Skip to content

Latest commit

 

History

History
108 lines (85 loc) · 3.75 KB

0094-lfarm.org

File metadata and controls

108 lines (85 loc) · 3.75 KB
Raw

lfarm

This library is similar to Python’s multiprocessing which is also able to run processes on different hosts.

Use lfarm when you need to run jobs on the cluster.

The cool lfarm’s feature is the ability to run jobs on workers, running on different Common Lisp implementation.

To demonstrate how it works, I started a LispWorks Personal edition on my laptop and SBCL with ClozureCL in the Docker containers.

(ql:quickload :lfarm-server)

;; This how I started it on LispWorks
(lfarm-server:start-server "127.0.0.1" 10003)

;; inside the Docker:
(lfarm-server:start-server "0.0.0.0" 10000)

After that, I created a workers pool in my SBCL REPL and called broadcast-task to run a function on every worker:

POFTHEDAY> (setf lfarm:*kernel* (lfarm:make-kernel '(("127.0.0.1" 10001)
                                                     ("127.0.0.1" 10002)
                                                     ("127.0.0.1" 10003))))
#<LFARM-CLIENT.KERNEL:KERNEL :WORKER-COUNT 3 {1004E7C7C3}>

POFTHEDAY> (lfarm:broadcast-task
            (lambda ()
              (list :lisp (lisp-implementation-type)
                    :lisp-version (lisp-implementation-version)
                    :system (string-trim
                             (list #\Newline #\Space)
                             (with-output-to-string (s)
                               (uiop:run-program "uname -s -m -n"
                                                 :output s))))))
#((:LISP "LispWorks Personal Edition"
   :LISP-VERSION "7.1.2"
   :SYSTEM "Darwin lispbook x86_64")
  (:LISP "SBCL"
   :LISP-VERSION "1.5.6"
   :SYSTEM "Linux 5d4cc1b96f7e x86_64")
  (:LISP "Clozure Common Lisp"
   :LISP-VERSION "Version 1.11.5/v1.11.5  (LinuxX8664)"
   :SYSTEM "Linux d2c83144fce3 x86_64"))

Another great feature is API compatibility with lparallel, reviewed yesterday.

I took our yesterday code and modified it to be able to distinguish a machine processed a task:

POFTHEDAY> (lfarm:pmapcar
            (lambda (idx)
              (loop for iteration from 1 to 10
                    do (format t "~A task is processing ~A iteration~%"
                               idx iteration)
                       (sleep (random 1.0))
                    finally (return (format nil "~A result processed by ~A"
                                            idx
                                            (lisp-implementation-type)))))
              (alexandria:iota 10))
("0 result processed by Clozure Common Lisp"
 "1 result processed by Clozure Common Lisp"
 "2 result processed by Clozure Common Lisp"
 "3 result processed by Clozure Common Lisp"
 "4 result processed by SBCL"
 "5 result processed by SBCL"
 "6 result processed by SBCL"
 "7 result processed by LispWorks Personal Edition"
 "8 result processed by LispWorks Personal Edition"
 "9 result processed by LispWorks Personal Edition")

The bad news is that lfarm is not suite well to the cloud environment, where workers can disappear at any time and become available later.

Some time ago, I tried to use lfarm in Ultralisp, to run some jobs in separate containers, but found that lfarm does not like when worker process disconnects.

For example, if some of the workers aren’t running, make-kernel will hang forever. Or if some of the workers will go offline, then broadcast-task also will hang.

Tomorrow I’ll tell you about another solution which helped me to solve this problem.