[JobRerunner] fix to also support jobs with dictionaries as parameters

As we found out, some jobs get a dictionary as parameter, and not just
an OpenStack ID as the code and comments suggest.

That makes things a bit complicated, as we still want to prevent jobs
running on the same id with the same callbacks from running, but need
to store the payload of the request (the dict) now as well for
re-execution.

Author: mutax

networking_nsxv3/common/synchronization.py

@@ -1,6 +1,8 @@
 """
 Synchronization - classes related concurrent execution scheduling and limits
 """
+from typing import Callable, Union, Optional
+
 import eventlet
 eventlet.monkey_patch()
 
@@ -66,12 +68,38 @@ def retry_next(self):
         return False
 
 
+CBPARAMS = Union[str, dict]
+
+
 class Runnable(object):
 
-    def __init__(self, idn, fn, priority=Priority.LOWEST):
+    def __init__(self, fnparams: CBPARAMS, fn: Callable[[CBPARAMS], None], priority=Priority.LOWEST):
         self.priority = priority
-        self.idn = idn
-        self.fn = fn
+        self._fnparams = fnparams
+        self._fn = fn
+
+        # contradicting to code comments, we sometimes get a dictionary
+        # as parameter for the callback. Some data from that
+        # dictionary is then actually used, e.g. revision and resource id.
+        # Not going to happen. Fixing that would require major
+        # refactoring of the rpc and realizer objects. Not going to happen.
+        #
+        # Instead, in these cases we will still use the openstack id and the
+        # name of the callback function to prevent parallel running of jobs.
+        # But when a job is submitted to the Rerunner, it will keep track of the
+        # dictionary contents if one exists and keep these for rerunning. So essentially
+        # we create subjobs.
+        # Realistically we need a better data structure than the active queue and Rerunner.
+
+        # fnparams is a str for most of the callbacks:
+        if isinstance(fnparams, str):
+            self.idn = fnparams
+        elif isinstance(fnparams, dict):
+            self.idn = fnparams['id']
+        else:
+            self.idn = str(fnparams)
+            LOG.warning('unexpected type %s for job parameters of %s', type(fnparams), fn.__name__)
+
         self._runcount = 0
         self._created = time.time()
         self._scheduled = None
@@ -81,7 +109,10 @@ def __init__(self, idn, fn, priority=Priority.LOWEST):
 
     @property
     def identifier(self) -> tuple:
-        return (self.idn, self.fn.__name__)
+        return (self.idn, self._fn.__name__)
+
+    def debugid(self)->str:
+        return str((self.idn, self._fn.__name__, str(self._fnparams)))
 
     def set_scheduled(self):
         """ called when we submit the job to the worker pool """
@@ -90,7 +121,7 @@ def set_scheduled(self):
         self._jobdone = None
         self._rescheduled = None
 
-    def set_start(self):
+    def _set_start(self):
         """ called in our wrapper when we actually start fn """
         # we need to reset the other timings
         # because we use the same job for rerunning
@@ -99,7 +130,7 @@ def set_start(self):
         self._rescheduled = None
         self._runcount += 1
 
-    def set_done(self):
+    def _set_done(self):
         """ called in our wrapper when fn returns """
         self._jobdone = time.time()
         self._rescheduled = None
@@ -132,6 +163,13 @@ def get_statline(self) -> str:
         return (f"timings for job {self} - runcount: {self._runcount} age: {age} "
                 f"scheduled: {scheduled} started: {started} runtime: {runtime} rescheduled: {rescheduled}")
 
+    def execute(self):
+        self._set_start()
+        try:
+            self._fn(self._fnparams)
+        finally:
+            self._set_done()
+
     def __repr__(self):
         # lets not just use the object id, maybe
         return str(self.identifier)
@@ -141,9 +179,11 @@ def __eq__(self, other):
         Note, the priority is not part of the comparison
         Thus if a runnable with higher priority is about to be
         added to the queue it will be rejected silently.
+        To prevent starvation, the queue will update the priority of the
+        existing element, in case it was lower than the item that was about to be added.
         """
         if isinstance(other, Runnable):
-            return (self.idn == other.idn and self.fn == other.fn)
+            return (self._fnparams  == other._fnparams  and self._fn == other._fn)
         else:
             return False
 
@@ -162,11 +202,6 @@ def __lt__(self, other):
 
         return self.priority < other.priority
 
-    def __hash__(self):
-        # with the original repr this is broken, as __eq__ takes the fn into
-        # account as well!
-        return hash((self.idn, self.fn))
-
 
 class UniqFiFoQueue(eventlet.queue.Queue):
     """
@@ -238,14 +273,147 @@ def _get(self):
         return heapq.heappop(self.queue)
 
 
+class JobList():
+    """
+        We are keeping track of the jobs and their parameters for the
+        JobRerunner, based on the type of job and openstack id.
+        There can be multiple similar jobs (same id, same method) but
+        with different parameters, if the parameter is a dict.
+        In these cases this JobList will keep track of the jobs, because
+        a job with the same additional parameters will compare equal.
+
+        add:
+
+        When a job is added the _runnables list can be empty, then the job
+        can run and we store it here for reference, with count 1.
+        If the list is not empty the job can either already exist or it
+        can be a job with different additional parameters.
+
+        If the job already exists, we increase the counter, and do not allow it to run.
+        If the job does not exist, we add it to the list with count = 1 and allow it to run.
+
+        done:
+
+        When a job is done, we will look at our list, and decrease the counter.
+        If the counter is 0, the job was not submitted a second time, and we can remove
+        it from the list.
+        If the counter is not 0, the job was requested to run again, and we keep it in the
+        list with the updated counter.
+
+        done will then choose a job from the list, that is supposed to run again,
+        remove it from the list and return it to the JobRerunner.
+
+        This might be the same job that was just finished or it could be a different one.
+        For now we will choose the oldest one based on age, which should be the same job that
+        was just done, but we might want to change that so we use a helper function for that
+        for now in the POC.
+
+    """
+    def __init__(self):
+        self._job_identifier: Optional[str] = None
+        self._runnables: List[tuple[int, Runnable]] = []
+
+    def __len__(self):
+        return len(self._runnables)
+
+    @property
+    def size(self):
+        return sum( count for count, _ in self._runnables)
+
+    def __repr__(self):
+        return f"Joblist: {self._job_identifier}, len={len(self)}, {self._runnables}"
+
+    def add(self, job:Runnable)->bool:
+
+        if self._job_identifier is not None:
+            if job.identifier != self._job_identifier:
+                raise ValueError("Can only add jobs of same type to a JobList")
+        else:
+            self._job_identifier = job.identifier
+
+        # search through our list and update the counter or append the job:
+        for index, (count, existing_job) in enumerate(self._runnables):
+            if job == existing_job:
+                count += 1
+                self._runnables[index] = (count, existing_job)
+                return False
+        else:
+            # this is the first of its kind, we can run it.
+            # note that a job that gets re-executed will be removed from
+            # the list (with the others of the same kind still present)
+            # so when it returns it will be the only one of its kind and
+            # can run. after it is finished a different one will be returned by done.
+            self._runnables.append((1, job))
+            return True
+
+    def _runnable_is_done(self, job:Runnable):
+        # search through our list and update the counter or remove the job:
+        for index, (count, existing_job) in enumerate(self._runnables):
+            if job == existing_job:
+                # we do not need this job with this parameters again, it is done,
+                # so we remove it from the list.
+                # Note: the list might not be empty!
+                LOG.debug("Job %s is done, updating JobList, request count was: %d", job.debugid(), count)
+                count -= 1
+                if count <= 0:
+                    if count < 0:
+                        LOG.warning("Job count in JobList was %d for %s", count, job.debugid())
+                    del self._runnables[index]
+                    return
+                self._runnables[index] = (count, existing_job)
+                return
+        else:
+            # we should never mark a job done, that was not added in the first place,
+            # if its not in the list, something is wrong
+            raise KeyError(f'No such job {job}')
+
+    def _runnable_pop_next(self) -> Optional[Runnable]:
+        """ find the next job to run and remove it from the list,
+            or return None if there is None to run.
+        """
+        if not self._runnables:
+            return None
+        # no job should currently be running, because we are in the "done" part of the
+        # workflow. So we can choose any job we like, remove it from the list and run it.
+        # when that job returns and needs no re-execution it will be removed from the list,
+        # and we will not pop it here again, so the next one in line will be returned.
+        # we always append to our list, so we can just pop the first one here and get the oldest.
+        count, job = self._runnables.pop()
+        LOG.debug("found job to run next with %d rerun requests open: %s", count, job)
+        return job
+
+    def done(self, job:Runnable)->Optional[Runnable]:
+        if self._job_identifier is not None:
+            if job.identifier != self._job_identifier:
+                raise ValueError("Can only remove jobs of same type from a JobList")
+        else:
+            raise KeyError("JobList is empty")
+
+        self._runnable_is_done(job)
+        return self._runnable_pop_next()
+
+    def get_count(self, job):
+
+        if self._job_identifier is not None:
+            if job.identifier != self._job_identifier:
+                raise ValueError("Can only remove jobs of same type from a JobList")
+
+        for count, existing_job in self._runnables:
+            if job == existing_job:
+                return count
+
+        return 0
+
+
 class JobRerunner():
     """ Thread save data structure to reschedule jobs when they are already running
 
     When a job is retrieved from the active queue and already running in a worker thread,
     there is a chance that another job for the same object is added to the active
-    queue and also started in a worker. While there then is some locking happening
-    that should prevent race conditions, this still blocks the worker thread(s),
-    which degrades performance and re-executes the job unnecessarily often.
+    queue and also started in a worker. To prevent race conditions, the worker threads
+    will use a lock to prevent two jobs running on the same objects, but this leads to
+    blocking of each of the affected workers, which degrades performance and also re-executes
+    the jobs unnecessarily often.
 
     To prevent this, we use this JobRerunner:
 
@@ -275,8 +443,8 @@ class JobRerunner():
     _lname_torerun = 'JobRerunner-torerun'
 
     def __init__(self):
-        self._running = dict()
-        self._to_rerun = collections.deque()
+        self._running: collections.defaultdict[(int, str),JobList] = collections.defaultdict(JobList)
+        self._to_rerun: collections.deque[Runnable] = collections.deque()
 
     def get_rerunnable(self) -> Runnable:
         # Let's also use the LockManager that is used for locking
@@ -296,26 +464,19 @@ def get_rerunnable(self) -> Runnable:
 
     def job_done(self, job: Runnable):
         LOG.debug("JobRerunner job_done called for %s", job)
+
         with LockManager.get_lock(self._lname_running):
-            count = self._running.get(job, 0)
 
-            if count == 1:
-                del self._running[job]
+            joblist: JobList = self._running[job.identifier]
+            next_job = joblist.done(job)
+            if not next_job:
+                del self._running[job.identifier]
                 LOG.info("JobRerunner (done, no reruns requested) %s", job.get_statline())
-            elif count > 1:
-                # we only allow exactly one job to run at a time,
-                # all jobs arriving later will increase the counter while
-                # the job is still running or they get re-queued.
-                # if a job is in the ready deque it will at some point
-                # re-appear and so we can forget about the counter.
-                LOG.info("JobRerunner (done, %d reruns requested) %s", count - 1, job.get_statline())
-                del self._running[job]
-                with LockManager.get_lock(self._lname_torerun):
-                    self._to_rerun.append(job)
             else:
-                # prevent the error from spreading
-                del self._running[job]
-                LOG.warning("JobRerunner job_done called too often for job %s", job)
+                # we got a job to rerun from our helper
+                LOG.info("JobRerunner (done, got rerun) done: %s next: %s", job.get_statline(), next_job.get_statline())
+                with LockManager.get_lock(self._lname_torerun):
+                    self._to_rerun.append(next_job)
 
     def add_job(self, job: Runnable) -> bool:
         """ Add job to list of jobs running/to be started or mark it for re-execution
@@ -327,22 +488,20 @@ def add_job(self, job: Runnable) -> bool:
 
         """
         with LockManager.get_lock(self._lname_running):
-            count = self._running.get(job, 0)
-            if count <= 0:
+            joblist:JobList = self._running[job.identifier]
+
+            if joblist.add(job):
                 # no job running, we can run the job
-                # if the counter is < 0 our accounting is wrong,
-                # so we fix it and run the job.
-                self._running[job] = 1
                 LOG.debug("JobRerunner no identical job is currently running, can start %s", job)
                 return True
-
-            self._running[job] = count + 1
-            LOG.debug("JobRerunner job %s already running, marked for rescheduling, count: %d ", job, count)
+            else:
+                count = joblist.get_count(job)
+                LOG.debug("JobRerunner job %s already running, marked for rescheduling, count: %d ", job, count)
 
             sum = 0
-            for job, scount in self._running.items():
-                sum += scount
-                LOG.debug("JobRerunner stat: job %s is running, submission count: %d", job, scount)
+            for identifier, joblist in self._running.items():
+                sum += joblist.size
+                LOG.debug("JobRerunner stat: job %s is running, submission count: %d", identifier, joblist.size)
 
         # let's log these as info for debugging, they should be sufficient in prod
         # to find issues with the JobRerunner:
@@ -379,9 +538,14 @@ def __init__(self, active_size=INFINITY, passive_size=INFINITY,
     def run(self, priority, ids, fn):
         """ Submit a job with priority
 
+        Note: the second parameter apparently sometimes is a dictionary, in contrast
+              to the documentation in the code!
+              Fixing this would requires too much refactoring at the moment -- mutax
+
         Keyword arguments:
         priority -- job priority of type Priority.class
-        ids -- list of IDs (identifiers) that will be passed to the 'fn'
+        ids -- list of OpenStack-IDs (identifiers) that will be passed to the 'fn'
+               OR list of dictionaries(!) of OpenStack objects (containing their id)
         fn -- a function about to be executed by the runner with an argument ID
         """
         if self._state != "started":
@@ -419,15 +583,13 @@ def _start(self):
                 # check if we are allowed to run it,
                 # if yes mark it as running and spawn it
                 if self._rerunner.add_job(job):
-                    LOG.info(MESSAGE.format("Processing", job.idn, Priority(job.priority).name, job.fn.__name__))
+                    LOG.info(MESSAGE.format("Processing", job.idn, Priority(job.priority).name, job))
 
                     # ideally we would be able to add a callback to the
                     # greenthread, but this is hidden in the pool, so
                     # let's wrap the function once more.
                     def wrap(rerun, ajob):
-                        ajob.set_start()
-                        ajob.fn(ajob.idn)
-                        ajob.set_done()
+                        ajob.execute()
                         rerun.job_done(ajob)
 
                     job.set_scheduled()