Data Replication, Snapshots, Mirroring, Auditing, and Metrics Collection


Data from one of the replica containers is first offloaded and then the data in all the replica containers is purged. MapR Filesystem only stores the metadata after data is offloaded. The offload is considered successful only when data on all active replicas have been purged (or removed from the storage pool to release the disk space on the MapR file system). If, during the offload, the node on which one of the replicas reside is down, the data on that container is purged once the node comes back up.

In the tiering architecture, although data is moved to the storage tier, the namespace of the volume continues to be 3-way replicated. So, the metadata related to namespace container has 3x cost.

The offloaded replica containers are recalled if/when the whole volume is recalled. When a replica is reinstituted to the cluster as a result of a recall operation, a re-synchronization happens to bring all the replicas up to date from the designated master container.

Note: The offload and recall settings on the master container are applicable to the replica containers as well.


You can associate a snapshot schedule with tiering-enabled volumes. When the data in the volume is offloaded, associated snapshots are also offloaded and MapR Filesystem only stores the metadata. If the whole volume is recalled, the snapshots are also recalled to MapR file system. When offloading recalled snapshots, the rules for data offload apply to snapshots as well.

Note: You may experience latencies when accessing snapshots associated with offloaded data.


You can create tiering-enabled source volumes and associate them with tiering-enabled mirror volumes. You cannot associate tiering-enabled mirror volumes with standard volumes that are not tiering-enabled and vice versa. Only homogeneous combination of mirror and standard volumes are supported; heterogeneous combination of mirror and standard volumes are not supported.

Note: Both mirror volume and source volume data can be set up to be offloaded to the same tier (i.e., the same cold tier) or different tiers (i.e., different cold tiers). MapR does not require the source and mirror volume to be configured to use the same tier or have the same tier settings. Warm tier enabled volumes can have the same tier settings; however, the volume's tier only stores the meta data and data in each volume is offloaded to an associated back-end volume.

When a synchronization of the tiering-enabled mirror volume with the (local or remote) tiering-enabled source volume is triggered (either manually or automatically based on a schedule), the mirror volume synchronizes with the source volume if source volume data is local (and not yet tiered). On the other hand, if the source volume data is tiered, the tiering-enabled mirror volume synchronizes with the tiered data fetched by the MAST Gateway that is assigned to the source volume. Incremental changes in the mirror volume are offloaded based on the offload rules associated with the tiering-enabled mirror volume.

Using Tiering-Enabled Mirror Volumes for Disaster Recovery
You can create a secondary, cost optimized disaster recovery cluster for a primary three-way replicated cluster. To do this, create two clusters — a primary tiering-enabled cluster with no active schedule to automatically offload data and an associated secondary cluster where primary cluster data is mirrored and then aggressively offloaded to the tier. While the primary or source cluster continues to be three-way replicated, if the the secondary, disaster recovery cluster data is:
  • Erasure coded (warm tier), it provides space savings in the range of 1.2x-1.5x.
  • On a third-party cloud storage (cold tier), it can be three-way replicated on a low-cost storage alternative.
In case of a disaster, you can recall data from the tier to the MapR cluster.
Note: If you promote a tiering-enabled mirror volume during an offload or recall operation of the data associated with the mirror volume, the offload or recall operation is aborted and the mirror volume is converted to a read-write volume; the tierjobstatus command for the offload or recall job shows AbortedInternal status.


The MapR audit feature lets you log audit records of cluster-administration operations and operations on the data in the volume. Scheduled (and automatically triggered) tiering operations such as offload and compaction are not audited. However, if auditing is enabled at the cluster level, the manually triggered volume-level tiering operations such as offload, recall, abort, etc. are audited in the CLDB audit logs. For example, you can see a record similar to the following in the /opt/mapr/logs/cldbaudit.log.json file for volume offload command:


If auditing is enabled for data in the tiering-enabled volume and files within, file-level tiering operations such as offload, recall, etc. triggered using the REST API, hadoop, and dot-interface are audited in the FS audit logs (/var/mapr/local/<hostname>/audit/5661/FSAudit.log-<*>.json file).See Auditing Data Access Operations for the list of file-level tiering operations that are audited. You can selectively enable or disable auditing of these operations. See Selective Auditing of MapR Filesystem, MapR Database Table, and MapR Event Store For Apache Kafka Operations Using the CLI for more information. For example, you can see records similar to the following in the /var/mapr/local/<hostname>/audit/5661/FSAudit.log-<*>.json file for file offload command:


Both the tier rule list and tier list commands are audited in the /opt/mapr/logs/cldbaudit.log.json file as well as the /opt/mapr/mapr-cli-audit-log/audit.log.json file. The record in the audit log might look something similar to the following:


When auditing operations like tierjobstatus and tierjobabort, the coalesce interval set at the volume level is not honored. You may see multiple records of the same operation from the same client in the log.

Read requests processed using cache-volumes or erasure-coded volumes are not audited because when the file is accessed, the request first goes to the front-end volume and the operation is audited there. The audit record contains the ID of the front-end volume (volid) and primary file ID (fid). However, the write to the cache-volume for a volume-level recall of data is audited in the audit logs on the file server hosting the cache-volume with the primary file ID (fid). The write to the cache-volume for a file-level recall of data is not audited.

In addition, you can enable auditing of offload and/or recall events at both the volume and file levels by enabling auditing for filetieroffloadevent and filetierrecallevent at the volume level. By default, auditing is disabled for filetieroffloadevent and filetierrecallevent. If you enable auditing for filetieroffloadevent and filetierrecallevent using the dataauditops parameter with the volume create or volume modify command, the following are audited in the FS audit log:

  • For filetieroffloadevent, files offloaded by running the file offload command or (only) files purged on MapR file system after running volume offload command.
  • For filetierrecallevent, files recalled by running the file recall or volume recall command.

For example, you can see a record similar to the following in the /var/mapr/local/<hostname>/audit/5661/FSAudit.log-<*>.json file if auditing is enabled at the volume-level for filetieroffloadevent:{"timestamp":{"$date":"2018-06-07T07:27:58.810Z"},"operation":"FILE_TIER_OFFLOAD_EVENT","uid":2000,"ipAddress":"1}

For more information:

Collecting Metrics

If volume metrics collection is enabled on the tiering-enabled volume, metrics for all read and write operations on the tiered volume are logged in the metrics log. For example, you can see a record similar to the following in the metrics log file:
Tiering-related operations do not generate metrics records. That is, volume and file level offload, recall, and abort operations are not logged in the metrics log. However, the volumes created to support tiering (such as the cache-volume, the metadata volume, and the erasure-coded volume) have metrics collection enabled and the metrics records for these volumes are logged with the ID of the associated parent or front-end volume. That is, read operations on the the cache-volume are logged with the ID of the associated front-end volume. For example, you can see records similar to the following in the metrics log file for the volume:
See Enabling Volume Metric Collection and Collecting Volume Metrics for more information.