[Notes] [Git][BuildStream/buildstream][tristan/element-processing-order] 5 commits: setup.py: Import server main from _cas/casserver.py

[Tristan Van Berkom <gitlab mg gitlab com>]

Title: GitLab

Tristan Van Berkom pushed to branch tristan/element-processing-order at BuildStream / buildstream

Commits:

• 081e8426
  by James Ennis at 2019-01-16T17:58:41Z

  setup.py: Import server main from _cas/casserver.py
  
  Since the CAS refactor, we have not been able to execute
  bst-artifact-server. This commit ensures that we can.
  
  This fix closes #867
  

• e385660c
  by James Ennis at 2019-01-16T18:33:20Z

  Merge branch 'jennis/fix_bst_artifact_server' into 'master'
  
  Ensure that we can execute `bst-artifact-server`
  
  Closes #867
  
  See merge request BuildStream/buildstream!1079

• 6991d6d6
  by Tristan Van Berkom at 2019-01-16T19:28:14Z

  _scheduler: Refactor of queues and resources.
  
  This branch makes the following changes:
  
    * jobs/job.py: No longer stores any interested resource list
  
      Jobs are ephemeral again, they only ever exist while they
      are running.
  
    * queues/queue.py: Revert to only handling lists of elements
  
      Elements pass through the queues, Queue.harvest_jobs()
      replaces Queue.pop_ready_jobs() and now the Queue stops
      creating jobs as soon as there are not enough resources
      for the job.
  
      Also removed unused `prepare()` abstract method.
  
    * queues/buildqueue.py: Adapt the part where we launch a job
  
      This part needs to be reworked anyway, just touch it up for
      now so that it doesnt break with the surrounding changes.
  
    * jobs/{cachesize,cleanup}job.py: Expose uniform complete callback
  
      Allows the scheduler to manage resource deallocation for these
      two job completions as a custom thing, at the same phase
      that the Queues take care of their own resource deallocation.
  
    * resources.py: No longer has knowledge of the job
  
      Since jobs are ephemeral, they are not a suitable place
      to store the resource identifiers, these must be provided
      by the callers wherever needed.
  
      Now the main Resources object is owned by the Scheduler
      but shared with Queues, each take care of managing the
      resources of the jobs they create through the same
      resource API.
  
    * scheduler.py: Reverted to only creating jobs on demand
  
      This changes the flow of the scheduler such that whenever
      jobs complete, the queues are interrogated for as many
      jobs which can run at the moment but not more; and this
      completely removes the waiting list.
  
      For the internal cache management jobs, we handle this
      with a little state instead of having a waiting list
      and only launch when the resources permit it.
  
  By abolishing the scheduler waiting list and creating jobs
  on demand, we fix the order of element processing and consequently
  fix issue #712.
  

• 04ba59bc
  by Tristan Van Berkom at 2019-01-16T19:28:14Z

  tests/frontend/order.py: Enable the test for build and fix the fetch tests
  
  With the scheduler changes, fetch jobs get automatically skipped
  so the output is changed, using a separate repo for each element
  fixes the test such that every fetch job gets a job launched.
  

• 32bdded8
  by Tristan Van Berkom at 2019-01-16T19:28:14Z

  _artifactcache/artifactcache.py: Rephrase failure message
  
  It was saying "There is not enough space to build the given element.",
  this makes me think the error is associated to a specific element, but
  this does not make sense to show up in a cleanup task.
  
  Instead say "There is not enough space to complete the build.", which
  should be more clear that even after cleaning up there is not enough
  space.
  

10 changed files:

• buildstream/_artifactcache.py
• buildstream/_scheduler/jobs/cachesizejob.py
• buildstream/_scheduler/jobs/cleanupjob.py
• buildstream/_scheduler/jobs/job.py
• buildstream/_scheduler/queues/buildqueue.py
• buildstream/_scheduler/queues/queue.py
• buildstream/_scheduler/resources.py
• buildstream/_scheduler/scheduler.py
• setup.py
• tests/frontend/order.py

Changes:

buildstream/_artifactcache.py

---

...	...	@@ -247,7 +247,7 @@ class ArtifactCache():
247	247	# FIXME: Asking the user what to do may be neater
248	248	default_conf = os.path.join(os.environ['XDG_CONFIG_HOME'],
249	249	'buildstream.conf')
250		- detail = ("There is not enough space to build the given element.\n"
	250	+ detail = ("There is not enough space to complete the build.\n"
251	251	"Please increase the cache-quota in {}."
252	252	.format(self.context.config_origin or default_conf))
253	253

buildstream/_scheduler/jobs/cachesizejob.py

---

...	...	@@ -34,8 +34,8 @@ class CacheSizeJob(Job):
34	34	if status == JobStatus.OK:
35	35	self._artifacts.set_cache_size(result)
36	36
37		- if self._complete_cb:
38		- self._complete_cb(result)
	37	+ if self._complete_cb:
	38	+ self._complete_cb(status, result)
39	39
40	40	def child_process_data(self):
41	41	return {}

buildstream/_scheduler/jobs/cleanupjob.py

---

...	...	@@ -20,8 +20,9 @@ from .job import Job, JobStatus
20	20
21	21
22	22	class CleanupJob(Job):
23		- def __init__(self, *args, **kwargs):
	23	+ def __init__(self, *args, complete_cb, **kwargs):
24	24	super().__init__(*args, **kwargs)
	25	+ self._complete_cb = complete_cb
25	26
26	27	context = self._scheduler.context
27	28	self._artifacts = context.artifactcache
...	...	@@ -32,3 +33,6 @@ class CleanupJob(Job):
32	33	def parent_complete(self, status, result):
33	34	if status == JobStatus.OK:
34	35	self._artifacts.set_cache_size(result)
	36	+
	37	+ if self._complete_cb:
	38	+ self._complete_cb(status, result)

buildstream/_scheduler/jobs/job.py

---

...	...	@@ -85,28 +85,11 @@ class Process(multiprocessing.Process):
85	85	# action_name (str): The queue action name
86	86	# logfile (str): A template string that points to the logfile
87	87	# that should be used - should contain {pid}.
88		-# resources (iter(ResourceType)) - A set of resources this job
89		-# wants to use.
90		-# exclusive_resources (iter(ResourceType)) - A set of resources
91		-# this job wants to use
92		-# exclusively.
93	88	# max_retries (int): The maximum number of retries
94	89	#
95	90	class Job():
96	91
97		- def __init__(self, scheduler, action_name, logfile, *,
98		- resources=None, exclusive_resources=None, max_retries=0):
99		-
100		- if resources is None:
101		- resources = set()
102		- else:
103		- resources = set(resources)
104		- if exclusive_resources is None:
105		- exclusive_resources = set()
106		- else:
107		- exclusive_resources = set(resources)
108		-
109		- assert exclusive_resources <= resources, "All exclusive resources must also be resources!"
	92	+ def __init__(self, scheduler, action_name, logfile, *, max_retries=0):
110	93
111	94	#
112	95	# Public members
...	...	@@ -114,12 +97,6 @@ class Job():
114	97	self.action_name = action_name # The action name for the Queue
115	98	self.child_data = None # Data to be sent to the main process
116	99
117		- # The resources this job wants to access
118		- self.resources = resources
119		- # Resources this job needs to access exclusively, i.e., no
120		- # other job should be allowed to access them
121		- self.exclusive_resources = exclusive_resources
122		-
123	100	#
124	101	# Private members
125	102	#

buildstream/_scheduler/queues/buildqueue.py

---

...	...	@@ -57,11 +57,10 @@ class BuildQueue(Queue):
57	57	logfile=logfile)
58	58	job = ElementJob(self._scheduler, self.action_name,
59	59	logfile, element=element, queue=self,
60		- resources=self.resources,
61	60	action_cb=self.process,
62	61	complete_cb=self._job_done,
63	62	max_retries=self._max_retries)
64		- self._done_queue.append(job)
	63	+ self._done_queue.append(element)
65	64	self.failed_elements.append(element)
66	65	self._scheduler._job_complete_callback(job, False)
67	66

buildstream/_scheduler/queues/queue.py

---

...	...	@@ -72,8 +72,9 @@ class Queue():
72	72	# Private members
73	73	#
74	74	self._scheduler = scheduler
75		- self._wait_queue = deque()
76		- self._done_queue = deque()
	75	+ self._resources = scheduler.resources # Shared resource pool
	76	+ self._wait_queue = deque() # Ready / Waiting elements
	77	+ self._done_queue = deque() # Processed / Skipped elements
77	78	self._max_retries = 0
78	79
79	80	# Assert the subclass has setup class data
...	...	@@ -115,16 +116,6 @@ class Queue():
115	116	def status(self, element):
116	117	return QueueStatus.READY
117	118
118		- # prepare()
119		- #
120		- # Abstract method for handling job preparation in the main process.
121		- #
122		- # Args:
123		- # element (Element): The element which is scheduled
124		- #
125		- def prepare(self, element):
126		- pass
127		-
128	119	# done()
129	120	#
130	121	# Abstract method for handling a successful job completion.
...	...	@@ -153,26 +144,18 @@ class Queue():
153	144	if not elts:
154	145	return
155	146
156		- # Note: The internal lists work with jobs. This is not
157		- # reflected in any external methods (except
158		- # pop/peek_ready_jobs).
159		- def create_job(element):
160		- logfile = self._element_log_path(element)
161		- return ElementJob(self._scheduler, self.action_name,
162		- logfile, element=element, queue=self,
163		- resources=self.resources,
164		- action_cb=self.process,
165		- complete_cb=self._job_done,
166		- max_retries=self._max_retries)
167		-
168		- # Place skipped elements directly on the done queue
169		- jobs = [create_job(elt) for elt in elts]
170		- skip = [job for job in jobs if self.status(job.element) == QueueStatus.SKIP]
171		- wait = [job for job in jobs if job not in skip]
172		-
173		- self.skipped_elements.extend([job.element for job in skip])
174		- self._wait_queue.extend(wait)
175		- self._done_queue.extend(skip)
	147	+ # Place skipped elements on the done queue right away.
	148	+ #
	149	+ # The remaining ready and waiting elements must remain in the
	150	+ # same queue, and ready status must be determined at the moment
	151	+ # which the scheduler is asking for the next job.
	152	+ #
	153	+ skip = [elt for elt in elts if self.status(elt) == QueueStatus.SKIP]
	154	+ wait = [elt for elt in elts if elt not in skip]
	155	+
	156	+ self.skipped_elements.extend(skip) # Public record of skipped elements
	157	+ self._done_queue.extend(skip) # Elements to be processed
	158	+ self._wait_queue.extend(wait) # Elements eligible to be dequeued
176	159
177	160	# dequeue()
178	161	#
...	...	@@ -184,69 +167,59 @@ class Queue():
184	167	#
185	168	def dequeue(self):
186	169	while self._done_queue:
187		- yield self._done_queue.popleft().element
	170	+ yield self._done_queue.popleft()
188	171
189	172	# dequeue_ready()
190	173	#
191		- # Reports whether there are any elements to dequeue
	174	+ # Reports whether any elements can be promoted to other queues
192	175	#
193	176	# Returns:
194		- # (bool): Whether there are elements to dequeue
	177	+ # (bool): Whether there are elements ready
195	178	#
196	179	def dequeue_ready(self):
197	180	return any(self._done_queue)
198	181
199		- # pop_ready_jobs()
200		- #
201		- # Returns:
202		- # ([Job]): A list of jobs to run
	182	+ # harvest_jobs()
203	183	#
204	184	# Process elements in the queue, moving elements which were enqueued
205		- # into the dequeue pool, and processing them if necessary.
206		- #
207		- # This will have different results for elements depending
208		- # on the Queue.status() implementation.
209		- #
210		- # o Elements which are QueueStatus.WAIT will not be affected
	185	+ # into the dequeue pool, and creating as many jobs for which resources
	186	+ # can be reserved.
211	187	#
212		- # o Elements which are QueueStatus.SKIP will move directly
213		- # to the dequeue pool
214		- #
215		- # o For Elements which are QueueStatus.READY a Job will be
216		- # created and returned to the caller, given that the scheduler
217		- # allows the Queue enough resources for the given job
	188	+ # Returns:
	189	+ # ([Job]): A list of jobs which can be run now
218	190	#
219		- def pop_ready_jobs(self):
	191	+ def harvest_jobs(self):
220	192	unready = []
221	193	ready = []
222	194
223	195	while self._wait_queue:
224		- job = self._wait_queue.popleft()
225		- element = job.element
	196	+ if not self._resources.reserve(self.resources, peek=True):
	197	+ break
226	198
	199	+ element = self._wait_queue.popleft()
227	200	status = self.status(element)
	201	+
228	202	if status == QueueStatus.WAIT:
229		- unready.append(job)
230		- continue
	203	+ unready.append(element)
231	204	elif status == QueueStatus.SKIP:
232		- self._done_queue.append(job)
	205	+ self._done_queue.append(element)
233	206	self.skipped_elements.append(element)
234		- continue
235		-
236		- self.prepare(element)
237		- ready.append(job)
	207	+ else:
	208	+ reserved = self._resources.reserve(self.resources)
	209	+ assert reserved
	210	+ ready.append(element)
238	211
239		- # These were not ready but were in the beginning, give em
240		- # first priority again next time around
241	212	self._wait_queue.extendleft(unready)
242	213
243		- return ready
244		-
245		- def peek_ready_jobs(self):
246		- def ready(job):
247		- return self.status(job.element) == QueueStatus.READY
248		-
249		- yield from (job for job in self._wait_queue if ready(job))
	214	+ return [
	215	+ ElementJob(self._scheduler, self.action_name,
	216	+ self._element_log_path(element),
	217	+ element=element, queue=self,
	218	+ action_cb=self.process,
	219	+ complete_cb=self._job_done,
	220	+ max_retries=self._max_retries)
	221	+ for element in ready
	222	+ ]
250	223
251	224	#####################################################
252	225	# Private Methods #
...	...	@@ -292,6 +265,10 @@ class Queue():
292	265	#
293	266	def _job_done(self, job, element, status, result):
294	267
	268	+ # Now release the resources we reserved
	269	+ #
	270	+ self._resources.release(self.resources)
	271	+
295	272	# Update values that need to be synchronized in the main task
296	273	# before calling any queue implementation
297	274	self._update_workspaces(element, job)
...	...	@@ -324,12 +301,8 @@ class Queue():
324	301	detail=traceback.format_exc())
325	302	self.failed_elements.append(element)
326	303	else:
327		- #
328		- # No exception occured in post processing
329		- #
330		-
331		- # All jobs get placed on the done queue for later processing.
332		- self._done_queue.append(job)
	304	+ # All elements get placed on the done queue for later processing.
	305	+ self._done_queue.append(element)
333	306
334	307	# These lists are for bookkeeping purposes for the UI and logging.
335	308	if status == JobStatus.SKIPPED:

buildstream/_scheduler/resources.py

---

...	...	@@ -34,28 +34,25 @@ class Resources():
34	34	ResourceType.UPLOAD: set()
35	35	}
36	36
37		- def clear_job_resources(self, job):
38		- for resource in job.exclusive_resources:
39		- self._exclusive_resources[resource].remove(hash(job))
	37	+ # reserve()
	38	+ #
	39	+ # Reserves a set of resources
	40	+ #
	41	+ # Args:
	42	+ # resources (set): A set of ResourceTypes
	43	+ # exclusive (set): Another set of ResourceTypes
	44	+ # peek (bool): Whether to only peek at whether the resource is available
	45	+ #
	46	+ # Returns:
	47	+ # (bool): True if the resources could be reserved
	48	+ #
	49	+ def reserve(self, resources, exclusive=None, *, peek=False):
	50	+ if exclusive is None:
	51	+ exclusive = set()
40	52
41		- for resource in job.resources:
42		- self._used_resources[resource] -= 1
43		-
44		- def reserve_exclusive_resources(self, job):
45		- exclusive = job.exclusive_resources
46		-
47		- # The very first thing we do is to register any exclusive
48		- # resources this job may want. Even if the job is not yet
49		- # allowed to run (because another job is holding the resource
50		- # it wants), we can still set this - it just means that any
51		- # job *currently* using these resources has to finish first,
52		- # and no new jobs wanting these can be launched (except other
53		- # exclusive-access jobs).
54		- #
55		- for resource in exclusive:
56		- self._exclusive_resources[resource].add(hash(job))
	53	+ resources = set(resources)
	54	+ exclusive = set(exclusive)
57	55
58		- def reserve_job_resources(self, job):
59	56	# First, we check if the job wants to access a resource that
60	57	# another job wants exclusive access to. If so, it cannot be
61	58	# scheduled.
...	...	@@ -68,7 +65,8 @@ class Resources():
68	65	# is currently not possible, but may be worth thinking
69	66	# about.
70	67	#
71		- for resource in job.resources - job.exclusive_resources:
	68	+ for resource in resources - exclusive:
	69	+
72	70	# If our job wants this resource exclusively, we never
73	71	# check this, so we can get away with not (temporarily)
74	72	# removing it from the set.
...	...	@@ -84,14 +82,14 @@ class Resources():
84	82	# at a time, despite being allowed to be part of the exclusive
85	83	# set.
86	84	#
87		- for exclusive in job.exclusive_resources:
88		- if self._used_resources[exclusive] != 0:
	85	+ for resource in exclusive:
	86	+ if self._used_resources[resource] != 0:
89	87	return False
90	88
91	89	# Finally, we check if we have enough of each resource
92	90	# available. If we don't have enough, the job cannot be
93	91	# scheduled.
94		- for resource in job.resources:
	92	+ for resource in resources:
95	93	if (self._max_resources[resource] > 0 and
96	94	self._used_resources[resource] >= self._max_resources[resource]):
97	95	return False
...	...	@@ -99,7 +97,70 @@ class Resources():
99	97	# Now we register the fact that our job is using the resources
100	98	# it asked for, and tell the scheduler that it is allowed to
101	99	# continue.
102		- for resource in job.resources:
103		- self._used_resources[resource] += 1
	100	+ if not peek:
	101	+ for resource in resources:
	102	+ self._used_resources[resource] += 1
104	103
105	104	return True
	105	+
	106	+ # release()
	107	+ #
	108	+ # Release resources previously reserved with Resources.reserve()
	109	+ #
	110	+ # Args:
	111	+ # resources (set): A set of resources to release
	112	+ #
	113	+ def release(self, resources):
	114	+ for resource in resources:
	115	+ assert self._used_resources[resource] > 0, "Scheduler resource imbalance"
	116	+ self._used_resources[resource] -= 1
	117	+
	118	+ # register_exclusive_interest()
	119	+ #
	120	+ # Inform the resources pool that `source` has an interest in
	121	+ # reserving this resource exclusively.
	122	+ #
	123	+ # The source parameter is used to identify the caller, it
	124	+ # must be ensured to be unique for the time that the
	125	+ # interest is registered.
	126	+ #
	127	+ # This function may be called multiple times, and subsequent
	128	+ # calls will simply have no effect until clear_exclusive_interest()
	129	+ # is used to clear the interest.
	130	+ #
	131	+ # This must be called in advance of reserve()
	132	+ #
	133	+ # Args:
	134	+ # resources (set): Set of resources to reserve exclusively
	135	+ # source (any): Source identifier, to be used again when unregistering
	136	+ # the interest.
	137	+ #
	138	+ def register_exclusive_interest(self, resources, source):
	139	+
	140	+ # The very first thing we do is to register any exclusive
	141	+ # resources this job may want. Even if the job is not yet
	142	+ # allowed to run (because another job is holding the resource
	143	+ # it wants), we can still set this - it just means that any
	144	+ # job *currently* using these resources has to finish first,
	145	+ # and no new jobs wanting these can be launched (except other
	146	+ # exclusive-access jobs).
	147	+ #
	148	+ for resource in resources:
	149	+ self._exclusive_resources[resource].add(source)
	150	+
	151	+ # unregister_exclusive_interest()
	152	+ #
	153	+ # Clear the exclusive interest in these resources.
	154	+ #
	155	+ # This should be called by the given source which registered
	156	+ # an exclusive interest.
	157	+ #
	158	+ # Args:
	159	+ # resources (set): Set of resources to reserve exclusively
	160	+ # source (str): Source identifier, to be used again when unregistering
	161	+ # the interest.
	162	+ #
	163	+ def unregister_exclusive_interest(self, resources, source):
	164	+
	165	+ for resource in resources:
	166	+ self._exclusive_resources[resource].remove(source)

buildstream/_scheduler/scheduler.py

---

...	...	@@ -28,7 +28,7 @@ from contextlib import contextmanager
28	28
29	29	# Local imports
30	30	from .resources import Resources, ResourceType
31		-from .jobs import CacheSizeJob, CleanupJob
	31	+from .jobs import JobStatus, CacheSizeJob, CleanupJob
32	32
33	33
34	34	# A decent return code for Scheduler.run()
...	...	@@ -38,14 +38,10 @@ class SchedStatus():
38	38	TERMINATED = 1
39	39
40	40
41		-# Our _REDUNDANT_EXCLUSIVE_ACTIONS jobs are special ones
42		-# which we launch dynamically, they have the property of being
43		-# meaningless to queue if one is already queued, and it also
44		-# doesnt make sense to run them in parallel
	41	+# Some action names for the internal jobs we launch
45	42	#
46	43	_ACTION_NAME_CLEANUP = 'cleanup'
47	44	_ACTION_NAME_CACHE_SIZE = 'cache_size'
48		-_REDUNDANT_EXCLUSIVE_ACTIONS = [_ACTION_NAME_CLEANUP, _ACTION_NAME_CACHE_SIZE]
49	45
50	46
51	47	# Scheduler()
...	...	@@ -81,8 +77,6 @@ class Scheduler():
81	77	#
82	78	# Public members
83	79	#
84		- self.active_jobs = [] # Jobs currently being run in the scheduler
85		- self.waiting_jobs = [] # Jobs waiting for resources
86	80	self.queues = None # Exposed for the frontend to print summaries
87	81	self.context = context # The Context object shared with Queues
88	82	self.terminated = False # Whether the scheduler was asked to terminate or has terminated
...	...	@@ -95,15 +89,23 @@ class Scheduler():
95	89	#
96	90	# Private members
97	91	#
	92	+ self._active_jobs = [] # Jobs currently being run in the scheduler
	93	+ self._starttime = start_time # Initial application start time
	94	+ self._suspendtime = None # Session time compensation for suspended state
	95	+ self._queue_jobs = True # Whether we should continue to queue jobs
	96	+
	97	+ # State of cache management related jobs
	98	+ self._cache_size_scheduled = False # Whether we have a cache size job scheduled
	99	+ self._cache_size_running = None # A running CacheSizeJob, or None
	100	+ self._cleanup_scheduled = False # Whether we have a cleanup job scheduled
	101	+ self._cleanup_running = None # A running CleanupJob, or None
	102	+
	103	+ # Callbacks to report back to the Scheduler owner
98	104	self._interrupt_callback = interrupt_callback
99	105	self._ticker_callback = ticker_callback
100	106	self._job_start_callback = job_start_callback
101	107	self._job_complete_callback = job_complete_callback
102	108
103		- self._starttime = start_time
104		- self._suspendtime = None
105		- self._queue_jobs = True # Whether we should continue to queue jobs
106		-
107	109	# Whether our exclusive jobs, like 'cleanup' are currently already
108	110	# waiting or active.
109	111	#
...	...	@@ -113,9 +115,9 @@ class Scheduler():
113	115	self._exclusive_waiting = set()
114	116	self._exclusive_active = set()
115	117
116		- self._resources = Resources(context.sched_builders,
117		- context.sched_fetchers,
118		- context.sched_pushers)
	118	+ self.resources = Resources(context.sched_builders,
	119	+ context.sched_fetchers,
	120	+ context.sched_pushers)
119	121
120	122	# run()
121	123	#
...	...	@@ -150,7 +152,7 @@ class Scheduler():
150	152	self._connect_signals()
151	153
152	154	# Run the queues
153		- self._schedule_queue_jobs()
	155	+ self._sched()
154	156	self.loop.run_forever()
155	157	self.loop.close()
156	158
...	...	@@ -240,12 +242,14 @@ class Scheduler():
240	242	# status (JobStatus): The status of the completed job
241	243	#
242	244	def job_completed(self, job, status):
243		- self._resources.clear_job_resources(job)
244		- self.active_jobs.remove(job)
245		- if job.action_name in _REDUNDANT_EXCLUSIVE_ACTIONS:
246		- self._exclusive_active.remove(job.action_name)
	245	+
	246	+ # Remove from the active jobs list
	247	+ self._active_jobs.remove(job)
	248	+
	249	+ # Scheduler owner facing callback
247	250	self._job_complete_callback(job, status)
248		- self._schedule_queue_jobs()
	251	+
	252	+ # Now check for more jobs
249	253	self._sched()
250	254
251	255	# check_cache_size():
...	...	@@ -255,78 +259,104 @@ class Scheduler():
255	259	# if needed.
256	260	#
257	261	def check_cache_size(self):
258		- job = CacheSizeJob(self, _ACTION_NAME_CACHE_SIZE,
259		- 'cache_size/cache_size',
260		- resources=[ResourceType.CACHE,
261		- ResourceType.PROCESS],
262		- complete_cb=self._run_cleanup)
263		- self._schedule_jobs([job])
	262	+
	263	+ # Here we assume we are called in response to a job
	264	+ # completion callback, or before entering the scheduler.
	265	+ #
	266	+ # As such there is no need to call `_sched()` from here,
	267	+ # and we prefer to run it once at the last moment.
	268	+ #
	269	+ self._cache_size_scheduled = True
264	270
265	271	#######################################################
266	272	# Local Private Methods #
267	273	#######################################################
268	274
269		- # _sched()
	275	+ # _spawn_job()
270	276	#
271		- # The main driving function of the scheduler, it will be called
272		- # automatically when Scheduler.run() is called initially,
	277	+ # Spanws a job
273	278	#
274		- def _sched(self):
275		- for job in self.waiting_jobs:
276		- self._resources.reserve_exclusive_resources(job)
	279	+ # Args:
	280	+ # job (Job): The job to spawn
	281	+ #
	282	+ def _spawn_job(self, job):
	283	+ job.spawn()
	284	+ self._active_jobs.append(job)
	285	+ if self._job_start_callback:
	286	+ self._job_start_callback(job)
277	287
278		- for job in self.waiting_jobs:
279		- if not self._resources.reserve_job_resources(job):
280		- continue
	288	+ # Callback for the cache size job
	289	+ def _cache_size_job_complete(self, status, cache_size):
281	290
282		- # Postpone these jobs if one is already running
283		- if job.action_name in _REDUNDANT_EXCLUSIVE_ACTIONS and \
284		- job.action_name in self._exclusive_active:
285		- continue
	291	+ # Deallocate cache size job resources
	292	+ self._cache_size_running = None
	293	+ self.resources.release([ResourceType.CACHE, ResourceType.PROCESS])
286	294
287		- job.spawn()
288		- self.waiting_jobs.remove(job)
289		- self.active_jobs.append(job)
	295	+ # Schedule a cleanup job if we've hit the threshold
	296	+ if status != JobStatus.OK:
	297	+ return
290	298
291		- if job.action_name in _REDUNDANT_EXCLUSIVE_ACTIONS:
292		- self._exclusive_waiting.remove(job.action_name)
293		- self._exclusive_active.add(job.action_name)
	299	+ context = self.context
	300	+ artifacts = context.artifactcache
294	301
295		- if self._job_start_callback:
296		- self._job_start_callback(job)
	302	+ if artifacts.has_quota_exceeded():
	303	+ self._cleanup_scheduled = True
297	304
298		- # If nothings ticking, time to bail out
299		- if not self.active_jobs and not self.waiting_jobs:
300		- self.loop.stop()
	305	+ # Callback for the cleanup job
	306	+ def _cleanup_job_complete(self, status, cache_size):
301	307
302		- # _schedule_jobs()
303		- #
304		- # The main entry point for jobs to be scheduled.
305		- #
306		- # This is called either as a result of scanning the queues
307		- # in _schedule_queue_jobs(), or directly by the Scheduler
308		- # to insert special jobs like cleanups.
	308	+ # Deallocate cleanup job resources
	309	+ self._cleanup_running = None
	310	+ self.resources.release([ResourceType.CACHE, ResourceType.PROCESS])
	311	+
	312	+ # Unregister the exclusive interest when we're done with it
	313	+ if not self._cleanup_scheduled:
	314	+ self.resources.unregister_exclusive_interest(
	315	+ [ResourceType.CACHE], 'cache-cleanup'
	316	+ )
	317	+
	318	+ # _sched_cleanup_job()
309	319	#
310		- # Args:
311		- # jobs ([Job]): A list of jobs to schedule
	320	+ # Runs a cleanup job if one is scheduled to run now and
	321	+ # sufficient recources are available.
312	322	#
313		- def _schedule_jobs(self, jobs):
314		- for job in jobs:
	323	+ def _sched_cleanup_job(self):
315	324
316		- # Special treatment of our redundant exclusive jobs
317		- #
318		- if job.action_name in _REDUNDANT_EXCLUSIVE_ACTIONS:
	325	+ if self._cleanup_scheduled and self._cleanup_running is None:
	326	+
	327	+ # Ensure we have an exclusive interest in the resources
	328	+ self.resources.register_exclusive_interest(
	329	+ [ResourceType.CACHE], 'cache-cleanup'
	330	+ )
	331	+
	332	+ if self.resources.reserve([ResourceType.CACHE, ResourceType.PROCESS],
	333	+ [ResourceType.CACHE]):
319	334
320		- # Drop the job if one is already queued
321		- if job.action_name in self._exclusive_waiting:
322		- continue
	335	+ # Update state and launch
	336	+ self._cleanup_scheduled = False
	337	+ self._cleanup_running = \
	338	+ CleanupJob(self, _ACTION_NAME_CLEANUP, 'cleanup/cleanup',
	339	+ complete_cb=self._cleanup_job_complete)
	340	+ self._spawn_job(self._cleanup_running)
323	341
324		- # Mark this action type as queued
325		- self._exclusive_waiting.add(job.action_name)
	342	+ # _sched_cache_size_job()
	343	+ #
	344	+ # Runs a cache size job if one is scheduled to run now and
	345	+ # sufficient recources are available.
	346	+ #
	347	+ def _sched_cache_size_job(self):
	348	+
	349	+ if self._cache_size_scheduled and not self._cache_size_running:
326	350
327		- self.waiting_jobs.append(job)
	351	+ if self.resources.reserve([ResourceType.CACHE, ResourceType.PROCESS]):
	352	+ self._cache_size_scheduled = False
	353	+ self._cache_size_running = \
	354	+ CacheSizeJob(self, _ACTION_NAME_CACHE_SIZE,
	355	+ 'cache_size/cache_size',
	356	+ complete_cb=self._cache_size_job_complete)
	357	+ self._spawn_job(self._cache_size_running)
328	358
329		- # _schedule_queue_jobs()
	359	+ # _sched_queue_jobs()
330	360	#
331	361	# Ask the queues what jobs they want to schedule and schedule
332	362	# them. This is done here so we can ask for new jobs when jobs
...	...	@@ -335,7 +365,7 @@ class Scheduler():
335	365	# This will process the Queues, pull elements through the Queues
336	366	# and process anything that is ready.
337	367	#
338		- def _schedule_queue_jobs(self):
	368	+ def _sched_queue_jobs(self):
339	369	ready = []
340	370	process_queues = True
341	371
...	...	@@ -344,10 +374,7 @@ class Scheduler():
344	374	# Pull elements forward through queues
345	375	elements = []
346	376	for queue in self.queues:
347		- # Enqueue elements complete from the last queue
348	377	queue.enqueue(elements)
349		-
350		- # Dequeue processed elements for the next queue
351	378	elements = list(queue.dequeue())
352	379
353	380	# Kickoff whatever processes can be processed at this time
...	...	@@ -362,41 +389,51 @@ class Scheduler():
362	389	# thus need all the pulls to complete before ever starting
363	390	# a build
364	391	ready.extend(chain.from_iterable(
365		- queue.pop_ready_jobs() for queue in reversed(self.queues)
	392	+ q.harvest_jobs() for q in reversed(self.queues)
366	393	))
367	394
368		- # pop_ready_jobs() may have skipped jobs, adding them to
369		- # the done_queue. Pull these skipped elements forward to
370		- # the next queue and process them.
	395	+ # harvest_jobs() may have decided to skip some jobs, making
	396	+ # them eligible for promotion to the next queue as a side effect.
	397	+ #
	398	+ # If that happens, do another round.
371	399	process_queues = any(q.dequeue_ready() for q in self.queues)
372	400
373		- self._schedule_jobs(ready)
374		- self._sched()
	401	+ # Spawn the jobs
	402	+ #
	403	+ for job in ready:
	404	+ self._spawn_job(job)
375	405
376		- # _run_cleanup()
377		- #
378		- # Schedules the cache cleanup job if the passed size
379		- # exceeds the cache quota.
	406	+ # _sched()
380	407	#
381		- # Args:
382		- # cache_size (int): The calculated cache size (ignored)
	408	+ # Run any jobs which are ready to run, or quit the main loop
	409	+ # when nothing is running or is ready to run.
383	410	#
384		- # NOTE: This runs in response to completion of the cache size
385		- # calculation job lauched by Scheduler.check_cache_size(),
386		- # which will report the calculated cache size.
	411	+ # This is the main driving function of the scheduler, it is called
	412	+ # initially when we enter Scheduler.run(), and at the end of whenever
	413	+ # any job completes, after any bussiness logic has occurred and before
	414	+ # going back to sleep.
387	415	#
388		- def _run_cleanup(self, cache_size):
389		- context = self.context
390		- artifacts = context.artifactcache
	416	+ def _sched(self):
391	417
392		- if not artifacts.has_quota_exceeded():
393		- return
	418	+ if not self.terminated:
	419	+
	420	+ #
	421	+ # Try the cache management jobs
	422	+ #
	423	+ self._sched_cleanup_job()
	424	+ self._sched_cache_size_job()
	425	+
	426	+ #
	427	+ # Run as many jobs as the queues can handle for the
	428	+ # available resources
	429	+ #
	430	+ self._sched_queue_jobs()
394	431
395		- job = CleanupJob(self, _ACTION_NAME_CLEANUP, 'cleanup/cleanup',
396		- resources=[ResourceType.CACHE,
397		- ResourceType.PROCESS],
398		- exclusive_resources=[ResourceType.CACHE])
399		- self._schedule_jobs([job])
	432	+ #
	433	+ # If nothing is ticking then bail out
	434	+ #
	435	+ if not self._active_jobs:
	436	+ self.loop.stop()
400	437
401	438	# _suspend_jobs()
402	439	#
...	...	@@ -406,7 +443,7 @@ class Scheduler():
406	443	if not self.suspended:
407	444	self._suspendtime = datetime.datetime.now()
408	445	self.suspended = True
409		- for job in self.active_jobs:
	446	+ for job in self._active_jobs:
410	447	job.suspend()
411	448
412	449	# _resume_jobs()
...	...	@@ -415,7 +452,7 @@ class Scheduler():
415	452	#
416	453	def _resume_jobs(self):
417	454	if self.suspended:
418		- for job in self.active_jobs:
	455	+ for job in self._active_jobs:
419	456	job.resume()
420	457	self.suspended = False
421	458	self._starttime += (datetime.datetime.now() - self._suspendtime)
...	...	@@ -488,19 +525,16 @@ class Scheduler():
488	525	wait_limit = 20.0
489	526
490	527	# First tell all jobs to terminate
491		- for job in self.active_jobs:
	528	+ for job in self._active_jobs:
492	529	job.terminate()
493	530
494	531	# Now wait for them to really terminate
495		- for job in self.active_jobs:
	532	+ for job in self._active_jobs:
496	533	elapsed = datetime.datetime.now() - wait_start
497	534	timeout = max(wait_limit - elapsed.total_seconds(), 0.0)
498	535	if not job.terminate_wait(timeout):
499	536	job.kill()
500	537
501		- # Clear out the waiting jobs
502		- self.waiting_jobs = []
503		-
504	538	# Regular timeout for driving status in the UI
505	539	def _tick(self):
506	540	elapsed = self.elapsed_time()

setup.py

---

...	...	@@ -124,7 +124,7 @@ def list_man_pages():
124	124	# So screw it, lets just use an env var.
125	125	bst_install_entry_points = {
126	126	'console_scripts': [
127		- 'bst-artifact-server = buildstream._artifactcache.casserver:server_main'
	127	+ 'bst-artifact-server = buildstream._cas.casserver:server_main'
128	128	],
129	129	}
130	130

tests/frontend/order.py

---

...	...	@@ -12,7 +12,21 @@ DATA_DIR = os.path.join(
12	12	)
13	13
14	14
15		-def create_element(repo, name, path, dependencies, ref=None):
	15	+# create_element()
	16	+#
	17	+# Args:
	18	+# project (str): The project directory where testing is happening
	19	+# name (str): The element name to create
	20	+# dependencies (list): The list of dependencies to dump into YAML format
	21	+#
	22	+# Returns:
	23	+# (Repo): The corresponding git repository created for the element
	24	+def create_element(project, name, dependencies):
	25	+ dev_files_path = os.path.join(project, 'files', 'dev-files')
	26	+ element_path = os.path.join(project, 'elements')
	27	+ repo = create_repo('git', project, "{}-repo".format(name))
	28	+ ref = repo.create(dev_files_path)
	29	+
16	30	element = {
17	31	'kind': 'import',
18	32	'sources': [
...	...	@@ -20,7 +34,9 @@ def create_element(repo, name, path, dependencies, ref=None):
20	34	],
21	35	'depends': dependencies
22	36	}
23		- _yaml.dump(element, os.path.join(path, name))
	37	+ _yaml.dump(element, os.path.join(element_path, name))
	38	+
	39	+ return repo
24	40
25	41
26	42	# This tests a variety of scenarios and checks that the order in
...	...	@@ -59,18 +75,6 @@ def create_element(repo, name, path, dependencies, ref=None):
59	75	@pytest.mark.parametrize("operation", [('show'), ('fetch'), ('build')])
60	76	def test_order(cli, datafiles, tmpdir, operation, target, template, expected):
61	77	project = os.path.join(datafiles.dirname, datafiles.basename)
62		- dev_files_path = os.path.join(project, 'files', 'dev-files')
63		- element_path = os.path.join(project, 'elements')
64		-
65		- # FIXME: Remove this when the test passes reliably.
66		- #
67		- # There is no reason why the order should not
68		- # be preserved when the builders is set to 1,
69		- # the scheduler queue processing still seems to
70		- # be losing the order.
71		- #
72		- if operation == 'build':
73		- pytest.skip("FIXME: This still only sometimes passes")
74	78
75	79	# Configure to only allow one fetcher at a time, make it easy to
76	80	# determine what is being planned in what order.
...	...	@@ -84,11 +88,8 @@ def test_order(cli, datafiles, tmpdir, operation, target, template, expected):
84	88	# Build the project from the template, make import elements
85	89	# all with the same repo
86	90	#
87		- repo = create_repo('git', str(tmpdir))
88		- ref = repo.create(dev_files_path)
89	91	for element, dependencies in template.items():
90		- create_element(repo, element, element_path, dependencies, ref=ref)
91		- repo.add_commit()
	92	+ create_element(project, element, dependencies)
92	93
93	94	# Run test and collect results
94	95	if operation == 'show':

—
View it on GitLab.
You're receiving this email because of your account on gitlab.com. If you'd like to receive fewer emails, you can adjust your notification settings.