Pacemaker 2.1.3 - Final

Release info

• Fri Jun 03 2022 Ken Gaillot [email protected] Pacemaker-2.1.4-rc1
• Changesets: 11
• Diff: 8 files changed, 31 insertions(+), 11 deletions(-)

Known regressions introduced in this release

• Asynchronous IPC clients can leak file descriptors (fixed in 2.1.8)
• The scheduler can crash when logging an invalid utilization value (fixed in 2.1.8)
• Resources that need to move because they were reordered in the CIB might not do so until the next natural transition (fixed in 2.1.5)
• Fence devices cannot be registered if the local node is not in the CIB (fixed in 2.1.5)
• crm_attribute --utilization incorrectly sets the value of node utilization attributes (fixed in 2.1.5)
• crm_attribute --quiet outputs "(null)" instead of nothing if the requested attribute has no value (fixed in 2.1.5)
• crm_mon can return a failure code if pacemaker is shutting down, which can cause resource failures with resource agents that run crm_mon in monitor or stop actions (fixed in 2.1.5)
• crm_mon can leak memory when run in daemon mode (fixed in 2.1.5)
• attrd_updater rejects (deprecated and unused) --quiet option (fixed in 2.1.5)
• Fencing devices configured with target-by-attribute no longer work (fixed in 2.1.4)
• When debug logging is enabled, the fencer can have a use-after-free memory error when processing self-fencing requests with topology (fixed in 2.1.4)
• ocf:pacemaker:HealthSMART meta-data has a typo (fixed in 2.1.4)

Features added since Pacemaker-2.1.2

• Internal failures of resource actions (such as an OCF agent not being found) are shown with a detailed exit reason in logs, crm_mon output, etc.
• Support for CIB entries with type="ping" is deprecated (this was an undocumented means of defining a quorum-only node)
• build: configure script supports experimental --enable-nls option to enable native language translations (currently only Chinese translations of certain help text are available)
• rpm: crm_attribute is now part of the pacemaker-cli package instead of the pacemaker package
• CIB: resources support allow-unhealthy-nodes meta-attribute to exempt the resource from bans due to node health checks (particularly useful for health monitoring resources themselves)
• CIB: multiple-active cluster property can be set to "stop_unexpected" to leave the expected instance running and stop only any unexpected ones
• CIB: bundles support resource utilization
• pacemakerd: regularly check that subdaemons are active and accepting IPC connections so sbd can self-fence a node with a malfunctioning subdaemon
• pacemaker-schedulerd: support --output-as/--output-to options including XML output with interactive options
• tools: cibadmin --show-access option to show CIB colorized for ACLs
• tools: crm_attribute supports standard --output-as/--output-to options
• tools: crm_mon output indicates if a node's health is yellow or red
• tools: for probes that failed because the service is not installed or locally configured, crm_mon displays the resource as stopped rather than failed
• tools: crm_rule supports standard --output-as/--output-to options, allows passing multiple -r options, and is no longer experimental
• tools: stonith_admin fencing commands display reasons for failures
• resource agents: HealthSMART supports OCF 1.1 standard and new OCF_RESKEY_dampen and OCF_RESKEY_smartctl parameters

Fixes since Pacemaker-2.1.2

• build: avoid circular library dependency (regression introduced in 2.1.0)
• systemd: if pacemakerd exits immediately after starting, wait 1 second before trying to respawn, and allow 5 attempts
• fencer: get fencing completion time correctly (regression introduced in 2.1.2)
• fencer: avoid memory leak when broadcasting history differences (regression introduced in 2.1.0)
• controller: correctly match "node down" events so remote nodes don't get fenced when connection is stopped (regression introduced in 2.1.2)
• executor: avoid possible double free during notify operation (regression introduced in 2.1.1)
• tools: get stonith_admin -T/--tag option working again (regression introduced in 2.0.3)
• resources: use correct syntax in Stateful meta-data (regression introduced in 2.1.0)
• corosync: repeat corosync_cfg_trackstart if first attempt fails
• libcrmcommon: retry IPC requests after EAGAIN errors
• executor,scheduler: treat "no secrets" fence results as a hard error
• fencing: handle dynamic target query failures better
• fencing: don't set stonith action to pending if fork fails
• pacemakerd: avoid race condition when subdaemaon is checked while exiting
• scheduler: avoid memory leak when displaying clones in certain conditions
• scheduler: properly set data set flags when scheduling actions
• tools: support command-line crm_attribute calls on Pacemaker remote nodes whose node name in the cluster differs from their local hostname
• tools: prevent possible crm_resource crashes if multiple commands specified

Public API changes since Pacemaker-2.1.2

• libcrmcommon: pcmk_ipc_api_t supports scheduler IPC
• libpacemaker: add pcmk_status() (equivalent to crm_mon)
• libcib: deprecate get_object_parent()
• libcib: deprecate get_object_path()
• libcib: deprecate get_object_root()
• libcrmcommon: add pcmk_cib_parent_name_for()
• libcrmcommon: add pcmk_cib_xpath_for()
• libcrmcommon: add pcmk_find_cib_element()
• libcrmcommon: deprecate crm_xml_add_boolean()
• libpe_status: add pe_flag_check_config
• libpe_status: add pe_node_shared_s:data_set
• libpe_status: add pe_rsc_restarting flag
• libpe_status: add pe_rsc_stop_unexpected flag
• libpe_status: add recovery_stop_unexpected to enum rsc_recovery_type
• libpe_status: deprecate node_ping
• libpe_status: deprecate pe_order_stonith_stop
• libpe_status: deprecate pe_rsc_starting and pe_rsc_stopping
• libstonithd: add exit_reason member to stonith_history_t
• libstonithd: deprecate stonith_t:call_timeout
• libstonithd: stonith_api_del_notification() with NULL second argument removes all notifications