NAME

DESCRIPTION

CONFIGURATION FILE SYNTAX

# Global section
control <addr>
listen <addr>
replication <mode>
checksum <algorithm>
compression <algorithm>
timeout <seconds>
exec <path>
metaflush on | off
pidfile <path>

on <node> {
	# Node section
        control <addr>
        listen <addr>
        pidfile <path>
}

on <node> {
	# Node section
        control <addr>
        listen <addr>
        pidfile <path>
}

resource <name> {
	# Resource section
	replication <mode>
	checksum <algorithm>
	compression <algorithm>
	name <name>
	local <path>
	timeout <seconds>
	exec <path>
	metaflush on | off

	on <node> {
		# Resource-node section
		name <name>
		# Required
		local <path>
		metaflush on | off
		# Required
		remote <addr>
		source <addr>
	}
	on <node> {
		# Resource-node section
		name <name>
		# Required
		local <path>
		metaflush on | off
		# Required
		remote <addr>
		source <addr>
	}
}

Most of the various available configuration parameters are optional. If parameter is not defined in the particular section, it will be inherited from the parent section. For example, if the listen parameter is not defined in the node section, it will be inherited from the global section. In case the global section does not define the listen parameter at all, the default value will be used.

CONFIGURATION FILE DESCRIPTION

The following statements are available:

control ⟨addr⟩

Address for communication with hastctl(8). Each of the following examples defines the same control address:

uds:///var/run/hastctl
unix:///var/run/hastctl
/var/run/hastctl
    

The default value is uds:///var/run/hastctl.

pidfile ⟨path⟩

File in which to store the process ID of the main hastd(8) process.

The default value is /var/run/hastd.pid.

listen ⟨addr⟩

Address to listen on in form of:

protocol://protocol-specific-address
    

Each of the following examples defines the same listen address:

0.0.0.0
0.0.0.0:8457
tcp://0.0.0.0
tcp://0.0.0.0:8457
tcp4://0.0.0.0
tcp4://0.0.0.0:8457
    

Multiple listen addresses can be specified. By default hastd listens on tcp4://0.0.0.0:8457 and tcp6://[::]:8457 if kernel supports IPv4 and IPv6 respectively.

replication ⟨mode⟩

Replication mode should be one of the following:

memsync

Report the write operation as completed when local write completes and when the remote node acknowledges the data receipt, but before it actually stores the data. The data on remote node will be stored directly after sending acknowledgement. This mode is intended to reduce latency, but still provides a very good reliability. The only situation where some small amount of data could be lost is when the data is stored on primary node and sent to the secondary. Secondary node then acknowledges data receipt and primary reports success to an application. However, it may happen that the secondary goes down before the received data is really stored locally. Before secondary node returns, primary node dies entirely. When the secondary node comes back to life it becomes the new primary. Unfortunately some small amount of data which was confirmed to be stored to the application was lost. The risk of such a situation is very small. The memsync replication mode is the default.

fullsync

Mark the write operation as completed when local as well as remote write completes. This is the safest and the slowest replication mode.

async

The write operation is reported as complete right after the local write completes. This is the fastest and the most dangerous replication mode. This mode should be used when replicating to a distant node where latency is too high for other modes.

checksum ⟨algorithm⟩

Checksum algorithm should be one of the following:

none

No checksum will be calculated for the data being send over the network. This is the default setting.

crc32

CRC32 checksum will be calculated.

sha256

SHA256 checksum will be calculated.

compression ⟨algorithm⟩

Compression algorithm should be one of the following:

none

Data send over the network will not be compressed.

hole

Only blocks that contain all zeros will be compressed. This is very useful for initial synchronization where potentially many blocks are still all zeros. There should be no measurable performance overhead when this algorithm is being used. This is the default setting.

lzf

The LZF algorithm by Marc Alexander Lehmann will be used to compress the data send over the network. LZF is very fast, general purpose compression algorithm.

timeout ⟨seconds⟩

Connection timeout in seconds. The default value is 20.

exec ⟨path⟩

Execute the given program on various HAST events. Below is the list of currently implemented events and arguments the given program is executed with:

<path> role <resource> <oldrole> <newrole>

Executed on both primary and secondary nodes when resource role is changed.

<path> connect <resource>

Executed on both primary and secondary nodes when connection for the given resource between the nodes is established.

<path> disconnect <resource>

Executed on both primary and secondary nodes when connection for the given resource between the nodes is lost.

<path> syncstart <resource>

Executed on primary node when synchronization process of secondary node is started.

<path> syncdone <resource>

Executed on primary node when synchronization process of secondary node is completed successfully.

<path> syncintr <resource>

Executed on primary node when synchronization process of secondary node is interrupted, most likely due to secondary node outage or connection failure between the nodes.

<path> split-brain <resource>

Executed on both primary and secondary nodes when split-brain condition is detected.

The ⟨path⟩ argument should contain full path to executable program. If the given program exits with code different than 0, hastd will log it as an error.

The ⟨resource⟩ argument is resource name from the configuration file.

The ⟨oldrole⟩ argument is previous resource role (before the change). It can be one of: init, secondary, primary.

The ⟨newrole⟩ argument is current resource role (after the change). It can be one of: init, secondary, primary.

metaflush on | off

When set to on, flush write cache of the local provider after every metadata (activemap) update. Flushing write cache ensures that provider will not reorder writes and that metadata will be properly updated before real data is stored. If the local provider does not support flushing write cache (it returns EOPNOTSUPP on the BIO_FLUSH request), hastd will disable metaflush automatically. The default value is on.

name ⟨name⟩

GEOM provider name that will appear as /dev/hast/<name>. If name is not defined, resource name will be used as provider name.

local ⟨path⟩

Path to the local component which will be used as backend provider for the resource. This can be either GEOM provider or regular file.

remote ⟨addr⟩

Address of the remote hastd daemon. Format is the same as for the listen statement. When operating as a primary node this address will be used to connect to the secondary node. When operating as a secondary node only connections from this address will be accepted.

A special value of none can be used when the remote address is not yet known (eg. the other node is not set up yet).

source ⟨addr⟩

Local address to bind to before connecting to the remote hastd daemon. Format is the same as for the listen statement.

FILES

/etc/hast.conf

The default hastctl(8) and hastd(8) configuration file.

/var/run/hastctl

Control socket used by the hastctl(8) control utility to communicate with the hastd(8) daemon.

EXAMPLES

listen tcp://0.0.0.0

on hasta {
	listen tcp://2001:db8::1/64
}
on hastb {
	listen tcp://2001:db8::2/64
}

resource shared {
	local /dev/da0

	on hasta {
		remote tcp://10.0.0.2
	}
	on hastb {
		remote tcp://10.0.0.1
	}
}
resource tank {
	on hasta {
		local /dev/mirror/tanka
		source tcp://10.0.0.1
		remote tcp://10.0.0.2
	}
	on hastb {
		local /dev/mirror/tankb
		source tcp://10.0.0.2
		remote tcp://10.0.0.1
	}
}

SEE ALSO

AUTHORS