1.8.0 - scenarios and boost

Bug fixes

• Role installation will be completely skipped
  if you specify a tag other than the tags for this role
• Fixed selecting control instance that doesn't belong to cluster or isn't alive.
  The following rules are currently used:
  • Members are checked in lexicographic order by URIs
  • Members not mentioned in hostvars aren't selected to be control
  • Members with status not alive aren't selected to be control
• Fixed setting needs_restart when configuration files don't exist
• Fixed error on configuring auth without users specified
• Reset role variables before each run

Features

• cartridge-replicasets tag added to the membership stage
• cartridge_wait_buckets_discovery parameter to wait for instance to discover buckets
• instance_discover_buckets_timeout parameter to configure time in seconds
  to wait for instance to discover buckets
• Ability to deploy TGZ packages
• cartridge_multiversion flag that allows to use specific version of application
  for each instance and perform rolling update correctly (using new update_instance step)
• rotate_dists step that allows to rotate application distributions
• cleanup step to remove temporary files from specific list
• Added availability to import steps by scenario name. Added some default scenarios.
  Added availability to create custom scenarios.
• availability to use tasks_from to import any step
• zone variable to edit instance zone
• edit_topology_timeout variable to wait until cluster become healthy
  after editing topology
• availability to specify instance memtx_dir, vinyl_dir and wal_dir params
  by cartridge_memtx_dir_parent, cartridge_vinyl_dir_parent, cartridge_wal_dir_parent variables.
• Control instance is selected considering two-phase commit version of instances.
  The reason is that all operations that modify cluster-wide config should be performed via instance
  that has lowest Cartridge version (in fact, only two-phase commit version matters).
• Availability to change advertise URIs of any instance

Changes

• cartridge.admin_edit_topology is called once for all replicasets and instances
  to expel. It can be called second time to set up failover priority for
  replicasets where new instances were joined.
  As a result, replicaset_healthy_timeout is removed as unused.
• Now list of instances for installing a package is selected once for all.
  Before this patch, the complexity of calculating the list of instances was O(N^2),
  now it is O(N). For 100 instances, it gives a 10x time reduction (60s -> 5s).
• Refactored package installing. Getting package info is performed in a library module,
  all tasks except installing package itself are common for RPM and DEB.
• Now check_instance_started function: check all instances, including the stateboard;
  wait Unconfigured or RolesConfigured status instead alive state; check that
  all buckets are discovered by routers if cluster was bootstrapped.
• Role divided into many steps. It's possible to combine them using a scenario
  in the config by cartridge_scenario. It is also possible to use custom steps in a scenario.
  Custom steps can be defined by cartridge_custom_steps_dir and cartridge_custom_steps.
• Now step connect_to_membership is executed only on one not expelled instance.
  Before the patch, the difficulty of performing connect_to_membership step was N^2.
  For 100 instances, the step took about 900 seconds. Now the complexity has decreased to N,
  so for 100 instances the execution time is about 5 seconds.