Zsolt Haraszti | a341031 | 2016-09-18 23:29:04 -0700 | [diff] [blame] | 1 | # |
| 2 | # Copyright 2016 the original author or authors. |
| 3 | # |
| 4 | # Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | # you may not use this file except in compliance with the License. |
| 6 | # You may obtain a copy of the License at |
| 7 | # |
| 8 | # http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | # |
| 10 | # Unless required by applicable law or agreed to in writing, software |
| 11 | # distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | # See the License for the specific language governing permissions and |
| 14 | # limitations under the License. |
| 15 | # |
| 16 | |
| 17 | from structlog import get_logger |
| 18 | from twisted.internet import reactor |
| 19 | from twisted.internet.base import DelayedCall |
| 20 | from twisted.internet.defer import inlineCallbacks, DeferredList |
| 21 | |
| 22 | from asleep import asleep |
| 23 | |
| 24 | |
| 25 | class Leader(object): |
| 26 | """ |
| 27 | A single instance of this object shall exist across the whole cluster. |
| 28 | This is guaranteed by the coordinator which instantiates this class |
| 29 | only when it secured the leadership lock, as well as calling the halt() |
| 30 | method in cases it looses the leadership lock. |
| 31 | """ |
| 32 | |
| 33 | ASSIGNMENT_PREFIX = 'service/voltha/assignments/' |
| 34 | WORKLOAD_PREFIX = 'service/voltha/workload/' |
| 35 | |
| 36 | log = get_logger() |
| 37 | |
| 38 | # Public methods: |
| 39 | |
| 40 | def __init__(self, coordinator): |
| 41 | self.coorinator = coordinator |
| 42 | self.halted = False |
| 43 | self.soak_time = 5 # soak till membership/workload changes settle |
| 44 | |
| 45 | self.workload = [] |
| 46 | self.members = [] |
| 47 | self.reassignment_soak_timer = None |
| 48 | |
| 49 | @inlineCallbacks |
| 50 | def start(self): |
| 51 | self.log.info('leader-started') |
| 52 | yield self._validate_workload() |
| 53 | yield self._start_tracking_assignments() |
| 54 | |
| 55 | def halt(self): |
| 56 | """Suspend leadership duties immediately""" |
| 57 | self.log.info('leader-halted') |
| 58 | self.halted = True |
| 59 | |
| 60 | # any active cancellations, releases, etc., should happen here |
| 61 | if isinstance(self.reassignment_soak_timer, DelayedCall): |
| 62 | self.reassignment_soak_timer.cancel() |
| 63 | |
| 64 | # Private methods: |
| 65 | |
| 66 | @inlineCallbacks |
| 67 | def _validate_workload(self): |
| 68 | """ |
| 69 | Workload is defined as any k/v entries under the workload prefix |
| 70 | in consul. Under normal operation, only the leader shall edit the |
| 71 | workload list. But we make sure that in case an administrator |
| 72 | manually edits the workload, we react to that properly. |
| 73 | """ |
| 74 | |
| 75 | # TODO for now we simply generate a fixed number of fake entries |
| 76 | yield DeferredList([ |
| 77 | self.coorinator.kv_put( |
| 78 | self.WORKLOAD_PREFIX + 'device_group_%04d' % (i + 1), |
| 79 | 'placeholder for device group %d data' % (i + 1)) |
| 80 | for i in xrange(100) |
| 81 | ]) |
| 82 | |
| 83 | def _start_tracking_assignments(self): |
| 84 | """ |
| 85 | We must track both the cluster member list as well as the workload |
| 86 | list. Upon change in either, we must rerun our sharding algorithm |
| 87 | and reassign work as/if needed. |
| 88 | """ |
| 89 | reactor.callLater(0, self._track_workload, 0) |
| 90 | reactor.callLater(0, self._track_members, 0) |
| 91 | |
| 92 | @inlineCallbacks |
| 93 | def _track_workload(self, index): |
| 94 | |
| 95 | try: |
| 96 | (index, results) = yield self.coorinator.kv_get( |
| 97 | self.WORKLOAD_PREFIX, index=index, recurse=True) |
| 98 | |
| 99 | workload = [e['Key'] for e in results] |
| 100 | |
| 101 | if workload != self.workload: |
| 102 | self.log.info('workload-changed', workload=workload) |
| 103 | self.workload = workload |
| 104 | self._restart_reassignment_soak_timer() |
| 105 | |
| 106 | except Exception, e: |
| 107 | self.log.exception('workload-track-error', e=e) |
| 108 | yield asleep(1.0) # to prevent flood |
| 109 | |
| 110 | finally: |
| 111 | if not self.halted: |
| 112 | reactor.callLater(0, self._track_workload, index) |
| 113 | |
| 114 | @inlineCallbacks |
| 115 | def _track_members(self, index): |
| 116 | |
| 117 | def is_member(entry): |
| 118 | key = entry['Key'] |
| 119 | member_id = key[len(self.coorinator.MEMBERSHIP_PREFIX):] |
| 120 | return '/' not in member_id # otherwise it is a nested key |
| 121 | |
| 122 | try: |
| 123 | (index, results) = yield self.coorinator.kv_get( |
| 124 | self.coorinator.MEMBERSHIP_PREFIX, index=index, recurse=True) |
| 125 | |
| 126 | members = [e['Key'] for e in results if is_member(e)] |
| 127 | |
| 128 | if members != self.members: |
| 129 | self.log.info('membership-changed', members=members) |
| 130 | self.members = members |
| 131 | self._restart_reassignment_soak_timer() |
| 132 | |
| 133 | except Exception, e: |
| 134 | self.log.exception('members-track-error', e=e) |
| 135 | yield asleep(1.0) # to prevent flood |
| 136 | |
| 137 | finally: |
| 138 | if not self.halted: |
| 139 | reactor.callLater(0, self._track_members, index) |
| 140 | |
| 141 | def _restart_reassignment_soak_timer(self): |
| 142 | |
| 143 | if self.reassignment_soak_timer is not None: |
| 144 | assert isinstance(self.reassignment_soak_timer, DelayedCall) |
Zsolt Haraszti | 8dc1f5e | 2016-09-18 23:35:39 -0700 | [diff] [blame^] | 145 | if not self.reassignment_soak_timer.called: |
| 146 | self.reassignment_soak_timer.cancel() |
Zsolt Haraszti | a341031 | 2016-09-18 23:29:04 -0700 | [diff] [blame] | 147 | |
| 148 | self.reassignment_soak_timer = reactor.callLater( |
| 149 | self.soak_time, self._reassign_work) |
| 150 | |
| 151 | print self.reassignment_soak_timer |
| 152 | |
| 153 | @inlineCallbacks |
| 154 | def _reassign_work(self): |
| 155 | self.log.info('reassign-work') |
| 156 | yield None |
| 157 | |
| 158 | # TODO continue from here |
| 159 | |
| 160 | # Plan |
| 161 | # Step 1: collect current assignments from consul |
| 162 | # Step 2: calculate desired assignment from current members and |
| 163 | # workload list (e.g., using consistent hashing or any other |
| 164 | # algorithm |
| 165 | # Step 3: find the delta between the desired and actual assignments: |
| 166 | # these form two lists: |
| 167 | # 1. new assignments to be made |
| 168 | # 2. obsolete assignments to be revoked |
| 169 | # graceful handling may be desirable when moving existing |
| 170 | # assignment from existing member to another member (to make |
| 171 | # sure it is abandoned by old member before new takes charge) |
| 172 | # Step 4: orchestrate the assignment by adding/deleting(/locking) |
| 173 | # entries in consul |
| 174 | # |
| 175 | # We must make sure while we are working on this, we do not re-enter |
| 176 | # into same method! |