Ceph Developer Monthly Report: New SeaStore Engine, RGW Enhancements, and Backend Improvements

This is the fifth edition of the 2018 Ceph Developer Monthly Report, summarizing the most interesting commits and important changes in the Ceph community for the month.

SeaStore Design Goals

Target NVMe devices only, excluding PEME and HDD.

Use SPDK for user‑space I/O.

Leverage the SeaStar framework with future&promise programming.

Rework the network messaging layer using SeaStar and DPDK to achieve minimal‑copy I/O paths.

SeaStore requires OSD refactoring and integration with SeaStar.

Object Storage Updates

Improvements to radosgw-admin bucket sync status (now radosgw-admin bucket sync markers) provide detailed shard‑level sync statistics across source zones.

$ radosgw-admin bucket sync status --bucket BUCKET
realm 43b972d8-9448-4bd5-a396-1415dfcc483a (dev)
zonegroup 18b95c94-333f-465c-9869-cc06869eda1e (na)
zone 1508260a-6c32-49db-8248-562aa1881c90 (na)
bucket BUCKET[7559751c-8905-4c4b-bfab-83412318e701.4134.1]

source zone 65ab7537-c7e9-48a3-97b2-e5808744140a (na)
full sync: 0/8 shards
incremental sync: 8/8 shards
bucket is behind on 2 shards
behind shards: [0,3]
source zone 7559751c-8905-4c4b-bfab-83412318e701 (na)
full sync: 8/8 shards
full sync: 123 objects completed
incremental sync: 0/8 shards
bucket is behind on 8 shards
behind shards: [0,1,2,3,4,5,6,7]

Block Storage Updates

rbd disk-usage now supports an --exact flag for precise on‑disk usage calculation.

librbd deep copy can optionally flatten cloned images, removing parent‑child relationships during copy.

rbd-mirror adds optional active/active replication for higher availability and scalability.

Unified Storage Layer Updates

OSD can now force‑create placement groups after failure with the --yes-i-really-mean-it confirmation flag.

Monitor metadata command now displays all available compression algorithms, improving configuration visibility.

Cluster Management Enhancements

Improved mgr module disable command with clearer error handling for already disabled, non‑existent, or already enabled modules.

Centralized configuration management for mgr modules using set_config / get_config and KV store APIs.

Mgr service daemons can now unregister from the ServiceMap during shutdown.

Dashboard now supports TLS, RGW user and bucket management, and adds a notification panel for missing RGW configuration.

New mgr modules: telegraf for metric collection, and telemetry for anonymous cluster reporting.

Foundation Libraries

Added QAT (Intel Quick Assist Technology) compression support, enabling hardware‑accelerated compression and decompression for large‑data workloads.

Tooling Improvements

Log write coalescing merges multiple log entries into a single disk write, reducing system‑call overhead. ceph-volume lvm list now includes physical devices associated with logical volumes.

Integrated Cosbench benchmark into Teuthology’s CBT task for comprehensive testing.