What's New in Infernalis The Rise of the Vampire Squid By Bryan Stillwell & Ian Colle - February 9, 2016
Agenda Welcome (food/drinks) Introductions Presentations Q&A What's New in Infernalis (Bryan Stillwell) What's Coming in Jewel (Ian Colle) Q&A Discussions
Introductions Who are you? What brings you here today?
Future Meetup Ideas How you're using Ceph CephFS RBD (RADOS Block Device) RGW (RADOS Gateway) RADOS Monitoring (Calamari/VSP/Grafana) Automation (Puppet/Ansible/Salt) Cache Tiering CRUSH Performance Tuning CBT (The Ceph Benchmarking Tool)
What's New in Infernalis (v9.2.x)
Old Version Numbering Codename Release Type Versions Argonaut Stable 0.48 -> 0.48.3 Bobtail 0.56 -> 0.56.7 Cuttlefish 0.61 -> 0.61.9 Dumpling LTS 0.67 -> 0.67.11 Emperor 0.72 -> 0.72.2 Firefly 0.80 -> 0.80.11 Giant 0.87 -> 0.87.2 Hammer 0.94 -> 0.94.5
New Version Numbering x.y.z x Stable release number (I == 9, J == 10, K == 11) y 0 = Development (early testers) 1 = Release Candidate (test clusters) 2 = Stable (production ready) z Revision/Bugfix release
New Version Numbering Codename Type Versions Infernalis Stable 9.2.x Jewel 10.2.x Kraken 11.2.x Note – all releases are Stable now. Removed concept of LTS.
General Improvements systemd (except for Ubuntu Trusty) systemctl start ceph-osd* systemctl stop ceph-osd@42 RedHat distros now have an SELinux policy
General Improvements Ceph daemons run as 'ceph' user chown -R ceph:ceph /var/lib/ceph Time consuming Must be done on mon nodes as well Problem with journals not getting chown'd Reboots fix, but only if you're using GPT
RADOS Improvements Cache tier improvements SHEC erasure coding no longer experimental Unified queue for handling client IO, recovery, scrubbing, and snapshot trimming Many ceph-objectstore-tool improvements Cleanup of ObjectStore API to facilitate new backends like BlueStore SHEC = Shingled Erasure Code contributed by Fujitsu (improves recovery efficiency) ceph-objectstore-tool
RGW Improvements Added support for SWIFT expirations Many SWIFT API compatibility improvements SWIFT expirations work well for storing database backups for example
RBD Improvements The size argument now supports suffixes Hammer (0.94.5): Infernalis (9.2.0): Default format also changed
RBD Improvements The rbd du command now shows actual usage (quickly if fast-diff enabled)
RBD Improvements Many stability improvements to object-map Can enable/disable object-map and exclusive-lock features dynamically Object-map was added during the Hammer release cycle. It tracks which blocks of the image are actually allocated and where.
RBD Improvements New rbd status command
RBD Improvements You can now store user metadata and set persistent librbd options associated with individual images Deep-flatten now supports snapshots The export-diff tool is now faster (now uses aio)
CephFS Improvements Snapshots can now be renamed Ongoing improvements around administration, diagnostics, and check and repair tools Dramatic improvements around caching a revocation of client cache state due to unused inodes The ceph-fuse client behaves better on 32bit hosts
What's Coming in Jewel (v10.2.x)
RADOS Improvements Continuing to make scrubbing smarter (don’t scrub when the OSD is busy, if scrub starts, finish as quickly as possible) Further ceph-objectstore-tool improvements Still evaluating tcmalloc 2.4 vs jemalloc performance BlueStore! – uses LevelDB for journaling – greatly improves performance of RADOS Further Cache-tiering improvements (but still caveat emptor)
RGW Improvements NFS for RGW Multisite v2 Keystone v3 Support Originally master-slave with sync agent, which only allowed writes in master location Now master-master Keystone v3 Support Swift per tenant namespace Static Web Hosting in S3 Bucket Passes TEMPEST
RBD Improvements Async RBD Mirroring iSCSI integration
CephFS Improvements Continued improvements around administration, diagnostics, and check and repair tools Active-Passive MDS now considered tech preview Stabilizing work on Active-Active
General Moved from Ceph Developer Summit (CDS) every 3-4 months to Ceph Design Monthly (CDM) on the first Wed of each month, alternating between European and Asian friendly times Next CDM 02MAR2016 2100EST Videos available here: https://www.youtube.com/channel/UCno-Fry25FJ7B4RycCxOtfw
Questions?