1 # This file is part of NIT (http://www.nitlanguage.org).
3 # Copyright 2014 Alexis Laferrière <alexis.laf@xymus.net>
5 # Licensed under the Apache License, Version 2.0 (the "License");
6 # you may not use this file except in compliance with the License.
7 # You may obtain a copy of the License at
9 # http://www.apache.org/licenses/LICENSE-2.0
11 # Unless required by applicable law or agreed to in writing, software
12 # distributed under the License is distributed on an "AS IS" BASIS,
13 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 # See the License for the specific language governing permissions and
15 # limitations under the License.
17 # Tester of Nit engines on an MPI cluster
24 # Any processor, worker or controller
26 # All data and methods within this class are shared by the controller and the
28 abstract class Processor
31 # Controller rank is always 0
32 var controller_rank
: Rank = 0.rank
34 # Where to store data for transfer between nodes
36 # Require: `buffer.length % 4 == 0`
37 var buffer
= new CIntArray(1024)
39 # Run in verbose mode, display more text
44 # OpenMPI sends a SIGTERM to all nodes upon receiving a SIGTERM or SIGINT
45 # on the first process.
46 handle_signal
(sigterm
, true)
49 # Tag of a new task packet of size `tasks_per_packet`
50 var task_tag
: Tag = 0.tag
52 # Tag to return a set of `Result` throught `buffer`
53 var result_tag
: Tag = 1.tag
55 # Tag to notify `Worker` when to quit
56 var quit_tag
: Tag = 2.tag
58 # Tag to request more work from the `Controller` by a `Worker`
59 var need_work_tag
: Tag = 4.tag
61 # Tag to notify `Controller` that the sender `Worker` is done
62 var done_tag
: Tag = 5.tag
64 # Number of tasks within each task assignation with `task_tag`
65 var tasks_per_packet
= 4
67 # Run the main logic of this node
70 # Engines targetted by this execution
71 var engines
: Array[String] is noinit
73 # All known engines, used to detect errors in `engines`
74 var all_engines
: Array[String] = ["nitg-s", "nitg-sg", "nitg-g", "nitg-e", "niti", "emscripten"]
76 # Programs to test in this execution
77 var test_programs
: Array[String] is noinit
79 # Root of the temporary directory
80 var tmp_dir
= "/dev/shm/"
83 var ccache_dir
= "/dev/shm/nit_ccache"
85 # Read command line options
88 var opt_ctx
= new OptionContext
89 var opt_engines
= new OptionString(
90 "Engines to test, separated with commas ({all_engines.join(", ")} or all)",
92 var opt_help
= new OptionBool("Print this help message", "--help", "-h")
93 var opt_verbose
= new OptionCount(
94 "Be verbose, repeat to increase verbose level (max with -vvv)",
96 var opt_cleanup
= new OptionBool(
97 "Clean up all nitester files (and do not run tests)",
100 opt_ctx
.add_option
(opt_engines
, opt_help
, opt_verbose
, opt_cleanup
)
104 if opt_help
.value
then opt_ctx
.usage_error
null
107 verbose
= opt_verbose
.value
110 if opt_cleanup
.value
then
111 assert tmp_dir
.file_exists
112 for file
in tmp_dir
.files
do if file
.has_prefix
("nit") then
113 var full_path
= tmp_dir
/ file
114 if full_path
== ccache_dir
then continue
116 assert full_path
.file_exists
118 var stat
= full_path
.file_lstat
122 full_path
.file_delete
131 var rest
= opt_ctx
.rest
132 if rest
.is_empty
then opt_ctx
.usage_error
"This tool needs at least one test_program.nit"
135 # gather and check engines
136 var engines_str
= opt_engines
.value
138 if engines_str
== null then
142 engines
= engines_str
.split
(',')
144 if engines
.has
("all") then
146 engines
= all_engines
150 # check validity of targetted engines
151 var unknown_engines
= new Array[String]
152 for engine
in engines
do if not all_engines
.has
(engine
) then unknown_engines
.add engine
154 if not unknown_engines
.is_empty
then
155 opt_ctx
.usage_error
"Unknown engines: {unknown_engines.join(", ")} (expected one or most of {all_engines.join(", ")})"
157 self.engines
= engines
160 # All tasks to be performed
161 var tasks
= new Array[Task]
163 # Gather and registar all tasks
167 for engine
in engines
do for prog
in test_programs
do
168 tasks
.add
new Task(engine
, prog
)
174 # Single controller to dispatch tasks, gather results and produce stats
178 # Id as `Int` of the next task to distribute
181 redef fun receive_signal
(signal
)
198 # Cumulated results from workers
199 var results
= new ResultSet
201 # Maintain communication with workers to distribute tasks and receiver results
204 var at_work
= new Array[Rank]
207 for r
in [1..comm_world
.size
[ do
208 var sent
= send_task_to
(r
.rank
)
212 mpi
.send_empty
(r
.rank
, quit_tag
, comm_world
)
216 var status
= new Status
217 # await results and send new tasks
218 while not at_work
.is_empty
do
221 # Double probe to avoid bug with some implementation of MPI
222 mpi
.probe
(new Rank.any
, new Tag.any
, comm_world
, status
)
223 mpi
.probe
(new Rank.any
, new Tag.any
, comm_world
, status
)
225 if status
.tag
== result_tag
then
226 # Receive results fron a worker
227 var count
= status
.count
(new DataType.int
)
228 mpi
.recv_into
(buffer
, 0, count
, status
.source
, status
.tag
, comm_world
)
230 # Parse results from C array to `Result` instances
232 # Each result is on 4 ints: task id, arg, alt and result.
234 # See the comments where the data is produced in `Worker::work_on_tasks` for more informationé
235 assert count
% 4 == 0
236 for t
in (count
/4).times
do
239 var task_id
= buffer
[tt
]
240 var arg
= buffer
[tt
+1]
241 var alt
= buffer
[tt
+2]
242 var res
= buffer
[tt
+3]
244 var result
= new Result(tasks
[task_id
], arg
, alt
)
246 if res
== 1 then result
.ok
= true
247 if res
== 2 then result
.ok_empty
= true
248 if res
== 3 then result
.no_sav
= true
249 if res
== 4 then result
.fixme
= true
250 if res
== 5 then result
.fail
= true
251 if res
== 6 then result
.soso
= true
252 if res
== 7 then result
.skip
= true
253 if res
== 0 then result
.unknown
= true
257 if verbose
> 0 and results
.length
% 25 == 0 then print_short_results
260 else if status
.tag
== need_work_tag
then
261 # A worker needs more work
262 mpi
.recv_empty
(status
.source
, status
.tag
, comm_world
)
263 var sent
= send_task_to
(status
.source
)
266 mpi
.send_empty
(status
.source
, quit_tag
, comm_world
)
268 else if status
.tag
== done_tag
then
269 # A worker is done and will quit
270 mpi
.recv_empty
(status
.source
, status
.tag
, comm_world
)
271 at_work
.remove
(status
.source
)
273 if verbose
> 1 then print
"worker {status.source} is done ({at_work.length} still at work)"
275 print
"Unexpected tag {status.tag}"
283 # Send a packet of tasks to worker at `rank`
284 fun send_task_to
(rank
: Rank): Bool
286 if next_task_id
>= tasks
.length
then return false
288 buffer
[0] = next_task_id
289 next_task_id
+= tasks_per_packet
291 mpi
.send_from
(buffer
, 0, 1, rank
, task_tag
, comm_world
)
293 if verbose
> 1 then print
"sent tasks [{buffer[0]}..{next_task_id}[ to worker {rank}"
297 # Display the accumulated results received from workers
301 print
"* {results.length} total"
302 print
"* {results.oks.length + results.ok_empties.length} oks & 0ks"
303 print
"* {results.fails.length} fails"
304 print
"* {results.no_savs.length} no savs"
305 print
"* {results.fixmes.length} fixmes"
306 print
"* {results.sosos.length} sosos"
307 print
"* {results.skips.length} skips"
308 print
"* {results.unknowns.length} unknowns (bug in tests.sh or nitester)"
311 fun print_short_results
do print
"oks & fails / total: {results.oks.length + results.ok_empties.length} " +
312 "& {results.fails.length} / {results.length}"
314 # Shutdown anormaly the running tests
317 print
"Shutting down"
318 mpi
.send_empty
(new Rank.any
, quit_tag
, comm_world
)
322 # A worker node which actually execute the tests
326 # The `Rank` of `self`
329 # Compilation directory
330 var comp_dir
= "/dev/shm/nit_compile{rank}" is lazy
332 # Output file directory
333 var out_dir
= "/dev/shm/nit_out{rank}" is lazy
335 # Directory to store the xml files produced for Jenkins
336 var xml_dir
= "~/jenkins_xml/"
338 # Output file of the `tests.sh` script
339 var tests_sh_out
= "/dev/shm/nit_local_out{rank}" is lazy
341 # Source Nit repository, must be already updated and `make` before execution
342 var nit_source_dir
= "~/nit"
344 # Compiled `Regex` to detect the argument of an execution
345 var re_arg
: Regex = "arg [0-9]+".to_re
347 # Compiled `Regex` to detect the alternative of an execution
348 var re_alt
: Regex = "_alt[0-9]+".to_re
359 # Setup the testing environment
361 # Clone the nit repository.
364 if verbose
> 0 then sys
.system
"hostname"
367 # Clean up the testing environment
369 # Delete all temporary files, except `ccache_dir`.
372 if comp_dir
.file_exists
then comp_dir
.rmdir
373 if out_dir
.file_exists
then out_dir
.rmdir
374 if tests_sh_out
.file_exists
then tests_sh_out
.file_delete
377 # Single C `int` to hold the next task id received from the `Controller`
378 var task_buffer
= new CIntArray(1)
380 # Manage communication with the `Controller` and execute dispatched `Task`s
383 var status
= new Status
387 # We double probe to prevent bug where a single probes does not receive the
389 mpi
.probe
(controller_rank
, new Tag.any
, comm_world
, status
)
390 mpi
.probe
(controller_rank
, new Tag.any
, comm_world
, status
)
392 if status
.tag
== task_tag
then
393 # Receive tasks to execute
394 mpi
.recv_into
(task_buffer
, 0, 1, status
.source
, status
.tag
, comm_world
)
395 var first_id
= task_buffer
[0]
396 for task_id
in [first_id
.. first_id
+ tasks_per_packet
] do
398 # If id is over all known tasks, stop right here
399 if task_id
>= tasks
.length
then break
400 var task
= tasks
[task_id
]
402 # Command line to execute test
403 var cmd
= "XMLDIR={xml_dir} ERRLIST={out_dir}/errlist TMPDIR={out_dir} " +
404 "CCACHE_DIR={ccache_dir} CCACHE_TEMPDIR={ccache_dir} CCACHE_BASEDIR={comp_dir} " +
405 "./tests.sh --compdir {comp_dir} --outdir {out_dir} " +
406 " --node --engine {task.engine} {task.test_program} > {tests_sh_out}"
411 # Test results were written to file, read them
412 var fstream
= new IFStream.open
(tests_sh_out
)
413 var content
= fstream
.read_all
416 # Parse result and prepare them for sending
418 # The structure is composed of 4 ints for each result.
422 # 4. test result as int
423 var c
= results_count
424 for line
in content
.split
('\n') do if not line
.is_empty
then
429 var arg_match
= line
.search
(re_arg
)
431 if arg_match
!= null then arg
= arg_match
.to_s
.substring_from
(4).to_i
434 var alt_match
= line
.search
(re_alt
)
436 if alt_match
!= null then alt
= alt_match
.to_s
.substring_from
(4).to_i
440 if line
.has
("[ok]") then res
= 1
441 if line
.has
("[0k]") then res
= 2
442 if line
.has
("[=== no sav ===]") then res
= 3
443 if line
.has
("[fixme]") then res
= 4
444 if line
.has
("[======= fail") then res
= 5
445 if line
.has
("[======= soso") then res
= 6
446 if line
.has
("[skip]") then res
= 7
450 if verbose
> 1 then print
"Unknown result: '{line}'"
456 if verbose
> 2 then print
"tests.sh output line: {line}"
458 # If result buffer is full, send to `Controller`
459 if c
*4 == buffer
.length
then
465 self.results_count
= c
468 mpi
.send_empty
(controller_rank
, need_work_tag
, comm_world
)
469 else if status
.tag
== quit_tag
then
470 # Notification from the `Controller` to quit
471 mpi
.recv_empty
(status
.source
, status
.tag
, comm_world
)
473 # Send remaining results
476 # Notify `Controller` that `self` is done and will quit
477 mpi
.send_empty
(controller_rank
, done_tag
, comm_world
)
480 print
"Unexpected tag {status.tag}"
487 # Total results listed in `buffer` and ready to send
488 var results_count
= 0
490 # Send all results in `buffer` to the `Controller`
493 if results_count
> 0 then
494 if verbose
> 1 then print
"sending {results_count} results"
495 mpi
.send_from
(buffer
, 0, results_count
*4, controller_rank
, result_tag
, comm_world
)
500 redef fun receive_signal
(signal
)
508 # A single test task, on a `test_program` with an `engine`
510 # Note that a task may involve more than one program to test considering the
511 # alts and args for the `test_program`.
513 # Engine to test executing `test_program`
516 # Program to execute with `engine`
517 var test_program
: String
519 redef fun to_s
do return "{engine} {test_program}"
524 # There may be more than one result per `Task`.
526 # `Task` associated to `self`
529 # Argument index of the execution resulting in `self`
532 # Alternative index of the execution resulting in `self`
535 # Is `self` result an _ok_?
538 # Is `self` result an _0k_?
541 # Is `self` result a _no sav_?
544 # Is `self` result a _fixme_?
547 # Is `self` result a _fail_?
550 # Is `self` result a _soso_?
553 # Is `self` skipped test?
556 # Is `self` an unknown result, probably an error
562 if no_sav
then err
= "no sav"
563 if ok
then err
= "ok"
564 if ok_empty
then err
= "0k"
565 if fixme
then err
= "fixme"
566 if fail
then err
= "fail"
568 return "{task} arg{arg} alt{alt} => {err}"
572 # A global and sorted collection of `Result`
574 super HashSet[Result]
576 var no_savs
= new HashSet[Result]
577 var oks
= new HashSet[Result]
578 var ok_empties
= new HashSet[Result]
579 var fixmes
= new HashSet[Result]
580 var fails
= new HashSet[Result]
581 var sosos
= new HashSet[Result]
582 var skips
= new HashSet[Result]
583 var unknowns
= new HashSet[Result]
586 var per_engines
= new HashMap[String, Result]
588 redef fun add
(result
)
590 if result
.no_sav
then no_savs
.add result
591 if result
.ok
then oks
.add result
592 if result
.ok_empty
then ok_empties
.add result
593 if result
.fixme
then fixmes
.add result
594 if result
.fail
then fails
.add result
595 if result
.soso
then sosos
.add result
596 if result
.skip
then skips
.add result
597 if result
.unknown
then unknowns
.add result
602 redef fun remove
(r
) do abort
604 redef fun clear
do abort
607 redef class OptionContext
609 # Print usage with a possible error `message`
610 private fun usage_error
(message
: nullable String)
613 if message
!= null then
614 print
"Error: {message}"
618 if comm_world
.rank
== 0 then
619 print
"Usage: mpirun nitester [Options] test_program.nit [other_test.nit [...]]"
628 # On `Worker` nodes, prefix all prints with `rank/comm_world.size`
629 redef fun print
(msg
: Object)
631 if comm_world
.rank
!= 0.rank
then
632 super "{comm_world.rank}/{comm_world.size}: {msg}"
636 # Running MPI instance
637 fun mpi
: MPI do return once
new MPI
643 var rank
= comm_world
.rank
645 var processor
: Processor
646 if rank
== 0.rank
then
647 # If rank == 0, this is the `Controller`
648 processor
= new Controller
651 processor
= new Worker(rank
)