Comprehensive Guide to Rubrik Data Management and SLA Domains
Atlas File System and SDFS Architecture
- The file system utilized within Rubrik clusters is known as Atlas, also referred to as the Scale Data File System (SDFS).
- Atlas is a masterless, distributed file system designed specifically for scalability, fault tolerance, and high performance.
- The file system operates as a single entity that combines the direct-attached storage (DAS) from all nodes within the cluster.
- Intelligence within SDFS allows it to scale dynamically as the cluster size changes.
- Backup and restore operations can be executed as single-node or multi-node processes.
Data Resiliency and Striping Mechanism
- Data is broken down into specific units to ensure resiliency and efficient distribution across the cluster.
- A file is divided into 32MB stripes.
- Reed-Solomon erasure coding, a form of error correction calculation, is performed on these stripes to provide data protection.
- These stripes are then written across 4 disks randomly distributed throughout the Rubrik cluster.
- Shards are created as the sum of the data on the 4 stripes written to the disks.
- The primary goal of this architecture is resiliency; if physical disks are lost, the system retains enough data on the shards to perform a full restore.
- When an application writes to SDFS on one node, the file system distributes those writes across disks in the cluster and sends the changes to the Callisto metadata store to update relevant metadata records.
The Snapshot Chain and Chain Reversal Process
- For most objects, Rubrik takes only one full snapshot or backup when protection is first initiated.
- Thereafter, only incremental changes (deltas) are captured from the workload, though exceptions are discussed in advanced series.
- Objects are stored as a single logical chain. Operations such as consolidation and chain reversal occur on this same chain.
- Chain Reversal Logic:
- This process moves the full snapshot point to the front of the chain periodically.
- It serves as a "synthetic full" (a representation of the full contents, not a new full backup).
- This allows for "incremental snapshots forever," providing a base for continuous incremental backups without needing a new full, which increases restoration (materialization) efficiency.
- The full is never stuck at a specific point in the chain.
- Execution Frequency:
- On a physical cluster, chain reversal occurs hourly.
- On virtual nodes, chain reversal occurs every 2 hours.
- Triggers for Chain Reversal:
- A maximum chain length of 64 snapshots is reached.
- A minimum change rate of 50% is detected (high data ingestion between snapshots).
- Admin Interaction: The chain reversal process is an automated background task that cannot be edited or changed by an administrator.
Data Availability and Fault Tolerance Domains
- Data availability concerns maintaining access to information even when components of the system fail. This is dependent on the deployment method and cluster scale.
- Availability Domains by Hardware:
- Rubrik Edge: This is a virtualized version of a single-node deployment. It lacks availability domains; if the node fails, data is unavailable. Replication and archival are strongly recommended in this configuration.
- Typical Cluster (Up to 12 nodes / 3 appliances): This configuration can withstand the loss of 1 node OR the loss of any 2 physical disks being offline simultaneously.
- Large Cluster (16 or more nodes / 4 or more appliances): This configuration can withstand the loss of an entire appliance (4 nodes) OR the loss of any 2 physical disks.
- Dynamic Rebalancing: If a section of the cluster is lost, the cluster automatically redefines its availability domain and rebalances data to determine where it can and cannot safely write.
SLA Domain Definition and Core Features
- SLA Domains are the backup policies that govern the behavior, cadence, and retention of data within the Rubrik environment.
- Core Features:
- Cadence: Defining how frequently snapshots are taken.
- Retention: Defining how long the snapshots are kept.
- Workload Requirements: Managing specific tasks like transaction log backups for databases (SQL/Oracle) and Continuous Data Protection (CDP) for VMware.
- Snapshot Windows: These are optional timeframes for backup activity.
- Remote Settings: Definitions for where data is sent for archival and replication.
- Protection Terminology:
- SLA Domain: The overall backup policy.
- Workload/Object: The entity being backed up (e.g., a VM, a database, or a file set).
- Snapshot: The actual backup representing a specific point-in-time.
The Lifecycle of Protected Data: Expiration, Consolidation, and Garbage Collection
- As defined by the SLA, segments of the snapshot chain move through several phases:
- Expiration: Parts of the chain reach their retention limit and are flagged for removal.
- Garbage Collection: A background job that identifies flagged data and performs the actual deletion.
- Consolidation: This process merges parts of the chain to remove logical gaps, making the chain shorter, more stable, and more space-efficient.
SLA Creation and Configuration Wizard
- To create an SLA, the administrator uses the Rubrik Security Cloud (RSC) interface, navigating to the SLA Domains section.
- Object Selection: The wizard requires selecting object types (e.g., Linux file sets, VMware, SQL) early on, as this unlocks specific configuration steps later (e.g., log backups).
- Cadence and Retention: Defined as intervals (e.g., four times a day).
- Snapshot Windows:
- Optional settings to restrict backups to specific hours.
- Includes a specific window for the First Full Snapshot, which is likely to be the most resource-intensive and long-running since it captures the entire object rather than just deltas.
- Advanced Frequencies: Allows for more granular scheduling, such as quarterly or monthly backups with distinct retention periods.
- Retention Lock: Prevents accidental or malicious deletion.
- Secondary Storage: Configuring archival (long-term storage) and replication (off-site copy) locations.
- Object Specifics: Configuring database logs or VMware CDP if applicable.
SLA Inheritance and Hierarchy
- SLAs can be applied to individual objects or at various levels of the object hierarchy.
- Inheritance Logic: Lower tiers derive their protection status from assignments made higher in the hierarchy.
- Example (vSphere): If an SLA is applied to an ESXi host, any VMs discovered on that host inherit that SLA automatically.
- Example (SQL): Applying protection at the instance level results in all underlying databases inheriting that protection.
- Applying Changes:
- Administrators can "Apply changes to existing snapshots" (retroactive application).
- If disabled, changes only apply to new snapshots taken moving forward.
Advanced Retention and Locking Features
- "Don't Protect" Options:
- Use Current SLA for Retention: Stops taking new backups but retains existing ones according to the original SLA timeline.
- Keep Forever: Keeps backups locally on the CDM cluster indefinitely; they will not expire unless manually deleted.
- Expire Immediately: Marks snapshots for immediate removal.
- Intelligent Data Lock: This is a built-in safety mechanism that prevents snapshots from being deleted for 7 days, even if "Expire Immediately" is selected. Support must be contacted to stop this clock in malicious scenarios.
- Retention Locked (WORM) SLAs:
- Compliance Mode: Support-led locking. Requires a support ticket and signed DocuSign documents. This is a prerequisite for the Rubrik Ransomware Warranty.
- Governance Mode: Self-governing via Quorum/Quorum Authorization (qAuth). Changes require approval from a designated group of approvers; this does not qualify for the Ransomware Warranty.
- Legal Hold: Stops the expiration clock on specific snapshots. It holds the data in place on both local CDM and archival locations. Removing the hold restores the original expiration schedule.
SLA Compliance Logic and Snapshot Intervals
- Interval Prioritization: Frequency settings determine when the next backup starts. In a set with 4-hourly, daily, and monthly snapshots, the system executes based on the most frequent cadence (4-hourly).
- Snapshot Overlap and Flags: The system does not take separate backups for daily and hourly requirements if they overlap. Instead, it takes the snapshot based on the most frequent cadence and changes the "flag" on the last snapshot of the period to mark it as the daily or monthly copy.
- Compliance Checks: The system checks if the data residing in the environment matches the definitions in the SLA.
- Example Logic:
- If an SLA requires snapshots every 4 hours, but a snapshot fails, the system becomes "Out of Compliance."
- To attempt recovery of compliance, the system may re-flag the next successful snapshot to fill the role of a missed higher-tier snapshot (like a daily).
Snapshot Inventory and Management
- The Snapshot Management section in the UI provides a view beyond the standard calendar page.
- Types of Manageable Snapshots:
- On-Demand Snapshots: Prompted by the user to occur as soon as possible, ignoring windows.
- Relic Snapshots: Backups for objects no longer in the production inventory (e.g., a deleted VM or dropped database) that are still retained as per SLA.
- Retrieved Snapshots: Data downloaded from an archive back to the local cluster.
- Unprotected Snapshots: Snapshots no longer governed by an active SLA.
- Deletion Restrictions: The "Delete Snapshots" option is grayed out if an object is actively protected. The object must be moved to "unprotected" or "forever" status before manual deletion is permitted (except for retention-locked snapshots).
System Monitoring and Logging Operations
- Spray Server: This is the name for the GUI or API service on the CDM level. It logs web UI interactions, REST API calls, and LDAP information.
- Syslog Support: Rubrik recommends connecting a third-party syslog server for prolonged log retention.
- Audit Logs: Every change to SLA settings or the environment is recorded in an audit log that can be exported to a CSV file or queried via API.
- Reference Metadata: An Access Denied error mentioned in context provided the following technical markers: RequestId is
BY125WYS0GDNYKGC, and HostId is IpZluNftaviNxeAQK3rZdcSz4bPPh9StbGFl3ecc37RrvkC0TDvveIiXkPqus+Hi16SBBNwkICQ=.