In order to speed up command kart command invocations a helper mode can be enabled on Linux and MacOS
This mode will start a background process which stays running and forks a new process to perform the command so as to not include python import time overhead.
The helper command
kart helper starts the background process and supports the
--socket is the
name of a UNIX socket to be used to communcate between the client
--socket defaults to $HOME/.kart.socket and
setting a seperate socket name is not currently supported in the
--timout specifies how long the helper will wait
for a command from the client before shutting down.
In normal operation
kart helper is started by the client
process which is part of the standard
kart startup process
but the client will use a previously started helper if it can
connect on the named UNIX socket.
Through an environment variable,
KART_HELPER_LOG, it is possible
to enable logging of all commands performed by the helper. This will
log the PID and working directory of the client along with the command
and args that were run to the filename in
Helper mode is enabled by specifying setting and environment variable,
KART_USE_HELPER. 0, not set and any value with leading spaces are
all treated as false, any other value is true.
The client executable is
kart_cli_helper but at install is linked as
kart so use is transparent. The main PyInstaller build executable is
kart_cli on supported platforms.
When helper mode is enabled the client,
kart, will try to connect
to a UNIX socket, if that is not possible it will try to start the
helper command of kart and wait until the socket is available.
During starting the helper command the client will double fork, close file descriptors and execute the helper command so that it is fully detached in the background to continue running after this invocation of the client. The helper command will start, open a socket and listen along with importing any expensive python libraries so as to ensure no imports are done when a fork is performed to run a command.
Once the client can connect to the socket it creates a semaphore to receive the exit code of kart and sends a JSON dictionary of the local environment at calling time, command arguments to run, semaphore ID and PID of the client process. This is sent to the helper along with stdin, stdout, stderr file descriptors and a file descriptor of the current working directory. The client then needs to wait as long as the longest kart command might take, currently it waits for 1 day.
On receiving a request from the client the helper mode forks a child,
sets up the environment and working directory as per the client process,
opens the stdin, stdout, stderr file descriptors from the client and
assigns them to sys.stdin, sys.stdout, sys.stderr so that output will
go directly to the calling process standard streams without copying
then runs the command through
cli() as usual.
Once the command completes the helper writes the exit code to the semphore and sends a SIGALRM to the client. The client handles the SIGALRM by reading the semphore to set the exit code and exiting.