalties are counted. Planned disruption is cheaper than unplanned, but it is charged at the same rate while it lasts, and it is paid twice. Once by the users who stop working and again by the administrator doing the change, which often happens at a time when nobody is available to help if it goes sideways.

In CloudFS 8.7.1, the shared virtual IP stays active on the primary node through the unpair. Clients keep their sessions. Nobody reconnects or rejoins the domain, and the change moves out of the maintenance window and into the working day. In the confirmation dialog, leave the Remove VIP toggle off, which is the default configuration, and the VIP is preserved. When you later pair a new LHA node, the previously configured VIP settings are detected and pre-populated, which takes a manual re-entry step and the typos that come with it out of the picture.

Manual failovers can also be aborted while in progress. A takeover started by mistake, or one overtaken by events, no longer must run to completion before you can put things back.

Where this leaves the conventional approach

Most of these operations are done on traditional file infrastructure according to a default approach. That is, a file server estate, a NAS with replication, or a distributed file system without these controls. The comparison is less about features than about who does the work.

Table 2. What changes with CloudFS 8.7.1

Twenty-eight changes customers asked for

The other important matter is the list of things customers told us they needed, and CloudFS 8.7.1 delivers 28 of them. They arrived through support cases, field escalations, and conversations with administrators who know firsthand what they need to move smarter and faster

Several are important to name due to the workflows they touch. Civil 3D and Revit add-in workflows over SMB now handle file renumbering the way the applications expect, and Samba sessions will not stall during the metadata operations that frequently involve AutoCAD files, which is the category of behavior that costs a design team time and usually gets blamed on the network.

Files return to their correct size when a deactivated node comes back online. Requests around NTACL tooling, NFS share creation, cross-node change notification on Windows Explorer, and email alert routing all represent requests that came in from the field and are answered in this release.

Ten CVEs are addressed as part of the same cycle, across high-, medium-, and low-severity, alongside two changes customers raised directly, namely Swagger UI now requires authentication before serving API documentation, and the NGINX rewrite module is updated. Security maintenance runs continuously rather than waiting for a major release.

Every item, with its issue ID and a description of what changed, is listed in the release notes. Two changes belong in front of your administrators before the change window.

• Username-based authentication now requires the domain. Scripts using username-based authentication and the net use command must supply credentials as DOMAIN\username. The automatic mapping from a bare username no longer happens, so any tooling passing a free-form username will fail to authenticate after the upgrade. This applies to automated scripts, the IO test framework, and manual share mapping. Worth grepping your automation for before you schedule.
• Linux CIFS mounts may need remounting. The Samba 4.22 upgrade changes how identifiers are generated, which can conflict with information cached on Linux clients holding existing CIFS mounts and surface as stale file handle errors. Unmount the affected share and remount it with the noserverino option, which has the client generate its own inode numbers rather than using the server’s.

Full release notes, including every new or improved item and the complete security update list, are in the Panzura Knowledge Base. Talk to Panzura Support to schedule your upgrade.

Frequently Asked Questions

• What is Panzura Express and what does the license include?

  Panzura Express is a full enterprise-class NAS delivered as software, built for single-site organizations. It packages the Panzura CloudFS file platform with a local high-availability node pair, Panzura Data Services for file estate search, activity auditing, and AI-powered Threat Control, and Panzura Nexus, which prepares file data for Microsoft 365 Copilot. Everything comes in one annual license, available in 25TB, 50TB, and 100TB capacity tiers and sold through Panzura channel partners. Because it deploys as software, there is no hardware to order and nothing waiting in a fulfillment queue.

• How does software-defined file storage avoid the 2026 hardware shortage?

  Hardware-based storage puts a physical purchase in the critical path, and today that path is congested: major hard-drive makers committed their production for the year in advance, and major-vendor server lead times have stretched to 32 to 40-plus weeks. Software-defined file storage removes the purchase entirely. Panzura Express deploys as software on infrastructure an organization already owns, or against cloud object storage, so capacity decisions become licensing decisions. There is no allocation queue to join, no quote that expires in a week, and no repricing clause between order and delivery.

• What is the difference between Panzura Express and buying a new NAS appliance?

  A NAS appliance bundles storage software with proprietary hardware, so every refresh, expansion, and failure ties you to a physical supply chain, one that currently carries sharply higher pricing and long lead times. Panzura Express delivers the same class of capability as software: enterprise file services, high availability, immutable snapshots, and ransomware protection, running on hardware you choose or in the cloud. Growth becomes a license change rather than a forklift upgrade. The appliance model asks you to buy the box again every few years. Panzura Express separates the file platform from the metal permanently.

• Can Panzura Express run on hardware we already own?

  Yes. Panzura Express is delivered as software and deploys on virtual infrastructure an organization already operates, on-premises or in the cloud, using S3-compatible object storage as its foundation. That is the point of the design: the file platform is independent of any particular box, so nothing in the deployment depends on ordering new equipment. For firms holding a storage refresh quote that has doubled, or waiting on extended server lead times, this changes the decision from a capital purchase with a delivery date to a software rollout that can begin immediately.

• How does Panzura Express protect against ransomware compared with traditional backup?

  Traditional backup protects data after the fact: it restores files once an attack has already succeeded, often losing hours or days of work in the gap between snapshots. Panzura Express builds protection into the file system itself. Immutable snapshots mean encrypted or deleted data can be rolled back to a clean state, while Threat Control uses AI-driven user behavior intelligence to detect ransomware activity as it happens and contain it. The result is dramatically faster recovery, without maintaining a separate backup product, license, and administration burden alongside primary storage.

• What size organization is Panzura Express designed for?

  Panzura Express is built for single-site organizations, typically small and mid-sized firms managing up to 100TB of file data — the segment that historically had to choose between entry-level NAS boxes and enterprise platforms priced beyond reach. Three capacity tiers of 25TB, 50TB, and 100TB let a firm match the license to its current footprint and move up a tier as data grows, without a migration project. Multi-site organizations with distributed teams are better served by the full Panzura CloudFS platform, which extends the same foundation across many locations.

• How does Panzura Express make file data ready for Microsoft 365 Copilot?

  AI assistants are only as good as the data they can reach, and file servers keep most organizational knowledge locked away from them. Panzura Express includes Panzura Nexus, which connects file data directly to Microsoft 365 Copilot while preserving existing permissions, so people and AI tools work from one authoritative, governed source of truth. Copilot can then draw on project files, documents, and institutional knowledge it otherwise could not see with access controls intact, so no one retrieves anything through AI that they could not open themselves.