On a FlashArray//XR2 or //XR3 , if a dedicated PCIe SAS HBA (Host Bus Adapter) is required for external shelf connectivity (expanding beyond the onboard SAS ports), it must be installed in Slot 1 .

The PCIe slot prioritization on the FlashArray//X R2/R3 controllers is strictly defined to ensure optimal performance and lane distribution:

Slot 0: Reserved primarily for frontend Host I/O cards (Fibre Channel or iSCSI/Ethernet). This slot usually connects to CPU 0 and offers high bandwidth for host traffic.

Slot 1: Designated as the primary expansion slot for backend connectivity (SAS cards for expansion shelves) or secondary Host I/O if no SAS expansion is needed.

Slot 2: On smaller models (X10/X20), this is often the only host I/O slot, but on the larger X50/X70/X90 chassis, it is used for additional Host I/O.

If an Implementation Engineer installs a SAS card in Slot 0, the system may fail to recognize the shelves correctly or may flag a configuration error, as the Purity software expects backend expansion traffic to route through the bus associated with Slot 1. Therefore, for SAS expansion cards, Slot 1 is the mandatory location.

To reduce SafeMode protections (such as shortening the retention period or disabling the eradication timer), two authorized SafeMode approvers must authenticate and approve the request via Pure1 step-up authentication .

SafeMode is a ransomware protection feature designed to prevent the accidental or malicious deletion of snapshots. Because reducing these protections weakens the array's security posture, Pure Storage enforces a strict "Ratchet" authorization process.

The Process: Unlike standard support requests, a single admin or local user cannot authorize this change (making Option C incorrect). The customer must have previously designated specific individuals as "SafeMode Approvers" in their Pure1 portal.

Authorization: When a request to weaken the policy is made, Pure Support triggers a verification workflow. Two of these designated approvers must log into Pure1 and perform a secondary authentication (often involving a PIN or TOTP) to explicitly "sign off" on the reduction. This "two-person rule" ensures that a compromised credential or a rogue insider cannot unilaterally expose the organization's backup data to destruction.

The status "unrecognized" on a drive module typically indicates that the physical hardware is detected by the chassis midplane, but the operating system (Purity) does not have the necessary driver definition or firmware logic to interact with it.

This most commonly occurs when newer generation DirectFlash Modules (e.g., larger capacity drives or newer NAND revisions) are inserted into an array running an older version of Purity. The hardware identifier of the new drive is unknown to the older software .

To resolve this, the Implementation Engineer must verify the Purity Compatibility Matrix for the specific drive part number. The array will likely need a Purity software upgrade to a version that supports the new media before the drives can be successfully admitted and used. It is rarely a hardware fault (which would show as "failed" or "missing") or a capacity limit (which would prevent admission but usually identify the drive correctly).

=========

In a Non-Disruptive Upgrade (NDU), ensuring that the host is correctly multipathed and balancing I/O across all paths is critical before taking a controller down. If a host is only using paths to the controller about to be rebooted (CT0, for example), taking that controller offline would cause an outage.

The command iobalance --sampletime 30 (or its wrapper purehost monitor --balance) is the standard tool used during the upgrade procedure to validate this state.

iobalance : This script samples the I/O throughput on all ports for a specified duration (e.g., 30 seconds).

It analyzes the distribution of I/O. If it detects that I/O is heavily skewed or missing entirely on the "secondary" paths (the ones that will survive the reboot), it flags a warning.

This allows the Implementation Engineer to pause and remediate the host multipathing configuration (e.g., fixing a bad cable or zoning issue) before proceeding with the controller removal, ensuring the NDU remains truly non-disruptive.

The FlashArray//XR4 architecture fundamentally changed how external capacity is attached compared to older SAS-based or early NVMe platforms. Because DirectFlash Shelves (DFS) utilize massive 100GbE RDMA over Converged Ethernet (RoCE) for their backend loops to maintain NVMe-level performance, specific port and slot assignments are heavily regulated.

On an //XR4 model (such as the //X70 R4 or //X90 R4), the controllers feature dedicated, integrated 100GbE ports on the motherboard (specifically eth18 and eth19) which are strictly reserved for the first DirectFlash Shelf loop.

However, when an Implementation Engineer needs to install a second DirectFlash Shelf, the integrated onboard ports are already fully saturated. Pure Storage does not support daisy-chaining NVMe-oF RoCE shelves in the same manner as legacy SAS enclosures. Instead, a new dedicated backend fabric loop must be established. To accomplish this, the engineer must install an additional 100GbE RoCE PCIe expansion card. According to the official //XR4 PCIe Slot Population Matrix, the designated location for this secondary backend expansion card is PCIe Slot 4 . Installing the card into other frontend-designated slots (like Slot 0 or Slot 3) will trigger a topology mismatch during the array's pre-flight hardware checks and prevent the new shelf from initializing.

On the FlashArray//XL , PCIe Slot 5 is explicitly reserved for Non-Disruptive Upgrade (NDU) procedures and specific control plane functions.

The FlashArray//XL chassis (5U) features a redesigned controller layout with 9 PCIe slots per controller (numbered 0-8). While slots 0-3 and 6-8 are generally available for Host I/O (FC/iSCSI) or Backend Expansion (SAS/NVMe-oF), Slot 5 is kept open or populated with specific interconnect cards used during controller upgrades.

During a "Data-in-Place" upgrade (e.g., moving from an //X to an //XL or upgrading //XL generations), this slot facilitates the high-speed link required to cluster the old and new controllers or to transfer state information. Implementation Engineers must ensure this slot is not populated with standard customer I/O cards (like FC HBAs) during the initial configuration, as doing so would block the ability to perform future non-disruptive hardware upgrades.

DirectFlash Shelves (DFS) are populated with DirectFlash Modules (DFMs) or DFMDs (DirectFlash Module with Distributed NVRAM, though usually just DFMs in shelves). For the Gen 2 DirectFlash Shelf , Pure Storage offers specific "starter packs" or partially populated configurations to allow for capacity flexibility while maintaining necessary performance and electrical balance.

The validated starter configurations for a Gen 2 DFS are 12-slot and 14-slot populations. This means the shelf must be populated with at least this number of drives to be a supported configuration. Populating fewer than this (e.g., 8 or 10) might result in insufficient bandwidth utilization across the shelf's internal fabric or fail strict configuration checks in the Purity operating system. Engineers must verify the drive count matches these supported increments during inventory to avoid configuration errors during initialization.

=========

To correctly configure the proxy settings for Phonehome and Remote Assist on a FlashArray, the Implementation Engineer requires the Protocol prefix (HTTP/HTTPS) and the Port number in addition to the IP address.

The Purity operating system requires a complete URL structure to establish a connection through a proxy server. A raw IP address like "10.24.18.120" is insufficient because the array does not know which communication protocol to use or which listening port the proxy server is monitoring.

Correct Format: The required input format is typically scheme://host:port. For example, https://10.24.18.120:8080 or http://10.24.18.120:3128.

Why it's critical: Without the port, the connection will likely time out or be rejected. Without the protocol (http vs. https), the handshake may fail. While credentials (username/password) might be required if the proxy enforces authentication, they are optional depending on the customer's network policy. However, the port and protocol are always technically mandatory to form a valid network socket connection.

Adding an external DirectFlash Shelf (DFS) or a legacy SAS Expansion Shelf to a Pure Storage FlashArray requires precise physical cabling to ensure backend high availability. Because the controllers utilize an active/active backend topology—where both controllers must maintain redundant loops to every single capacity shelf—a single misplaced cable can cause a severe hardware fault.

Once the physical installation is complete, the Implementation Engineer must definitively validate the topology before handing the array over to production. To do this, they log into the local KVM console or SSH session using the restricted puresetup account. From here, the official Pure Storage post-installation procedure dictates running two specific Python validation scripts: storage_view.py enclosures and storage_view.py config .

While both are used, storage_view.py config is the primary command utilized to map the specific SAS or NVMe-oF (RoCE) port bindings. The script actively polls the physical ports on the rear of the controllers and traces the loops down through the I/O Modules (IOMs) on the shelves. If the cabling is flawless, the CLI will explicitly return: "No errors detected." If a cable is crossed (e.g., Controller 0 plugged into IOM 1 instead of IOM 0), the script will instantly flag the topology mismatch, allowing the engineer to correct it on the spot.

In a Non-Disruptive Upgrade (NDU) scenario, the goal is to replace the compute and control logic of the array while preserving the existing data and minimizing hardware waste. When upgrading from a FlashArray//X70R2 to an //X70R4, the primary components that change are the Controllers themselves. The chassis, power supplies, and NVRAM modules (depending on the specific generation jump and health) can often be reused or are not the primary target of the replacement in a standard NDU kit.

If a complete array is shipped by mistake (often called a "greenfield" or "net-new" array shipment) instead of the specific upgrade kit, the engineer should only extract and use the Controllers from the new shipment. The existing chassis and its passive components (like the midplane and power supplies) are already racked, cabled, and functioning. Swapping the chassis or power supplies would turn the operation into a disruptive migration or a much more complex "chassis replacement" procedure, which is unnecessary for a controller upgrade.

Therefore, the engineer should unpack the new controllers, perform the NDU procedure to replace the old controllers one by one, and leave the rest of the new hardware (chassis, empty slots, etc.) as spare or return it. This approach adheres to the NDU procedure of swapping the "brains" of the storage while keeping the "body" (chassis and drives) intact.

=========