Pike Series Release Notes

Pike Series Release Notes

2.15.1

Bug Fixes

  • Fixed a bug introduced in 2.15.0 where the object reconstructor would exit with a traceback if no EC policy was configured.
  • Fixed deadlock when logging from a tpool thread.

    The object server runs certain IO-intensive methods outside the main pthread for performance. Previously, if one of those methods tried to log, this can cause a crash that eventually leads to an object server with hundreds or thousands of greenthreads, all deadlocked. The fix is to use a mutex that works across different greenlets and different pthreads.

  • The object reconstructor can now rebuild an EC fragment for an expired object.

Other Notes

  • Various other minor bug fixes and improvements.

2.15.0

New Features

  • Add Composite Ring Functionality

    A composite ring comprises two or more component rings that are combined to form a single ring with a replica count equal to the sum of the component rings. The component rings are built independently, using distinct devices in distinct regions, which means that the dispersion of replicas between the components can be guaranteed.

    Composite rings can be used for explicit replica placement and “replicated EC” for global erasure codes policies.

    Composite rings support ‘cooperative’ rebalance which means that during rebalance all component rings will be consulted before a partition is moved in any component ring. This avoids the same partition being simultaneously moved in multiple components.

    We do not yet have CLI tools for creating composite rings, but the functionality has been enabled in the ring modules to support this advanced functionality. CLI tools will be delivered in a subsequent release.

    For further information see the docs

  • The EC reconstructor process has been dramatically improved by adding support for multiple concurrent workers. Multiple processes are required to get high concurrency, and this change results in much faster rebalance times on servers with many drives.

    Currently the default is still only one process, and no workers. Set reconstructor_workers in the [object-reconstructor] section to some whole number <= the number of devices on a node to get that many reconstructor workers.

  • Add support to increase object ring partition power transparently to end users and with no cluster downtime. Increasing the ring part power allows for incremental adjustment to the upper bound of the cluster size. Please review the full docs for more information.
  • Added support for per-policy proxy config options. This allows per-policy affinity options to be set for use with duplicated EC policies and composite rings. Certain options found in per-policy conf sections will override their equivalents that may be set in the [app:proxy-server] section. Currently the options handled that way are sorting_method, read_affinity, write_affinity, write_affinity_node_count, and write_affinity_handoff_delete_count.
  • Enabled versioned writes on Dynamic Large Objects (DLOs).
  • Write-affinity aware object deletion

    Previously, when deleting objects in multi-region swift deployment with write affinity configured, users always get 404 when deleting object before it’s replicated to appropriate nodes.

    Now Swift will use write_affinity_handoff_delete_count to define how many local handoff nodes should swift send request to get more candidates for the final response. The default value “auto” means Swift will calculate the number automatically based on the number of replicas and current cluster topology.

  • Require that known-bad EC schemes be deprecated

    Erasure-coded storage policies using isa_l_rs_vand and nparity >= 5 must be configured as deprecated, preventing any new containers from being created with such a policy. This configuration is known to harm data durability. Any data in such policies should be migrated to a new policy. See See Launchpad bug 1639691 for more information.

  • Optimize the Erasure Code reconstructor protocol to reduce IO load on servers.
  • Fixed a bug where SSYNC would fail to replicate unexpired object.
  • Fixed a bug in domain_remap when obj starts/ends with slash.
  • Fixed a socket leak in copy middleware when a large object was copied.
  • Fixed a few areas where the swiftdir option was not respected.
  • swift-recon now respects storage policy aliases.
  • cname_lookup middleware now accepts a nameservers config variable that, if defined, will be used for DNS lookups instead of the system default.
  • Make mount_check option usable in containerized environments by adding a check for an “.ismount” file at the root directory of a device.
  • Remove deprecated vm_test_mode option.
  • The object and container server config option slowdown has been deprecated in favor of the new objects_per_second and containers_per_second options.
  • The output of devices from swift-ring-builder has been reordered by region, zone, ip, and device.
  • Imported docs content from openstack-manuals project.
  • Various other minor bug fixes and improvements.

2.14.0

New Features

  • EC Fragment Duplication - Foundational Global EC Cluster Support.
  • name_check and cname_lookup keys have been added to /info.
  • Add Vary headers for CORS responses.
  • Always set Swift processes to use UTC.
  • Removed per-device reconstruction stats. Now that the reconstructor is shuffling parts before going through them, those stats no longer make sense.
  • domain_remap now accepts a list of domains in “storage_domain”.
  • Do not follow CNAME when host is in storage_domain.
  • Enable cluster-wide CORS Expose-Headers setting via “cors_expose_headers”.
  • Cache all answers from nameservers in cname_lookup.

Bug Fixes

  • Fixed error where a container drive error resulted in double space usage on rest drives. When drive with container or account database is unmounted, the bug would create handoff replicas on all remaining drives, increasing the drive space used and filling the cluster.
  • Fixed UnicodeDecodeError in the object reconstructor that would prevent objects with non-ascii names from being reconstructed and caused the reconstructor process to hang.
  • Fixed encoding issue in ssync where a mix of ascii and non-ascii metadata values would cause an error.
  • Log the correct request type of a subrequest downstream of copy.
  • Prevent logged traceback in object-server on client disconnect for chunked transfers to replicated policies.
  • Fixed a race condition in updating hashes.pkl where a partition suffix invalidation may have been skipped.
  • Include received fragment index in reconstructor log warnings.
  • Log correct status code for conditional requests.

Other Notes

  • Drop support for auth-server from common/manager.py and swift-init.
  • Various other minor bug fixes and improvements.
Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.