Module cartridge.failover¶
Gather information regarding instances leadership.
Failover can operate in two modes:
- In disabled mode the leader is the first server configured in
topology.replicasets[].master
array. - In
eventual
mode the leader isn’t elected consistently. Instead, every instance in cluster thinks the leader is the first healthy server in replicaset, while instance health is determined according to membership status (the SWIM protocol). - In stateful mode leaders appointments are polled from the external storage. (Added in v2.0.2-2)
This module behavior depends on the instance state.
From the very beginning it reports is_rw() == false,
is_leader() == false
, get_active_leaders() == {}
.
The module is configured when the instance enters ConfiguringRoles state for the first time. From that moment it reports actual values according to the mode set in clusterwide config.
(Added in v1.2.0-17)
Functions¶
Local Functions¶
schedule_clear ()¶
Cancel all pending reconfigure_all tasks.
schedule_add ()¶
Schedule new reconfigure_all task.
_get_appointments_disabled_mode ()¶
Generate appointments according to clusterwide configuration. Used in „disabled“ failover mode.
_get_appointments_eventual_mode ()¶
Generate appointments according to membership status. Used in „eventual“ failover mode.
_get_appointments_stateful_mode ()¶
Get appointments from external storage. Used in „stateful“ failover mode.
accept_appointments (replicaset_uuid)¶
Accept new appointments.
Get appointments wherever they come from and put them into cache. Cached active_leaders table is never modified, but overriden by it’s modified copy (if necessary).
Parameters:
- replicaset_uuid: ({[string]=string}) to leader_uuid map
Returns:
(boolean) Whether leadership map has changed
fencing_check ()¶
Perform the fencing healthcheck.
Fencing is actuated when the instance disconnects from both the state provider and a replica, i.e. the check returns false.
Returns:
(boolean) true / false
failover_loop ()¶
Repeatedly fetch new appointments and reconfigure roles.
cfg ()¶
Initialize the failover module.
get_active_leaders ()¶
Get map of replicaset leaders.
Returns:
{[replicaset_uuid] = instance_uuid,…}
is_vclockkeeper ()¶
Check if current instance has persisted his vclock.
Returns:
(boolean) true / false
consistency_needed ()¶
Check if current configuration implies consistent switchover.
Returns:
(boolean) true / false
force_inconsistency (replicaset_uuid)¶
Force inconsistent leader switching. Do it by resetting vclockkepers in state provider.
Parameters:
- replicaset_uuid: ({[string]=string,…}) to leader_uuid mapping
Returns:
(boolean) true
Or
(nil)
(table) Error description
wait_consistency (replicaset_uuid)¶
Wait when promoted instances become vclockkepers.
Parameters:
- replicaset_uuid: ({[string]=string,…}) to leader_uuid mapping
Returns:
(boolean) true
Or
(nil)
(table) Error description