controller.py 58.1 KB
Newer Older
1
2
3
4
#!/usr/bin/python3


import asyncio
5
import collections
6
from   concurrent.futures   import ThreadPoolExecutor
7
import contextlib
8
import datetime
9
import enum
BAIRE Anthony's avatar
BAIRE Anthony committed
10
import itertools
BAIRE Anthony's avatar
BAIRE Anthony committed
11
import json
12
13
14
import logging
import re
import os
15
import shlex
16
17
18
import signal
import socket
import sys
19
20
21
import time
import threading
import traceback
22
23
24

import docker
import MySQLdb
BAIRE Anthony's avatar
BAIRE Anthony committed
25
26
from   sqlalchemy import desc
import sqlalchemy.orm.scoping
27
import yaml
28

29
import config_reader
30
from database import *
31
from shared_swarm import SharedSwarmClient, ShuttingDown
32

33
34
HOST_PATH="/vol/host/"

35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
##################################################
# default number of concurrent tasks

# pushing images from the sandbox to the registry
NB_PUSH_SANDBOX = 2

# pulling images from the registry to the sandbox
NB_PULL_SANDBOX = 2

# pulling images from the registry to the swarm
NB_PULL_SWARM   = 2

# sandbox actions (start, stop, commit, ...)
NB_SANDBOX_TASKS = 4

# jobs
NB_JOB_TASKS     = 4


# default thread pool (used by most of the tasks)
default_executor = ThreadPoolExecutor(10)

##################################################

BAIRE Anthony's avatar
BAIRE Anthony committed
59
log = logging.getLogger("controller")
60

61
62
63
64
65
assert MySQLdb.threadsafety >= 1


class Error(Exception):
    pass
66

BAIRE Anthony's avatar
BAIRE Anthony committed
67
68
69
70
71
72
73
74
75
76
77
78
79
def docker_check_error(func, *k, **kw):
    """Wrapper for docker-py methods that produce a stream

    Methods producing a stram (eg: push, build) do not report all errors
    by raising exceptions. Some errors are reported later in the stream.

    This function parses the stream and raise Error() if needed.
    """
    for elem in func(*k, stream=True, **kw):
        js = json.loads(elem.decode())
        if "error" in js:
            raise Error("push error: " + js["error"])

BAIRE Anthony's avatar
BAIRE Anthony committed
80
81
82
83
84
85
86
87
88
@contextlib.contextmanager
def docker_warning(msg, *k, ignore=None):
    """Catch docker errors and issue a warning instead"""
    try:
        yield
    except docker.errors.APIError as e:
        if ignore is None or not isinstance(e, ignore):
            k += e,
            log.warning(msg + " (%s)", *k)
BAIRE Anthony's avatar
BAIRE Anthony committed
89

BAIRE Anthony's avatar
BAIRE Anthony committed
90
91
@contextlib.contextmanager
def report_error(fmt, *k):
BAIRE Anthony's avatar
doc    
BAIRE Anthony committed
92
93
94
95
96
97
98
    """Context manager for logging exceptions

    This function logs exceptions (when leaving the context) with log.error()
    (if the exception inherit from Error) or log.exception() otherwise.

    The log message is prepended with the string generated by: fmt % k
    """
BAIRE Anthony's avatar
BAIRE Anthony committed
99
100
101
102
103
104
105
106
    try:
        yield
    except Exception as e:
        msg = fmt % k
        log_func = log.error if isinstance(e, Error) else log.exception
        log_func("%s (%s)", msg,
                traceback.format_exception_only(type(e), e)[-1].strip())
        raise
107

BAIRE Anthony's avatar
BAIRE Anthony committed
108
109
110
111
112
113
114
115
116
117
118
119
120
def disable_future_warning(fut):
    """Add a dummy callback to a future to prevent asyncio warnings

    Return: the future

    The asynicio module log a warning message when the result of a future is
    not used. This function installs a dummy callback to the future so as to
    avoid this warning. This is useful for futures whose result *may* be
    ignored by the application.
    """
    fut.add_done_callback(lambda f: f.exception())
    return fut

121

122
123
124
125
126
127
128
129
130
def auto_create_task(func):
    """Decorator for forcing the creation of a task when a coroutine function is called

    Return a wrappers that calls the function, create the task on the fly and
    return it. Also it installs a no-op callback to avoid warnings in case the
    result is not used.
    """
    assert asyncio.iscoroutinefunction(func)

BAIRE Anthony's avatar
BAIRE Anthony committed
131
132
    return (lambda *k, **kw:
        disable_future_warning(asyncio.async(func(*k, **kw))))
BAIRE Anthony's avatar
BAIRE Anthony committed
133

134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
def cascade_future(src, dst):
    """propagate the result of a future to another future
    
    This function installs a callback to the future `src`, that propagates its
    result to the future `dst`.
    """
    def callback(fut):
        ex = fut.exception()
        if ex is None:
            dst.set_result(fut.result())
        else:
            dst.set_exception(ex)

    if src.done():
        callback(src)
    else:
        src.add_done_callback(callback)

BAIRE Anthony's avatar
todo    
BAIRE Anthony committed
152

153
class Manager:
BAIRE Anthony's avatar
BAIRE Anthony committed
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
    """A class for scheduling asynchronous jobs on a collection of keys

    (the one-line summary is not very helpful, sorry)
    

    Job submission

    The job is scheduled on key KEY by calling .process(KEY), the function
    returns an asyncio.Future object (which will receive the result)

    If .process(KEY) is called while a job is already running for KEY, then the
    manager arranges the job to be run a second time in a row for the same key


    Job implementation

    This is an abstract class. The job shall be implemented as the coroutine
    named ._process() in the inherited class.

    The manager guarantees that ._process() cannot be called multiple times
    concurrently for the same key

    In case a shutdown is requested (see .shutdown()):
     - all jobs that are not yet started are cancelled
     - running jobs continue until they return or until they try to acquire the
       internal semaphore (which raises ShuttingDown())


    Concurrency

    The job tasks are all started immediately (whatever the number of existing
    jobs)

    It is possible to limit the concurrency to a maximum number of parallel
    tasks. Manager provides an internal semaphore (the number of tokens is set
    in .__init__())

    The semaphore can be locked in two ways:
     - by locking the manager:

            with (yield from self): ... concurrency limited to `nb_tokens`
            tasks ...

     - by calling .run_in_executor (this is for running non-async code):

            yield from self.run_in_executor(func, args...)

    Shutdown

    When .shutdown() is called the Manager ensures that all jobs are properly
    terminated
     - it cancels all jobs that are not yet started
     - it prevents starting new jobs
     - it lets running pending jobs, but interrupts them if they when ther try
       to acquire the internal semaphore

    All cancel/interrupted tasks have their future raise ShuttingDown().

    .shutdown() returns after all pending jobs are terminated.


    Thread safety

    - Manager is *not* thread safe, all public methods must be called from the
      same thread
    """
220

221
    class _Handle:
BAIRE Anthony's avatar
BAIRE Anthony committed
222
        __slots__ = "key", "cur", "nxt", "rescheduled"
223

224
    def __init__(self, nb_tokens=1, *, executor = default_executor, interruptible = False):
BAIRE Anthony's avatar
BAIRE Anthony committed
225
        # {key: _TaskHandler}
226
        self._handles = {}
227
        self._semaphore = asyncio.Semaphore(nb_tokens)
228
        self._shutdown = asyncio.Future()
229
        self._executor = executor
230
        self._interruptible = interruptible
231
232
233
234
235
236
237
238
239
240

    def _create_task(self, hnd):
        assert hnd.nxt is None
        def reset():
            assert not hnd.cur.done()
            nxt = hnd.nxt
            if nxt is not None:
                cascade_future(hnd.cur, nxt)
                hnd.nxt = None

BAIRE Anthony's avatar
BAIRE Anthony committed
241
242
243
244
245
246
            # recreate the 'rescheduled' future and return it
            hnd.rescheduled = disable_future_warning(asyncio.Future())
            return hnd.rescheduled

        hnd.rescheduled = disable_future_warning(asyncio.Future())
        hnd.cur = asyncio.async(self._process(hnd.key, reset, hnd.rescheduled))
247
        hnd.cur.add_done_callback(lambda fut: self._done(hnd))
BAIRE Anthony's avatar
BAIRE Anthony committed
248
        log.debug("task scheduled %r %r", self, hnd.key)
249
        return hnd.cur
250

BAIRE Anthony's avatar
BAIRE Anthony committed
251
252
253
254
255
    def process(self, key):
        """Schedule the job to be run on key `key`

        returns an asyncio.Future (that will provide the result of ._process())
        """
256
        if self._shutdown.done():
257
258
            return self._shutdown

BAIRE Anthony's avatar
BAIRE Anthony committed
259
        hnd = self._handles.get(key)
260
261
        if hnd is None:
            # create handle
BAIRE Anthony's avatar
BAIRE Anthony committed
262
263
            self._handles[key] = hnd = self._Handle()
            hnd.key = key
264
265
            hnd.cur = None
            hnd.nxt = None
BAIRE Anthony's avatar
BAIRE Anthony committed
266
            hnd.rescheduled = None
267
268
269
270
271
272
273
274

        if hnd.cur is None:
            # create task
            return self._create_task(hnd)
        else:
            # reschedule task
            if hnd.nxt is None:
                hnd.nxt = asyncio.Future()
BAIRE Anthony's avatar
BAIRE Anthony committed
275
276
            if not hnd.rescheduled.done():
                hnd.rescheduled.set_result(None)
277
278
279
280
            return hnd.nxt

    def _done(self, hnd):

BAIRE Anthony's avatar
BAIRE Anthony committed
281
        assert hnd is self._handles.get(hnd.key)
282
283
284
285
        assert hnd.cur.done()

        try:
            hnd.cur.result()
286
287
        except ShuttingDown:
            pass
288
        except Exception:
BAIRE Anthony's avatar
BAIRE Anthony committed
289
            log.exception("task %r %r unhandled exception", self, hnd.key)
290
291
292
293

        nxt     = hnd.nxt
        hnd.cur = hnd.nxt = None
        if nxt is None:
BAIRE Anthony's avatar
BAIRE Anthony committed
294
            del self._handles[hnd.key]
295
        else:
296
            cascade_future(self._create_task(hnd), nxt)
297

298

299
300
    @asyncio.coroutine
    def __iter__(self):
BAIRE Anthony's avatar
BAIRE Anthony committed
301
302
303
304
305
306
307
308
309
310
        """Coroutine for locking the internal semaphore
        
        Usage:
            with (yield from manager):
              ...

        Warning:
            after a shutdown is initiated, this function will always raise
            ShuttingDown()
        """
311
        ctx = yield from iter(self._semaphore)
312
        if self._shutdown.done():
313
314
315
316
317
318
            with ctx:
                raise ShuttingDown()
        return ctx


    @asyncio.coroutine
319
    def run_in_executor(self, *k, lock=True):
BAIRE Anthony's avatar
BAIRE Anthony committed
320
321
322
323
324
        """Run a function in a separate thread (with limited concurrency)

        This function locks the internal semaphore and runs the provided
        functionc call in a separate thread (using the executor)
        """
325
326
327
328
329
330
331
332

        def run():
            coro = asyncio.get_event_loop().run_in_executor(self._executor, *k)
            if self._interruptible:
                return next(asyncio.as_completed((coro, self._shutdown)))
            else:
                return coro
        
333
334
        if lock:
            with (yield from self):
335
                return (yield from run())
336
        else:
337
            return (yield from run())
338
339


BAIRE Anthony's avatar
BAIRE Anthony committed
340
    @asyncio.coroutine
BAIRE Anthony's avatar
BAIRE Anthony committed
341
    def _process(self, key, reset, rescheduled):
BAIRE Anthony's avatar
doc    
BAIRE Anthony committed
342
343
344
345
346
347
348
        """Actual implementation of the job (to be reimplemented in inherited classes)

        The Manager class guarantees that this function cannot be called
        multiple times concurrently on the same key (in case the same key is
        submitted multiple times, they Manager will call this function a second
        time after it has terminated).

BAIRE Anthony's avatar
BAIRE Anthony committed
349
350
351
352
        `rescheduled` is a future whose result is set when the job is being
        rescheduled (if process(key) is called before _process(key ...)
        terminates.

BAIRE Anthony's avatar
doc    
BAIRE Anthony committed
353
354
        `reset` is a function that may be called to reset the 'dirty' state of
        this key (this is to avoid calling ._process() a second time if not
BAIRE Anthony's avatar
BAIRE Anthony committed
355
        necessary) in that case, a new `rescheduled` future is returned.
BAIRE Anthony's avatar
doc    
BAIRE Anthony committed
356
        """
357
358
        raise NotImplementedError()

359
    @asyncio.coroutine
360
    def shutdown(self):
BAIRE Anthony's avatar
BAIRE Anthony committed
361
362
363
364
        """Initiate a graceful shutdown

        This coroutine terminates once all tasks are properly terminated.
        """
365
366
        exc = ShuttingDown()
        self._shutdown.set_exception(exc)
367
        self._shutdown.exception()  # to avoid asyncio warnings
368
369
370
        # cancel all 'next' tasks
        for hnd in self._handles.values():
            if hnd.nxt is not None:
371
                hnd.nxt.set_exception(exc)
372
373
                hnd.nxt = None

374
375
376
377
        if not self._interruptible:
            yield from asyncio.gather(
                    *(h.cur for h in self._handles.values() if h.cur is not None),
                    return_exceptions=True)
BAIRE Anthony's avatar
BAIRE Anthony committed
378

379
380

class SandboxManager(Manager):
BAIRE Anthony's avatar
BAIRE Anthony committed
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
    """Manager for sandbox operation

    This manager handles all sandbox operations (start, stop, commit).

    Operations are requested asynchronously in the database:
     - Webapp.sandbox_state=starting (for starting a sandbox)
     - Webapp.sandbox_state=stopping (for stopping a sandbox)
     - WebappVersion with state=sandbox (for committing a new image)

    All state changes on Webapp.sandbox_state are atomic (e.g: if the user
    requests a stop while the sandbox is starting then the manager finishes
    with starting the sandbox but does not update the state, and it runs
    immediately again to stop the sandbox)

    Whatever is the value of Webapp.sandbox_state, the manager first examines
    commit requests and makes the commit if requested.

    If the container already exists before starting the webapp (not possible in
    normal operations), then a recovery image is committed first.

    When a commit is successful. The image manager is notified for pushing the
    image to the registry and (if it is not a recovery image) to pull it to the
    swarm (because this image will very likely be used soon).

    State changes:

     - sandbox start:
            starting->running     (normal case)
            starting->start_error (error case)

     - sandbox stop:
            stopping->idle        (normal case)
            stopping->stop_error  (error case)

     - image commit:
            sandbox->committed  (normal case)
            sandbox->error      (error case)
            (none)->committed   (recovery version) 
    """
420

421
    def __init__(self, ctrl, nb_threads = NB_SANDBOX_TASKS):
422
        super().__init__(nb_threads)
423
424
425
        self.ctrl = ctrl


BAIRE Anthony's avatar
BAIRE Anthony committed
426
427
    def inspect_sandbox(self, webapp):
        try:
BAIRE Anthony's avatar
BAIRE Anthony committed
428
429
            return self.ctrl.sandbox.inspect_container(
                    self.ctrl.gen_sandbox_name(webapp))
BAIRE Anthony's avatar
BAIRE Anthony committed
430
431
432
        except docker.errors.NotFound:
            return None

BAIRE Anthony's avatar
BAIRE Anthony committed
433
434
435
436
437
438

    @staticmethod
    def filter_commit_version(query, webapp_id):
        """Narrow a WebappVersion query to select the candidate versions to be committed"""
        return (query.
                filter_by(webapp_id=webapp_id,
BAIRE Anthony's avatar
BAIRE Anthony committed
439
                         state = int(VersionState.SANDBOX))
BAIRE Anthony's avatar
BAIRE Anthony committed
440
441
442
                    )

    def _start(self, webapp, version):
443
444
445
446
447
448
        """Start a webapp sandbox

        (to be executed in a thread pool)
        """

        ctrl = self.ctrl
BAIRE Anthony's avatar
BAIRE Anthony committed
449
        ses  = ctrl.session
450
451

        # prepare sandbox parameters
BAIRE Anthony's avatar
BAIRE Anthony committed
452
453
454

        # docker image
        if version is None:
455
            image = "%s:%s" % (ctrl.gen_factory_name(webapp.docker_os),
BAIRE Anthony's avatar
BAIRE Anthony committed
456
                    webapp.docker_os.version)
457
        else:
BAIRE Anthony's avatar
BAIRE Anthony committed
458
            image = "%s:%s" % (webapp.image_name, version.number)
BAIRE Anthony's avatar
BAIRE Anthony committed
459
460

        log.debug("sandbox %r: using image %r", webapp.docker_name, image)
461
462
463
464
465
466
467
468
469
470
471
472

        # safety checks
        # (because docker_name is used it the paths of the external volumes
        if ("/" in webapp.docker_name) or (webapp.docker_name in ("", ".", "..")):
            raise Error("malformatted docker_name")

        uid = webapp.id + 2000
        if uid < 2000:
            # just for safety
            raise Error("bad webapp id")

        # remove stale container (if any)
473
474
        if self.inspect_sandbox(webapp) is not None:
            self._stop(webapp)
475

BAIRE Anthony's avatar
BAIRE Anthony committed
476
        container = webapp.sandbox_name
477
478
        try:
            # prepare the sandbox
479
480
481
482
            # (create ssh keys)
            ctrl.check_host_path("isdir", ctrl.toolbox_path)
            ctrl.check_host_path("isdir", ctrl.sandbox_path)

BAIRE Anthony's avatar
BAIRE Anthony committed
483
            ctrl.sandbox.create_container("busybox:latest", name=container,
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
                    command = ["/bin/sh", "-c", """
set -ex

export PATH="$PATH:/.toolbox/bin"

# clean sandbox dir
rm -rf {sbx}

# create dirs
for dir in {sbx} {etc} {run}
do
    mkdir -p            ${{dir}}
    chown {uid}:65534   ${{dir}}
    chmod 0700          ${{dir}}
done

# xauth file
touch               {run}/XAuthority
chown {uid}:65534   {run}/XAuthority
chmod 0600          {run}/XAuthority

# generate ssh keys
(for type in ecdsa ed25519 rsa
do
    key={etc}/ssh_host_${{type}}_key
    [ -f $key ] || ssh-keygen -N '' -f $key -t $type >&2
    
    echo -n '{hostname}. ' | cat - ${{key}}.pub
done) > {etc}/ssh_known_hosts

# known_host file for allgo-shell
ssh-keygen -H -f {etc}/ssh_known_hosts
chmod 0644       {etc}/ssh_known_hosts
rm -f            {etc}/ssh_known_hosts.old

# authentication key for allgo-shell
rm -f               {etc}/identity
ssh-keygen -N '' -f {etc}/identity
chown {uid}:65534   {etc}/identity

# forced shell for the sshd config 
cat > {etc}/shell <<EOF
#!/bin/sh

export PATH="\$PATH:/.toolbox/bin"

uid=\`id -u\`
shell="\`getent passwd \$uid 2>/dev/null| cut -d : -f 7\`"
if [ -z "\$shell" ] ; then
    shell=/bin/sh
fi

if [ -n "\$SSH_ORIGINAL_COMMAND" ] ; then
    exec "\$shell" -c "\$SSH_ORIGINAL_COMMAND"
else
    exec "\$shell"
fi

EOF
chmod 755 {etc}/shell

# sshd config
cat > {etc}/sshd_config <<EOF
Port 22
Protocol 2

# turned off because it requires creating a 'sshd' user inside the sandbox
UsePrivilegeSeparation no

StrictModes no

ForceCommand /.sandbox/etc/ssh/shell

PermitRootLogin without-password

PubkeyAuthentication yes
AuthorizedKeysFile  /.sandbox/etc/ssh/identity.pub .ssh/authorized_keys .ssh/authorized_keys2

ChallengeResponseAuthentication no
PasswordAuthentication          no

X11Forwarding yes
X11DisplayOffset 10
PrintMotd no
PrintLastLog no
TCPKeepAlive yes

# Allow client to pass locale environment variables
AcceptEnv LANG LC_*

Subsystem sftp internal-sftp

UsePAM no
EOF
                    """.format(uid=uid,
                        hostname = "%s-sandbox-%s" % (ctrl.env, webapp.docker_name),
                        sbx = "/mnt/%s"         % webapp.docker_name,
                        etc = "/mnt/%s/etc/ssh" % webapp.docker_name,
                        run = "/mnt/%s/run"     % webapp.docker_name,
                        )],
584
                    host_config = ctrl.sandbox.create_host_config(
585
586
587
588
                        binds   = {
                            ctrl.sandbox_path: {"bind": "/mnt"},
                            ctrl.toolbox_path: {"bind": "/.toolbox", "mode": "ro"},
                    }))
589
590
            ctrl.sandbox.start(container)
            if ctrl.sandbox.wait(container):
591
592
                log.debug("sandbox %s output:\n%s", webapp.docker_name,
                        ctrl.sandbox.logs(container).decode(errors="replace"))
593
594
595
596
                raise Error("sandbox preparation failed")
            ctrl.sandbox.remove_container(container)

            # create and start the sandbox
597
598
599
600
601

            etc_dir = os.path.join(ctrl.sandbox_path, webapp.docker_name, "etc")
            run_dir = os.path.join(ctrl.sandbox_path, webapp.docker_name, "run")
            ctrl.check_host_path("isdir", etc_dir)
            ctrl.check_host_path("isdir", run_dir)
602
603
604
605

            if version is None and webapp.entrypoint:
                # prepend instructions to initialise a dummy entrypoint
                dn, bn = os.path.split(webapp.entrypoint)
BAIRE Anthony's avatar
todo    
BAIRE Anthony committed
606
                # FIXME: do nothing if entrypoint already exists
607
                prepare = """
608
                    {mkdir}
609
                    test -f {entrypoint} || cat > {entrypoint} <<EOF
610
611
612
613
614
615
616
#!/bin/sh
echo
echo "This is app '{name}' called with parameters '\$@'"
echo
echo "The workdir contains:"
ls -l
EOF
617
                    chmod 0755 -- {entrypoint}
618

619
                """.format( entrypoint  = shlex.quote(webapp.entrypoint),
620
                            name        = webapp.docker_name,
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
                            mkdir       = (("mkdir -p -- %s" % shlex.quote(dn)) if dn else ""))
            else:
                prepare = ""

            command = ["/bin/sh", "-c", """
set -x
export PATH="$PATH:/.toolbox/bin"

{prepare}

# xauth file (needed for X11 forwarding)
touch       /root/.Xauthority
chmod 600   /root/.Xauthority

exec /.toolbox/bin/sshd -D
            """.format(prepare=prepare)]
637

638
            ctrl.sandbox.create_container(image, name=container, hostname=container,
639
                    command = command,
640
641
                    host_config = ctrl.sandbox.create_host_config(
                        binds = {
642
643
644
                            etc_dir: {"bind": "/.sandbox/etc", "mode": "ro"},
                            run_dir: {"bind": "/.sandbox/run", "mode": "rw"},
                            ctrl.toolbox_path: {"bind": "/.toolbox", "mode": "ro"},
645
646
647
                            },
                        # TODO: maybe drop other caps
                        cap_drop = ["NET_RAW"],
648
                        restart_policy = {"Name": "unless-stopped"},
649
                        network_mode = ctrl.sandbox_network,
650
651
652
653
654
                        ))

            ctrl.sandbox.start(container)

        except:
BAIRE Anthony's avatar
BAIRE Anthony committed
655
656
            with docker_warning("cleanup error: unable to remove container %r",
                    container, ignore=docker.errors.NotFound):
657
658
659
                ctrl.sandbox.remove_container(container, force=True)
            raise

BAIRE Anthony's avatar
BAIRE Anthony committed
660
    def _commit(self, webapp, versions):
661
662
663
664
665
666
667
668
669
670
671
        """Commit a webapp sandbox

        (to be executed in a thread pool)

        The image version is looked up in webapp_versions (where state==sandbox).
        
        In case of any error, a recovery version is committed instead (to avoid
        loosing the work done inside the sandbox) and the candidates are put in
        error state.
        """
        ctrl = self.ctrl
BAIRE Anthony's avatar
BAIRE Anthony committed
672
        ses  = ctrl.session
673

BAIRE Anthony's avatar
BAIRE Anthony committed
674
675
676
677
678
679
        # pre-commit checks
        # - ensure that there is exactly one candidate webapp_version with
        #   state=sandbox
        # - otherwise:
        #   - put all candidates in error state
        #   - create a recovery version
680

BAIRE Anthony's avatar
BAIRE Anthony committed
681
682
683
684
685
        # version be committed
        version = None
        
        # error msg (if any)
        error = None
686

BAIRE Anthony's avatar
BAIRE Anthony committed
687
688
        # version ids to be recovered
        recover = ()
689

BAIRE Anthony's avatar
BAIRE Anthony committed
690
691
692
693
694
695
        if len(versions) == 1:
            # normal case (sandbox commit)
            version = versions[0]
            if not version.number:
                error   = "empty version number"
                recover = version.id,
696

BAIRE Anthony's avatar
BAIRE Anthony committed
697
698
699
        elif not versions:
            # sandbox rollback (when user drops a sandbox without committing a new image)
            error     = "dangling sandbox"
700

BAIRE Anthony's avatar
BAIRE Anthony committed
701
702
703
704
705
        else:
            # multiple candidates (should never happen)
            error    = "multiple candidate versions (%s)" % (
                ", ".join(map(repr, sorted(v.number for v in versions))))
            recover = tuple(v.id for v in versions)
706

BAIRE Anthony's avatar
BAIRE Anthony committed
707
708
        # TODO: make 'sandbox' a reserved name

BAIRE Anthony's avatar
BAIRE Anthony committed
709
710
711
        if error:
            changelog = "pre-commit error: " + error
            log.error("sandbox %r version id %r: %s", webapp.docker_name, recover, changelog)
712

BAIRE Anthony's avatar
BAIRE Anthony committed
713
            with ses.begin():
714
715
                # put all candidates in 'error state'
                if recover:
BAIRE Anthony's avatar
BAIRE Anthony committed
716
                    ses.execute('''UPDATE webapp_versions
BAIRE Anthony's avatar
fix sql    
BAIRE Anthony committed
717
718
                            SET changelog=CONCAT(changelog, " [", :changelog, "]"), state=:state
                            WHERE id IN :ids''', dict(changelog=changelog, ids=recover,
BAIRE Anthony's avatar
BAIRE Anthony committed
719
                                state=int(VersionState.ERROR)))
720
721

                # create a recovery version
BAIRE Anthony's avatar
BAIRE Anthony committed
722
723
724
725
726
                version = WebappVersion(
                        webapp_id = webapp.id,
                        number    = time.strftime("recovery-%Y%m%d-%H%M%S"),
                        changelog = changelog,
                        published = False,
BAIRE Anthony's avatar
BAIRE Anthony committed
727
                        state     = int(VersionState.SANDBOX))
BAIRE Anthony's avatar
BAIRE Anthony committed
728
729
730
                ses.add(version)
            ses.refresh(version)
            ses.expunge(version)
731

BAIRE Anthony's avatar
BAIRE Anthony committed
732
        assert version is not None
733

BAIRE Anthony's avatar
BAIRE Anthony committed
734
        # commit the docker image
735

BAIRE Anthony's avatar
BAIRE Anthony committed
736
737
        log.debug("dicts %r %r", webapp.__dict__, version.__dict__)
        log.info("commit sandbox %r version %r", webapp.docker_name, version.number)
738

BAIRE Anthony's avatar
BAIRE Anthony committed
739
740
741
742
        container = webapp.sandbox_name
        next_state = image_size = None
        try:
            # stop the container (if stopping or if creating a new sandbox)
BAIRE Anthony's avatar
BAIRE Anthony committed
743
            if webapp.sandbox_state in (SandboxState.STOPPING, SandboxState.STARTING):
BAIRE Anthony's avatar
BAIRE Anthony committed
744
745
               ctrl.sandbox.stop(container)
               ctrl.sandbox.wait(container)
746

BAIRE Anthony's avatar
BAIRE Anthony committed
747
748
            # commit
            cid = ctrl.sandbox.commit(container, webapp.image_name, version.number)
BAIRE Anthony's avatar
BAIRE Anthony committed
749
            next_state = VersionState.COMMITTED
BAIRE Anthony's avatar
BAIRE Anthony committed
750
751
752
753
754
755
756
            image_size = ctrl.sandbox.inspect_image(cid)["Size"]

            return version, error

        except docker.errors.NotFound:
            error = "commit error: container not found %r" % container
            log.error("%s", error)
BAIRE Anthony's avatar
BAIRE Anthony committed
757
            next_state = VersionState.ERROR
BAIRE Anthony's avatar
BAIRE Anthony committed
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
            image_size = 0
            ses.execute('''UPDATE webapp_versions
                    SET changelog=CONCAT(changelog, " [commit error: sandbox is down]")
                    WHERE id=%d''' % version.id)

            # here we do not propagate the error to allow starting/stopping the
            # sandbox immediately (without going through sandbox_state=:error)
            return None, error

        except Exception as e:
            log.exception("sandbox %r version %r: unexpected commit error (sandbox may still be recovered)",
                    webapp.docker_name, version.number)
            raise

        finally:
            # NOTE: if anything unexpected happens, the version is
            # left in state 'sandbox' and we propagate the exception to
            # ensure the work done inside the sandbox is not dropped
            # and the sandbox is pute in 'error' state
            #
            # The error will not be reported to the user. If this is an issue,
            # then the solution would be to create another error state (to be
            # used when the sandbox is still there).
            # 
            if next_state is not None:
                with ses.begin():
                    ses.execute("UPDATE webapp_versions SET state=%d, docker_image_size=%d WHERE id=%d"
                            % (next_state, image_size, version.id))
BAIRE Anthony's avatar
BAIRE Anthony committed
786

787

BAIRE Anthony's avatar
BAIRE Anthony committed
788
    @asyncio.coroutine
BAIRE Anthony's avatar
BAIRE Anthony committed
789
    def _manage_commit(self, webapp, versions, *, force=False):
BAIRE Anthony's avatar
BAIRE Anthony committed
790
791
792
793
794
795
796
797
        """Manage sandbox commit (if needed) and notify the image manager

        The commit is performed if one of these conditions is fulfilled:
        - a commit was requested (there is at least one WebappVersion entry
          with state=sandbox for this app)
        - a docker container exists (for this sandbox) and force is true
        """

BAIRE Anthony's avatar
BAIRE Anthony committed
798
        if not (versions or (force and self.inspect_sandbox(webapp) is not None)):
BAIRE Anthony's avatar
BAIRE Anthony committed
799
800
801
            return
        
        # do the commit
BAIRE Anthony's avatar
BAIRE Anthony committed
802
        version, error = yield from self.run_in_executor(self._commit, webapp, versions)
BAIRE Anthony's avatar
BAIRE Anthony committed
803
804

        # trigger push/pull operations (but do not wait)
BAIRE Anthony's avatar
BAIRE Anthony committed
805
        if version is not None:
806
807

            # push to the registry
BAIRE Anthony's avatar
BAIRE Anthony committed
808
            self.ctrl.image_manager.push(version.id)
809
810
811
812

            if not error:
                # preemptive pull to the swarm
                # (the image may be needed soon)
BAIRE Anthony's avatar
BAIRE Anthony committed
813
                self.ctrl.image_manager.pull(version.id, swarm=True)
814
815


BAIRE Anthony's avatar
BAIRE Anthony committed
816
817
    def _stop(self, webapp):
        """Stop a webapp sandbox
818

BAIRE Anthony's avatar
BAIRE Anthony committed
819
        (to be executed in a thread pool)
820
821
822
        """
        try:
            # FIXME: remove volumes (v=True) too ?
BAIRE Anthony's avatar
BAIRE Anthony committed
823
            self.ctrl.sandbox.remove_container(webapp.sandbox_name, force=True)
824
825
826
        except docker.errors.NotFound:
            pass

827
828

    @asyncio.coroutine
BAIRE Anthony's avatar
BAIRE Anthony committed
829
    def _process(self, webapp_id, reset, rescheduled):
830
831
832
        ctrl = self.ctrl

        log.debug("process sandbox %d", webapp_id)
BAIRE Anthony's avatar
BAIRE Anthony committed
833
834
835
836
837
838
839
840

        ses = ctrl.session
        with ses.begin():
            # current state of the sandbox + load docker os
            webapp = ses.query(Webapp).filter_by(id=webapp_id).one()
            webapp.docker_os

            # version to be started
841
            sandbox_version = webapp.sandbox_version
BAIRE Anthony's avatar
BAIRE Anthony committed
842
843
844
845
846
847
848
849
850
851

            # requested commits
            commit_versions = self.filter_commit_version(ses.query(WebappVersion), webapp_id).all()

            ses.expunge_all()

        # docker name of the sandbox & image
        webapp.sandbox_name = ctrl.gen_sandbox_name(webapp)
        webapp.image_name   = ctrl.gen_image_name(webapp)

BAIRE Anthony's avatar
BAIRE Anthony committed
852
        phase = "inspect"
853
        next_state = fail_state = None
854
        try:
BAIRE Anthony's avatar
BAIRE Anthony committed
855
            if webapp.sandbox_state == SandboxState.STARTING:
856
                # start the sandbox
BAIRE Anthony's avatar
BAIRE Anthony committed
857
                phase = "start"
BAIRE Anthony's avatar
BAIRE Anthony committed
858
859
                next_state = SandboxState.RUNNING
                fail_state = SandboxState.START_ERROR
860

BAIRE Anthony's avatar
BAIRE Anthony committed
861
                # commit (if a sandbox exists)
BAIRE Anthony's avatar
BAIRE Anthony committed
862
                yield from self._manage_commit(webapp, commit_versions, force=True)
BAIRE Anthony's avatar
BAIRE Anthony committed
863

BAIRE Anthony's avatar
BAIRE Anthony committed
864
                if sandbox_version is not None:
865
866
867
868
869
870
                    # ensure version belongs to this application
                    if sandbox_version.webapp_id != webapp.id:
                        raise Error("invalid version id %d (belongs to webapp %d)" % (
                            sandbox_version.id, sandbox_version.webapp_id))

                    # pull requested image
BAIRE Anthony's avatar
BAIRE Anthony committed
871
                    yield from ctrl.image_manager.pull(sandbox_version.id)
872
873
874
875
876
                else:
                    # pull image
                    yield from ctrl.image_manager.sandbox_pull_manager.process((
                        ctrl.gen_factory_name(webapp.docker_os),
                        webapp.docker_os.version))
BAIRE Anthony's avatar
BAIRE Anthony committed
877
878

                # start sandbox
BAIRE Anthony's avatar
BAIRE Anthony committed
879
                yield from self.run_in_executor(self._start, webapp, sandbox_version)
880

BAIRE Anthony's avatar
BAIRE Anthony committed
881
            elif webapp.sandbox_state == SandboxState.STOPPING:
882
                # stop the sandbox
BAIRE Anthony's avatar
BAIRE Anthony committed
883
                phase = "stop"
BAIRE Anthony's avatar
BAIRE Anthony committed
884
885
                next_state = SandboxState.IDLE
                fail_state = SandboxState.STOP_ERROR
886

BAIRE Anthony's avatar
BAIRE Anthony committed
887
                # commit (if requested)
BAIRE Anthony's avatar
BAIRE Anthony committed
888
                yield from self._manage_commit(webapp, commit_versions)
889

BAIRE Anthony's avatar
BAIRE Anthony committed
890
                yield from self.run_in_executor(self._stop, webapp)
891
892

            else:
BAIRE Anthony's avatar
BAIRE Anthony committed
893
894
                # commit (if requested)
                phase = "commit"
BAIRE Anthony's avatar
BAIRE Anthony committed
895
                yield from self._manage_commit(webapp, commit_versions)
BAIRE Anthony's avatar
BAIRE Anthony committed
896

897
898
899
900
        except ShuttingDown:
            next_state = None
            log.info("sandbox %r %s aborted (controller shutdown)", webapp.docker_name, phase)

901
902
903
904
        except BaseException as e:
            next_state = fail_state

            log_func = log.error if isinstance(e, (docker.errors.APIError, Error)) else log.exception
BAIRE Anthony's avatar
BAIRE Anthony committed
905
            log_func ("sandbox %r %s error (%s)", webapp.docker_name, phase,
906
907
908
909
910
911
912
913
                    traceback.format_exception_only(type(e), e)[-1].strip())

        finally:
            if next_state is not None:
                # atomically update the sandbox state in the db
                # (in case another action is requested during the process, eg: the user
                #  stops the sandbox while it is not fully started)
                log.info("sandbox %r is now in state %r", webapp.docker_name, next_state.name)
BAIRE Anthony's avatar
BAIRE Anthony committed
914
915
916
                with ses.begin():
                    ses.execute("UPDATE webapps SET sandbox_state=%d WHERE id=%d AND sandbox_state=%d" %
                            (next_state, webapp_id, webapp.sandbox_state))
917

918
919
920
            log.debug("done    sandbox %d", webapp_id)


BAIRE Anthony's avatar
BAIRE Anthony committed
921
class JobManager(Manager):
922
    class JobInfo:
923
        __slots__ = "job_id", "ver_id", "ctr_id", "version", "ctr_name", "client", "cpu", "mem", "node_id"
924

925
926
    def __init__(self, ctrl, bigmem_apps=()):
        super().__init__(0)
BAIRE Anthony's avatar
BAIRE Anthony committed
927
928
        self.ctrl = ctrl

929
930
931
932
933
934
935
936
        self.bigmem_apps = bigmem_apps


    @asyncio.coroutine
    def __iter__(self):
        raise NotImplementedError()


937
    def _create_job(self, info):
BAIRE Anthony's avatar
BAIRE Anthony committed
938
939
        ctrl = self.ctrl
        ses  = ctrl.session
940
941
942
        tmp_img = None

        assert info.ctr_id is None
BAIRE Anthony's avatar
BAIRE Anthony committed
943
944
945

        try:
            with ses.begin():
946
                job = ses.query(Job).filter_by(id=info.job_id).one()
BAIRE Anthony's avatar
BAIRE Anthony committed
947
948
949
                webapp = job.webapp

                log.info("start job %d (%s:%s)",
950
                        info.job_id, webapp.docker_name, info.version)
BAIRE Anthony's avatar
BAIRE Anthony committed
951

BAIRE Anthony's avatar
BAIRE Anthony committed
952
                job.state = int(JobState.RUNNING)       # pragma: nobranch (TODO: remove (coverage bug))
BAIRE Anthony's avatar
BAIRE Anthony committed
953
954
955

            
            repo = ctrl.gen_image_name(webapp)
956
            image = "%s:%s" % (repo, info.version)
BAIRE Anthony's avatar
BAIRE Anthony committed
957
958
959
960

            job_path = ctrl.gen_job_path(job)
            log.debug("job.path: %r", job_path)

961
962
963
            if info.ver_id is None:
                assert info.version == "sandbox"
                image = tmp_img = info.client.commit(ctrl.gen_sandbox_name(webapp), repo, info.version)["Id"]
BAIRE Anthony's avatar
BAIRE Anthony committed
964
965
966
967
968
            
            # TODO use another workdir
            # TODO use another uid

            ctrl.check_host_path("isdir", job_path)
969
            hc = ctrl.sandbox.create_host_config(
970
                        binds = {job_path: {"bind": "/tmp"}},
971
972
973
                        cap_drop = ["all"],
                        # FIXME: CAP_DAC_OVERRIDE needed because all nfs files have uid,gid=1000,1000
                        cap_add = ["dac_override"],
974
975
976
                        cpu_quota   = (None if info.cpu is None else (info.cpu * 1024)),
                        cpu_period  = (None if info.cpu is None else 1024),
#                        cpu_shares = info.cpu,
977
#                        mem_reservation = ctrl.mem_soft_limit,
978
                        mem_limit = info.mem,
979
980
981
982
                    )
            if ctrl.mem_soft_limit:
                # TODO: upgrade docker-py (and use create_host_config)
                hc["MemoryReservation"] = ctrl.mem_soft_limit
983
984
985
986
            # NOTE: cpu_shares has a different meaining in docker swarm and docker engine
            #  - swarm:  nb of cpus
            #  - engine: 1/1024 share of the total cpu resouces of the machine
            # engine requires  cpu_share>1
987
            if ctrl.cpu_shares:
BAIRE Anthony's avatar
BAIRE Anthony committed
988
                # TODO: upgrade docker-py (and use create_host_config)
989
                hc["CpuShares"] = info.cpu
990
            log.debug("host_config %r", hc)
991
            info.ctr_id = info.client.create_container(image, name=info.ctr_name,
992
                    working_dir = "/tmp",
993
                    # NOTE: the command line is a little complex, but this is
BAIRE Anthony's avatar
todos    
BAIRE Anthony committed
994
                    #   to ensure that (TODO write tests for this):
995
996
997
998
999
1000
1001
1002
1003
                    #   - no output is lost (we go though a pipe in case the
                    #     app has multiple processes writing to stdout/stderr
                    #     concurrently)
                    #   - we get the exit code of the app (not the exit code of
                    #     cat)
                    #   - we are failsafe (if fifo creation fails then the app
                    #     is run anyway, with the exit code of cat)
                    #   - we have no unusual dependencies (only sh, cat and
                    #     mkfifo)
1004
                    command = ["/bin/sh", "-c", """
1005
                                fifo=/.allgo.fifo.{job_id}
1006
1007
1008
1009
1010
1011
                                if mkfifo "$fifo"
                                then
                                    exec cat <"$fifo" >allgo.log &
                                    exec "$@" >"$fifo" 2>&1 &
                                    wait %1
                                    wait %2
1012
                                    rm "$fifo"
1013
1014
1015
                                else
                                    "$@" 2>&1 | cat >allgo.log
                                fi
1016
1017
1018
1019
                                failcnt="`cat /sys/fs/cgroup/memory/memory.failcnt`"
                                if [ "$failcnt" -ne 0 ] ; then
                                    echo "WARNING: out of memory (memory.failcnt=$failcnt)" >>allgo.log
                                fi
BAIRE Anthony's avatar
BAIRE Anthony committed
1020
                        """.format(job_id=job.id),
1021
                        "job%d" % job.id, webapp.entrypoint] + shlex.split(job.param),
BAIRE Anthony's avatar
BAIRE Anthony committed
1022

1023
                    labels = {"allgo.tmp_img": tmp_img or ""},
1024
                    environment=["constraint:node==" + info.node_id],
1025
                    host_config = hc)["Id"]
1026
            info.client.start(info.ctr_id)
1027
1028
1029
1030
1031

            with ses.begin():
                # save the container_id into the db
                job.container_id = info.ctr_id

1032
        except:
BAIRE Anthony's avatar
todo    
BAIRE Anthony committed
1033
            #TODO introduce a state JobState.ERROR
1034
1035
            self._remove_job(info, tmp_img=tmp_img)
            raise
BAIRE Anthony's avatar
BAIRE Anthony committed
1036
1037


1038
1039
    def _remove_job(self, info, *, tmp_img=None):
        ses = self.ctrl.session
BAIRE Anthony's avatar
BAIRE Anthony committed
1040

1041
1042
1043
        # TODO: report launch errors to the user
        # TODO: report exit code to the user
        # TODO: use another uid
BAIRE Anthony's avatar
BAIRE Anthony committed
1044

1045
1046
1047
1048
1049
1050
1051
        def parse_docker_timestamp(value):
            return datetime.datetime.strptime(
                    # limit the precision to the microsecond
                    # (othewise strptime fails)
                    re.sub(r"(\.\d{,6})\d*Z$", r"\1Z", value),
                    # iso8601 format
                    "%Y-%m-%dT%H:%M:%S.%fZ")
BAIRE Anthony's avatar
BAIRE Anthony committed
1052

1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
        with ses.begin():
            job = ses.query(Job).filter_by(id=info.job_id).one()

            exec_time = 0.0
            if info.ctr_id is not None:
                try:
                    js = info.client.inspect_container(info.ctr_id)
                except docker.errors.NotFound:
                    pass
                else:
                    started_at  = js["State"].get("StartedAt", "0001-")
                    finished_at = js["State"].get("FinishedAt")
BAIRE Anthony's avatar
BAIRE Anthony committed
1065

1066
1067
1068
1069
1070
1071
                    # default docker date is '0001-01-01T00:00:00Z'
                    if not started_at.startswith("0001-"):
                        try:
                            exec_time =( parse_docker_timestamp(finished_at)
                                        - parse_docker_timestamp(started_at)
                                        ).total_seconds()
BAIRE Anthony's avatar
BAIRE Anthony committed
1072
1073
                        except Exception: # pragma: nocover
                            log.exception("job %d: unable to compute exec time", info.job_id)
BAIRE Anthony's avatar
BAIRE Anthony committed
1074

1075
                    if tmp_img is None:
BAIRE Anthony's avatar
BAIRE Anthony committed
1076
                        tmp_img = js["Config"]["Labels"].get("allgo.tmp_img") or None
BAIRE Anthony's avatar
BAIRE Anthony committed
1077

1078
1079
                with docker_warning("job %d: cleanup error: unable to remove container", info.job_id):
                    info.client.remove_container(info.ctr_id)
BAIRE Anthony's avatar
BAIRE Anthony committed
1080
1081

            if tmp_img is not None:
1082
1083
1084
1085
1086
                with docker_warning("job %d: cleanup error: unable to remove tmp image", info.job_id):
                    info.client.remove_image(tmp_img)

            job.exec_time = exec_time
            job.state     = int(JobState.DONE)
1087
            job.container_id = None
1088
1089
1090
1091
1092
1093
1094
1095

        log.info("stop  job %d (duration %fs)", info.job_id, exec_time)


    @asyncio.coroutine
    def _finish_job(self, info):
        # wait for container to terminate
        if info.ctr_id is not None:
1096
            yield from info.client.wait_async(info.ctr_id)
BAIRE Anthony's avatar
BAIRE Anthony committed
1097

1098
1099
        # remove container
        yield from self.run_in_executor(self._remove_job, info, lock=False)
BAIRE Anthony's avatar
BAIRE Anthony committed
1100

1101
    @asyncio.coroutine
BAIRE Anthony's avatar
BAIRE Anthony committed
1102
    def _process(self, job_id, reset, rescheduled):
BAIRE Anthony's avatar
BAIRE Anthony committed
1103
1104
1105
1106
1107
1108
1109
        ctrl = self.ctrl
        ses  = ctrl.session
        log.debug("process job id %d", job_id)

        with ses.begin():
            # query db
            job = ses.query(Job).filter_by(id=job_id).first()
BAIRE Anthony's avatar
BAIRE Anthony committed
1110
            if job is None:     # pragma: nocover
1111
1112
                # unknown job
                log.warning("unknown job id %d", job_id)
BAIRE Anthony's avatar
BAIRE Anthony committed
1113
                return
BAIRE Anthony's avatar
BAIRE Anthony committed
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123

            state = JobState(job.state)

            if job.webapp is None:
                log.error("job %d: webapp id %r not found", job_id, job.webapp_id)
                if state == JobState.WAITING:       # pragma: nobranch
                    job.state = int(JobState.DONE)
                    job.exec_time = 0
                # TODO report error to the user ?
                return
1124
1125

            docker_name = job.webapp.docker_name
1126
1127
1128
1129
            
            info = self.JobInfo()
            info.job_id     = job_id
            info.ctr_id     = None
1130
            info.node_id    = None
1131
1132
            info.version    = job.version
            info.ctr_name   = ctrl.gen_job_name(job)
BAIRE Anthony's avatar
BAIRE Anthony committed
1133

1134
1135
1136
1137
1138
            # NOTE: .cpu .mem are the amount of cpu/mem requested when creating
            # a new job. They do not apply to already created jobs (by a
            # previous instance of the controller)
            info.cpu        = ctrl.cpu_shares
            info.mem        = ctrl.bigmem_hard_limit if docker_name in self.bigmem_apps else ctrl.mem_hard_limit
1139

BAIRE Anthony's avatar
BAIRE Anthony committed
1140
1141

            if job.version == "sandbox":
1142
                info.client  = ctrl.sandbox
BAIRE Anthony's avatar
BAIRE Anthony committed
1143
            else:
1144
1145
1146
1147
1148
                info.client  = ctrl.swarm


            if state == JobState.WAITING:
                # job is not yet started
1149
1150
1151
1152
                if job.container_id is not None:
                    log.warning("job %d is in state WAITING but already has a container id: %r (will be ignored)",
                            job.id, job.container_id)

1153
1154
1155
1156
1157
1158
                if job.version == "sandbox":
                    # to be run in the sandbox
                    info.ver_id = None
                else:
                    # to be run in the swarm

1159
                    # Find the wanted WebappVersion
1160
1161
1162
1163
                    #TODO: replace version_id with webapp_version_id
                    ver = ses.query(WebappVersion).filter_by(
                            webapp_id = job.webapp_id,
                            number    = job.version).filter(
1164
                                WebappVersion.state.in_((
BAIRE Anthony's avatar
BAIRE Anthony committed
1165
1166
                                    int(VersionState.COMMITTED),
                                    int(VersionState.READY)))
1167
1168
1169
                            ).order_by(
                                    WebappVersion.state.desc(),
                                    WebappVersion.id.desc()).first()
1170
1171
1172
1173
1174
1175
1176
1177
1178
                    if ver is None:
                        log.error("job %d: webapp %r version %r not found",
                                job_id, job.webapp.docker_name, job.version)

                        job.state = int(JobState.DONE)
                        # TODO report error to the user
                        return
                    info.ver_id = ver.id

BAIRE Anthony's avatar
BAIRE Anthony committed
1179
            elif state == JobState.RUNNING:     # pragma: nobranch
1180
1181
1182
1183
1184
1185
                # job is already started

                # we do not care about the actual version_id *but* we need to
                # know whether we are in the swarm or in the sandbox
                info.ver_id = None if job.version == "sandbox" else -1

1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
                if job.container_id is None:
                    # look up container id (if it is not yet in the db)
                    # FIXME: to be removed after migration (the container id is now stored in the db at creation time)
                    try:
                        info.ctr_id = job.container_id = info.client.inspect_container(
                                "%s-job-%d-%s" % (ctrl.env, job.id, job.webapp.docker_name))["Id"]
                    except docker.errors.NotFound:
                        pass
                else:
                    # check the presence of the container and validates its id against its name
                    info.ctr_id = ctrl.check_job_container(info.client, job)
1197
                    
1198
            else:
1199
                # unexpected state
1200
1201
                if state != JobState.DONE:
                    log.warning("job id %d is in unexpected state %s", job_id, state.name)
1202
1203
                return

BAIRE Anthony's avatar
BAIRE Anthony committed
1204

1205
1206
1207
1208
1209
        if state == JobState.WAITING:
            # job is not yet started

            # pull the image to the swarm
            if info.ver_id is not None:
BAIRE Anthony's avatar
BAIRE Anthony committed
1210
1211
                # NOTE: race condition: will fail if ver.state==sandbox
                #  jobs must be submitted after the image is committed