Pacemaker 2.0.0 - Release Candidate 1

• Deprecated features removed since Pacemaker-1.1.18

  • All of these have newer forms, and the cluster will automatically convert
    many older syntax usages in configurations to newer syntax as needed
  • Drop support for heartbeat and corosync 1 (whether using CMAN or plugin)
  • Drop support for rolling upgrades from Pacemaker versions older than 1.1.11
  • Drop support for PCMK_legacy and LRMD_MAX_CHILDREN environment variables
  • Drop support for legacy option aliases including default-action-timeout,
    default-resource-stickiness, resource-failure-stickiness,
    default-resource-failure-stickiness, is-managed-default,
    and all names using underbar instead of dash
  • Drop support for undocumented resource isolation feature
  • Drop support for "requires" operation meta-attribute
  • Drop support for operation meta-attributes in instance_attributes
  • Drop support for the pcmk_*_cmd, pcmk_arg_map, and pcmk_poweroff_action
    fence resource parameters
  • Drop support for processing very old saved CIB files (including
    pre-0.6.0 start failure entries, pre-0.6.5 operation history entries,
    pre-0.7 transition keys, pre-1.1.4 migration history entries,
    pre-1.0 XML configuration schemas, pre-1.1.6 ticket state entries, and
    pre-1.1.7 failed recurring operation history entries)
  • tools: drop support for deprecated command-line options to crmadmin,
    crm_attribute, crm_resource, crm_verify, crm_mon, and stonith_admin
  • tools: drop support for built-in SMTP and SNMP in crm_mon

• Features added since Pacemaker-1.1.18

  • The default location of the Pacemaker detail log is now
    /var/log/pacemaker/pacemaker.log (instead of being directly in /var/log),
    and Pacemaker will no longer use Corosync's logging preferences;
    configure script options are available to change default log locations
  • The master XML tag is deprecated in favor of a new "promotable"
    meta-attribute set to true with a standard clone tag, and the "master-max"
    and "master-node-max" master meta-attributes are deprecated in favor of
    new "promoted-max" and "promoted-node-max" clone meta-attributes;
    documentation now refers to these as promotable clones rather than
    master/slave, stateful or multistate clones, and to promotion scores
    instead of master scores
  • record-pending now defaults to TRUE (pending actions are shown in status)
  • A negative stonith-watchdog-timeout now tells the cluster to automatically
    calculate the value based on SBD_WATCHDOG_TIMEOUT (which was the behavior
    of 0 before 1.1.15; 0 retains its post-1.1.15 behavior of disabling use of
    the watchdog as a fencing device)
  • The undocumented restart-type and role_after_failure resource
    meta-attributes are now deprecated
  • systemd unit file removes limit on spawned processes
  • tools: all tools now return consistent, well-defined exit codes;
    crm_error has a --exit option to list these
  • tools: crm_resource --cleanup now cleans only failed operation history;
    crm_resource --reprobe retains the previous behavior of cleaning all
    operation history
  • All Python code in Pacemaker now supports both Python 2.7 and Python 3
  • Regression testing code has been consolidated and overhauled; most obvious
    change is new command names
  • Administration-related documentation has been moved from the
    Pacemaker Explained document to a new Pacemaker Administration document

• Changes since Pacemaker-1.1.18

  • Numerous changes to public API of libraries
  • Restore systemd unit dependency on DBus (regression in 1.1.17)
  • attrd: ensure node name is broadcast at start-up (CLBZ#5330, regression in 1.1.18)
  • crmd: quorum gain without a node join should cause new transition
  • pengine: ensure stop operations occur after stopped remote connections have been brought up
  • pengine: only allowed nodes need to be considered when ordering resource startup after all recovery
  • pengine: ordering bundle child stops/demotes after container fencing causes graph loops
  • pengine: remote connection resources are safe to to require only quorum
  • pengine: unfence before probing or starting fence devices (regression in 1.1.18)
  • pengine: correctly observe colocation constraints with bundles in the Master role
  • pengine: restart resource after failed demote when appropriate
  • Pacemaker Remote: always use most recent remote proxy
  • tools: prevent notify actions from causing crm_resource --wait to hang
  • tools: crm_master did not work without explicit --lifetime (regression in 1.1.18)

Pacemaker 1.1.18 - Final

Release info

• Changesets: 644
• Diff: 167 files changed, 9753 insertions(+), 5596 deletions(-)

Known regressions introduced in this release

• Agent meta-data is not always refreshed after a resource start, in case the agent was updated (fixed in 2.1.0)
• A resource may have unnecessary downtime when it needs to be moved to a Pacemaker Remote node that fails to come up (fixed in 2.0.1)
• attrd's node name might not be broadcast at start-up (fixed in 2.0.0)
• Unfencing might be incorrectly scheduled after probing or starting fence devices (fixed in 2.0.0)
• crm_simulate -sL does not show master scores (fixed in 2.0.0)
• crm_master does not work without explicit --lifetime (fixed in 2.0.0)

Features added since Pacemaker-1.1.17

• warnings are now logged when using legacy syntax to be removed in 2.0
• agents: ifspeed agent is now installed when building
• agents: ifspeed agent can optionally detect interface name from IP address
• alerts: support alert filters
• alerts: experimental support for alerts for node attribute changes
• crmd,pengine: support unfencing of remote nodes
• pengine: bundles now support all constraint types
• pengine: bundles now support rkt containers
• pengine: bundles support new container-attribute-target parameter
• pengine,tools: logs and crm_mon show why resources changed state
• stonith-ng: support new fencing resource parameter pcmk_delay_base
• tools: new crm_resource option --why explains why resources are stopped

Changes since Pacemaker-1.1.17

• many documentation improvements
• agents: ifspeed properly calculates speed of hfi1 interfaces
• agents: ClusterMon now interprets "update" less than 1000 as seconds
• attrd: don't lose attributes set between attrd start-up and cluster join
• attrd: fix multiple minor memory leaks
• crmd: correctly record that unfencing is complete
• crmd: error more quickly if remote start fails due to missing key
• lrmd: remote resource operations return immediate error if key setup fails
• lrmd: allow pre-1.1.15 cluster nodes to connect to current Pacemaker Remote
• pengine: guest nodes are now probed like other nodes
• pengine: probe remote nodes for guest node resources
• pengine: do not probe guest/bundle connections until guest/bundle is active
• pengine: allow resources to stop prior to probes completing
• pengine: bundles wait only for other containers on same node to be probed
• pengine: have bundles log to stderr so 'docker logs'/'journalctl -M' works
• pengine: only pass requests for promote/demote flags onto the bundle's child
• pengine: do not map ports into Docker container when net=host is specified
• pengine: allow resources inside bundles to receive clone notifications
• pengine: default to non-interleaved bundle ordering for safety
• pengine: ensure bundle nodes and child resources are correctly cleaned up
• pengine: prevent graph loops when fencing the host underneath a bundle
• pengine: fix multiple memory issues (use-after-free, use-of-NULL) with bundles
• pengine: resources in bundles respect failcounts
• pengine: ensure nested container connections run on the same host
• pengine: ensure unrecoverable remote nodes are fenced even with no resources
• pengine: handle resource migrating behind a migrating remote connection
• pengine: don't prefer to keep unique instances on same node
• pengine: exclude exclusive resources and nodes from symmetric default score
• pengine: if ignoring failure, also ignore migration-threshold
• pengine: restore the ability to send the transition graph via the disk if it gets too big
• pengine: validate no-quorum-policy=suicide correctly
• pengine: avoid crash when alerts section has comments
• pengine: detect permanent master scores at start-up
• pengine: do not re-add a node's default score for each location constraint
• pengine: make sure calculated resource scores are consistent on different architectures
• pengine: retrigger unfencing for changed device parameters only when necessary
• pengine: don't schedule reload and restart in same transition (CLBZ#5309, regression introduced in 1.1.15)
• stonith-ng: make fencing-device reappear properly after reenabling
• stonith-ng: include pcmk_on_action in meta-data so 'on' can be overridden
• tools: allow crm_report to work with no log files specified
• tools: fix use-after-free in crm_diff introduced in 1.1.17
• tools: allow crm_resource to operate on anonymous clones in unknown states
• tools: crm_resource --cleanup on appropriate nodes if we don't know state of resource
• tools: prevent disconnection from crmd during crm_resource --cleanup
• tools: improve messages for crm_resource --force-* options
• tools: crm_mon: avoid infinite process spawning if -E script can't be run
• tools: crm_mon: don't show previous exit-reason for failed action with none
• libcrmservice: list systemd unit files, not only active units (CLBZ#5299)
• libcrmservice: parse long description correctly for LSB meta-data

Pacemaker-1.1.18-rc4

• Features added since Pacemaker-1.1.18-rc3

  • agents: ifspeed agent can optionally detect interface name from IP address

• Changes since Pacemaker-1.1.18-rc3

  • crmd: error more quickly if remote start fails due to missing key
  • pengine: allow resources to stop prior to probes completing
  • pengine: probe remote nodes for guest node resources
  • pengine: bundles need to wait only for other containers on the same node to be probed
  • pengine: correctly defer processing of resources inside containers
  • pengine: do not always expire failed operations of nested remotes
  • pengine: do not probe connection resources until the container is active
  • pengine: don't keep unique instances on same node
  • pengine: don't probe offline or unclean guest nodes
  • pengine: exclude exclusive resources and nodes from symmetric default score
  • pengine: have bundles log to stderr so that 'docker logs' works
  • pengine: if ignoring failure, also ignore migration-threshold
  • pengine: only pass requests for promote/demote flags onto the bundle's child
  • pengine: resources are allowed to stop before their state is known everywhere
  • pengine: restore the ability to send the transition graph via the disk if it gets too big
  • pengine: do not map ports into Docker container when net=host is specified
  • tools: allow crm_resource to operate on anonymous clones in unknown states
  • tools: crm_resource --cleanup on appropriate nodes if we don't know state of resource
  • tools: prevent disconnection from crmd during crm_resource --cleanup

Pacemaker 1.1.18 - Release Candidate 3

• Changes since Pacemaker-1.1.18-rc2
  • crmd: fix regression for legacy stacks introduced in rc1
  • lrmd: allow pre-1.1.15 cluster nodes to connect to current Pacemaker Remote
  • pengine: don't schedule reload and restart in same transition
    (fixes CLBZ#5309, a regression introduced in 1.1.15)
  • pengine: fix use-after-free memory issue re-introduced by merge
  • tools: improve messages for crm_resource --force-* options

Pacemaker 1.1.18 - Release Candidate 2

• Changes since Pacemaker-1.1.18-rc1
  • fix regressions introduced in rc1 (don't break crm_resource --lifetime,
    avoid memory leak when caching meta-data,
    keep previous format of crm_mon XML output for bundles)
  • crm_mon: avoid infinite process spawning if -E script can't be run
  • crm_mon: don't show previous exit-reason for failed action with none
  • libcrmservice: list systemd unit files, not only active units (BZ#5299)
  • libcrmservice: parse long description correctly for LSB meta-data

Pacemaker 1.1.18 - Release Candidate 1

• Features added since Pacemaker-1.1.17

  • warnings are now logged when using legacy syntax to be removed in 2.0
  • agents: ifspeed agent is now installed when building
  • alerts: support alert filters
  • alerts: support alerts for node attribute changes (experimental)
  • crmd,pengine: support unfencing of remote nodes
  • pengine: bundles now support all constraint types
  • pengine: bundles now support rkt containers
  • pengine: bundles support new container-attribute-target parameter
  • pengine: guest nodes are now probed like any other node
  • pengine,tools: logs and crm_mon show why resources changed state
  • stonith-ng: support new fencing resource parameter pcmk_delay_base
  • tools: new crm_resource option --why explains why resources are stopped

• Changes since Pacemaker-1.1.17

  • many documentation improvements
  • agents: ifspeed properly calculates speed of hfi1 interfaces
  • agents: ClusterMon now interprets "update" less than 1000 as seconds
  • attrd: don't lose attributes set between attrd start-up and cluster join
  • attrd: fix multiple minor memory leaks
  • crmd: correctly record that unfencing is complete
  • lrmd: remote resource operations return immediate error if key setup fails
  • pengine: validate no-quorum-policy=suicide correctly
  • pengine: avoid crash when alerts section has comments
  • pengine: detect permanent master scores at start-up
  • pengine: do not re-add a node's default score for each location constraint
  • pengine: allow resources inside bundles to receive clone notifications
  • pengine: default to non-interleaved bundle ordering for safety
  • pengine: ensure bundle nodes and child resources are correctly cleaned up
  • pengine: prevent graph loops when fencing the host underneath a bundle
  • pengine: probe bundle connection resources
  • pengine: bundles now respect PCMK_authkey_location
  • pengine: fix multiple memory issues (use-after-free, use-of-NULL) with bundles
  • pengine: resources in bundles respect failcounts
  • pengine: ensure nested container connections run on the same host
  • pengine: ensure unrecoverable remote nodes are fenced even if no resources can run on them
  • pengine: handle resource migrating behind a migrating remote connection
  • pengine: make sure calculated resource scores are consistent on different architectures
  • pengine: retrigger unfencing for changed device parameters only when necessary
  • stonith-ng: make fencing-device reappear properly after reenabling
  • stonith-ng: include pcmk_on_action in meta-data so 'on' can be overridden
  • tools: allow crm_report to work with no log files specified
  • tools: fix use-after-free in crm_diff introduced in 1.1.17

Pacemaker 1.1.17 - Final

Release info

• Thu Jul 06 2017 Ken Gaillot [email protected] Pacemaker-1.1.17-1
• Changesets: 539
• Diff: 177 files changed, 11525 insertions(+), 5036 deletions(-)

Known regressions introduced in this release

• When run from a Pacemaker Remote node, crm_resource --cleanup incorrectly treats an unspecified node as the local node instead of all nodes (fixed in 2.1.0)
• Pacemaker's systemd unit incorrectly dropped a dependency on DBus (fixed in 2.0.0)
• Mixed-case node names are not handled properly when modifying attributes (fixed in 2.0.0)
• Scheduler can crash when logging ignored failure timeout (fixed in 2.0.0)
• crm_failcount does not handle INFINITY correctly (fixed in 2.0.0)
• crm_diff can have a use-after-free memory error (fixed in 1.1.18)

Features added since Pacemaker-1.1.16

• New "bundle" resource type for Docker container use cases (experimental)
• New "PCMK_node_start_state" environment variable to start node in standby
• New "value-source" rule expression attribute in location constraints to compare a node attribute against a resource parameter
• New "stonith-max-attempts" cluster option to specify how many times fencing can fail for a target before the cluster will no longer immediately re-attempt it (previously hard-coded at 10)
• New "cluster-ipc-limit" cluster option to avoid IPC client eviction in large clusters
• Failures are now tracked per operation type, as well as per node and resource (the "fail-count" and "last-failure" node attribute names now end in "#OPERATION_INTERVAL")
• attrd: Pacemaker Remote node attributes and regular expressions are now supported on legacy cluster stacks (heartbeat, CMAN, and corosync plugin)
• tools: New "crm_resource --validate" option
• tools: New "stonith_admin --list-targets" option
• tools: New "crm_attribute --pattern" option to match a regular expression
• tools: "crm_resource --cleanup" and "crm_failcount" can now take --operation and --interval options to operate on a single operation type

Changes since Pacemaker-1.1.16

• Fix multiple memory issues (leaks, use-after-free) in libraries
• pengine: unmanaging a guest node resource puts guest in maintenance mode
• cib: broadcasts of cib changes should always pass ACL checks
• crmd,libcrmcommon: update throttling when CPUs are hot-plugged
• crmd: abort transition whenever we lose quorum
• crmd: avoid attribute write-out on join when atomic attrd is used
• crmd: check for too many stonith failures only when aborting for that reason
• crmd: correctly clear failure counts only for a specified node
• crmd: don't fence old DC if it's shutting down as soon-to-be DC joins
• crmd: forget stonith failures when forgetting node
• crmd: all nodes should track stonith failure counts in case they become DC
• crmd: update cache status for guest node whose host is fenced
• dbus: prevent lrmd from hanging on dbus calls
• fencing: detect newly added constraints for stonith devices
• pengine: order remote actions after connection recovery (regression introduced in 1.1.15)
• pengine: quicker recovery from failed demote
• libcib: determine remote nodes correctly from node status entries
• libcrmcommon: avoid evicting IPC client if messages spike briefly
• libcrmcommon: better XML comment handling prevents infinite election loop
• libcrmcommon: set month correctly in date/time string sent to alert agents
• libfencing,fencing: intelligently remap "action" wrongly specified in config
• libservices: ensure completed ops aren't on blocked ops list
• libservices: properly detect and cancel in-flight systemd/upstart ops
• libservices: properly watch writable DBus handles
• libservices: systemd service that is reloading doesn't cause monitor failure
• pacemaker_remoted: allow graceful shutdown while unmanaged
• pengine,libpe_status: don't clear same fail-count twice
• pengine: consider guest node unclean if its host is unclean
• pengine: do not re-add a node's default score for each location constraint
• pengine: avoid restarting services when recovering remote connection
• pengine: better guest node recovery when host fails
• pengine: guest node fencing doesn't require stonith enabled
• pengine: allow probes of guest node connection resources
• pengine: properly handle allow-migrate explicitly set for remote connection
• pengine: fence failed remote nodes even if no resources can run on them
• tools: resource agents will now get the correct node name on Pacemaker Remote nodes when using crm_node and crm_attribute
• tools: avoid grep crashes in crm_report when looking for system logs
• tools: crm_resource -C now clears last-failure as well as fail-count
• tools: implement crm_failcount command-line options correctly
• tools: properly ignore version with crm_diff --no-version

Pacemaker 1.1.17 - Release Candidate 4

• Features added since Pacemaker-1.1.17-rc3

  • cib,libcib: support option for IPC eviction threshold

• Changes since Pacemaker-1.1.17-rc3

  • cib: Broadcasts of cib changes should always pass ACLs check
  • libcib: get remoteness correctly from node status
  • libpe_status: avoid memory leaks when creating bundle remote resource
  • libpe_status: disallow resources on bundle nodes
  • pengine: Bundle location constraints should only apply to the IP and docker resources
  • pengine: Clones within bundles may also have notifications enabled
  • pengine: Correctly implement pe_order_implies_first_printed
  • pengine: Do not re-add a node's default score for each location constraint

Pacemaker 1.1.17 - Release Candidate 3

• Features added since Pacemaker-1.1.17-rc2

  • Support inheritable meta-attributes for bundles

• Changes since Pacemaker-1.1.17-rc2

  • crmd: avoid attribute write-out on join when atomic attrd is used

Pacemaker-1.1.17-rc2

• Update source tarball to revision: 8fdb646

• Changesets: 19

• Diff: 10 files changed, 99 insertions(+), 58 deletions(-)

• Changes since Pacemaker-1.1.17-rc1

  • fencing: Detect newly added constraints for stonith devices
  • pengine,libpe_status: multiple bug fixes for new bundle feature
  • pengine: Order remote actions after connection recovery
    (regression introduced in 1.1.15)