View Source heart (kernel v10.0.1)
Heartbeat monitoring of an Erlang runtime system.
This modules contains the interface to the heart
process. heart
sends
periodic heartbeats to an external port program, which is also named heart
.
The purpose of the heart
port program is to check that the Erlang runtime
system it is supervising is still running. If the port program has not received
any heartbeats within HEART_BEAT_TIMEOUT
seconds (defaults to 60 seconds), the
system can be rebooted.
An Erlang runtime system to be monitored by a heart program is to be started
with command-line flag -heart
(see also erl(1)
). The
heart
process is then started automatically:
% erl -heart ...
If the system is to be rebooted because of missing heartbeats, or a terminated
Erlang runtime system, environment variable HEART_COMMAND
must be set before
the system is started. If this variable is not set, a warning text is printed
but the system does not reboot.
To reboot on Windows, HEART_COMMAND
can be set to heart -shutdown
(included
in the Erlang delivery) or to any other suitable program that can activate a
reboot.
The environment variable HEART_BEAT_TIMEOUT
can be used to configure the heart
time-outs; it can be set in the operating system shell before Erlang is started
or be specified at the command line:
% erl -heart -env HEART_BEAT_TIMEOUT 30 ...
The value (in seconds) must be in the range 10 < X <= 65535
.
When running on OSs lacking support for monotonic time, heart
is susceptible
to system clock adjustments of more than HEART_BEAT_TIMEOUT
seconds. When this
happens, heart
times out and tries to reboot the system. This can occur, for
example, if the system clock is adjusted automatically by use of the Network
Time Protocol (NTP).
If a crash occurs, an erl_crash.dump
is not written unless environment
variable ERL_CRASH_DUMP_SECONDS
is set:
% erl -heart -env ERL_CRASH_DUMP_SECONDS 10 ...
If a regular core dump is wanted, let heart
know by setting the kill signal to
abort using environment variable HEART_KILL_SIGNAL=SIGABRT
. If unset, or not
set to SIGABRT
, the default behavior is a kill signal using SIGKILL
:
% erl -heart -env HEART_KILL_SIGNAL SIGABRT ...
If heart should not kill the Erlang runtime system, this can be indicated
using the environment variable HEART_NO_KILL=TRUE
. This can be useful if the
command executed by heart takes care of this, for example as part of a specific
cleanup sequence. If unset, or not set to TRUE
, the default behaviour will be
to kill as described above.
% erl -heart -env HEART_NO_KILL 1 ...
Furthermore, ERL_CRASH_DUMP_SECONDS
has the following behavior on heart
:
ERL_CRASH_DUMP_SECONDS=0
- Suppresses the writing of a crash dump file entirely, thus rebooting the runtime system immediately. This is the same as not setting the environment variable.ERL_CRASH_DUMP_SECONDS=-1
- Setting the environment variable to a negative value does not reboot the runtime system until the crash dump file is completely written.ERL_CRASH_DUMP_SECONDS=S
-heart
waits forS
seconds to let the crash dump file be written. AfterS
seconds,heart
reboots the runtime system, whether the crash dump file is written or not.
In the following descriptions, all functions fail with reason badarg
if
heart
is not started.
Summary
Functions
Removes the validation callback call before heartbeats.
Clears the temporary boot command. If the system terminates, the normal
HEART_COMMAND
is used to reboot.
Get the validation callback. If the callback is cleared, none
will be
returned.
Gets the temporary reboot command. If the command is cleared, the empty string is returned.
Returns {ok, Options}
where Options
is a list of current options enabled for
heart. If the callback is cleared, none
will be returned.
This validation callback will be executed before any heartbeat is sent to the
port program. For the validation to succeed it needs to return with the value
ok
.
Sets a temporary reboot command. This command is used if a HEART_COMMAND
other
than the one specified with the environment variable is to be used to reboot the
system. The new Erlang runtime system uses (if it misbehaves) environment
variable HEART_COMMAND
to reboot.
Valid options set_options
are
Types
-type heart_option() :: check_schedulers.
Functions
-spec clear_callback() -> ok.
Removes the validation callback call before heartbeats.
-spec clear_cmd() -> ok.
Clears the temporary boot command. If the system terminates, the normal
HEART_COMMAND
is used to reboot.
Get the validation callback. If the callback is cleared, none
will be
returned.
-spec get_cmd() -> {ok, Cmd} when Cmd :: string().
Gets the temporary reboot command. If the command is cleared, the empty string is returned.
-spec get_options() -> {ok, Options} | none when Options :: [atom()].
Returns {ok, Options}
where Options
is a list of current options enabled for
heart. If the callback is cleared, none
will be returned.
-spec set_callback(Module, Function) -> ok | {error, {bad_callback, {Module, Function}}} when Module :: atom(), Function :: atom().
This validation callback will be executed before any heartbeat is sent to the
port program. For the validation to succeed it needs to return with the value
ok
.
An exception within the callback will be treated as a validation failure.
The callback will be removed if the system reboots.
-spec set_cmd(Cmd) -> ok | {error, {bad_cmd, Cmd}} when Cmd :: string().
Sets a temporary reboot command. This command is used if a HEART_COMMAND
other
than the one specified with the environment variable is to be used to reboot the
system. The new Erlang runtime system uses (if it misbehaves) environment
variable HEART_COMMAND
to reboot.
Limitations: Command string Cmd
is sent to the heart
program as an ISO
Latin-1 or UTF-8 encoded binary, depending on the filename encoding mode of the
emulator (see file:native_name_encoding/0
). The size of the encoded binary
must be less than 2047 bytes.
-spec set_options(Options) -> ok | {error, {bad_options, Options}} when Options :: [heart_option()].
Valid options set_options
are:
check_schedulers
- If enabled, a signal will be sent to each scheduler to check its responsiveness. The system check occurs before any heartbeat sent to the port program. If any scheduler is not responsive enough the heart program will not receive its heartbeat and thus eventually terminate the node.
Returns with the value ok
if the options are valid.