Doctor is capable of automatically detecting and trying to remediate performance or availability issues with the endpoint it is monitoring.
excerpt of config values specific to autohealing process below:
$ doctor --help
Usage of doctor:
--autoheal whether doctor should take active measures to attempt to heal the kava process (e.g. place on standby if it falls significantly behind live)
--autoheal_blockchain_service_name string the name of the systemd service running the blockchain. this is the service that gets restarted in the autoheal process (default "kava")
--autoheal_restart_delay_seconds int number of seconds autohealing routines will wait to restart the endpoint, effective from the last time it was restarted and over riding the values downtime_restart_threshold_seconds no_new_blocks_restart_threshold_seconds (default 2700)
--autoheal_sync_latency_tolerance_seconds int how far behind live the node is allowed to fall before autohealing actions are attempted (default 120)
--autoheal_sync_to_live_tolerance_seconds int how close to the current time the node must resync to before being considered in sync again (default 12)
--default_monitoring_interval_seconds int default interval doctor will use for the various monitoring routines (default 5)
--downtime_restart_threshold_seconds int how many continuous seconds the endpoint being monitored has to be offline or unresponsive before autohealing will be attempted (default 300)
--health_check_timeout_seconds int max number of seconds doctor will wait for a health check response from the endpoint (default 10)
--no_new_blocks_restart_threshold_seconds int how many continuous seconds the endpoint being monitored has not produce a new bloc before autohealing will be attempted (default 300)Default Kava Mainnet Doctor Config
Only if autoheal is set to true will autohealing routines be triggered.
Routines can run concurrently, e.g. a node may fall behind live and put on standby by one routine until it catches up, and if the node goes offline or stops making new blocks during the time it is on standby another routine will restart the kava process, and if the node syncs back to live the first auto healing process will put the node back in service with the autoscaling group.
If doctor detects that the node is offline for more than downtime_restart_threshold_seconds, it will attempt to restart the kava process on the node.
Out of Sync Heuristic and Auto Healing Workflow
If doctor detects that the node has fallen more than autoheal_sync_latency_tolerance_seconds behind the current time (comparing the latest block time for the node and the current time), it will attempt to place the node in standby with the autoscaling group so it won't have to serve requests and can sync faster, and if the node returns to within autoheal_sync_to_live_tolerance_seconds of the current time it will be placed back in service.
If doctor detects that the node has not synched a new block in more than no_new_blocks_restart_threshold_seconds, it will attempt to restart the kava process on the node.
The autohealing process assumes the chain is running via a systemd service. It uses a systemd restart to restart the chain. The name of this service is configurable via the configuration option autoheal_blockchain_service_name. By default, doctor uses the service name kava.