CSY/reowolf Changeset - 0d46914f7c2e · Centrum Wiskunde & Informatica (CWI)

Changeset - 0d46914f7c2e

Parent rev.

Child rev.

[Not reviewed]

0 6 0

MH - 4 years ago 2021-11-22 22:20:53
contact@maxhenger.nl

WIP on fixing bug where only partial sync region is available after error

6 files changed with 54 insertions and 32 deletions:

src/runtime2/connector.rs

src/runtime2/consensus.rs

src/runtime2/mod.rs

src/runtime2/scheduler.rs

src/runtime2/tests/mod.rs

src/runtime2/tests/sync_failure.rs

0 comments (0 inline, 0 general)

src/runtime2/connector.rs

➞

Show inline comments

@@ @@ -69,97 +69,97 @@ pub(crate) enum ConnectorScheduling { @@
     Later,              // Schedule for running, at some later point in time
     NotNow,             // Do not reschedule for running
     Exit,               // Connector has exited
+}
 pub(crate) struct ConnectorPDL {
     mode: Mode,
     eval_error: Option<EvalError>,
     tree: ExecTree,
     consensus: Consensus,
     last_finished_handled: Option<BranchId>,
+}
 // TODO: Remove remaining fields once 'fires()' is removed from language.
 struct ConnectorRunContext<'a> {
     branch_id: BranchId,
     consensus: &'a Consensus,
     prepared: PreparedStatement,
+}
 impl<'a> RunContext for ConnectorRunContext<'a>{
     fn performed_put(&mut self, _port: PortId) -> bool {
         return match self.prepared.take() {
             PreparedStatement::None => false,
             PreparedStatement::PerformedPut => true,
             taken => unreachable!("prepared statement is '{:?}' during 'performed_put()'", taken)
         };
+    }
     fn performed_get(&mut self, _port: PortId) -> Option<ValueGroup> {
         return match self.prepared.take() {
             PreparedStatement::None => None,
             PreparedStatement::PerformedGet(value) => Some(value),
             taken => unreachable!("prepared statement is '{:?}' during 'performed_get()'", taken),
         };
+    }
     fn fires(&mut self, port: PortId) -> Option<Value> {
         todo!("Remove fires() now");
         let port_id = PortIdLocal::new(port.0.u32_suffix);
         let annotation = self.consensus.get_annotation(self.branch_id, port_id);
         return annotation.expected_firing.map(|v| Value::Bool(v));
+    }
     fn created_channel(&mut self) -> Option<(Value, Value)> {
         return match self.prepared.take() {
             PreparedStatement::None => None,
             PreparedStatement::CreatedChannel(ports) => Some(ports),
-            taken => unreachable!("prepared statement is '{:?}' during 'created_channel)_'", taken),
+            taken => unreachable!("prepared statement is '{:?}' during 'created_channel()'", taken),
         };
+    }
     fn performed_fork(&mut self) -> Option<bool> {
         return match self.prepared.take() {
             PreparedStatement::None => None,
             PreparedStatement::ForkedExecution(path) => Some(path),
             taken => unreachable!("prepared statement is '{:?}' during 'performed_fork()'", taken),
         };
+    }
+}
 impl Connector for ConnectorPDL {
     fn run(&mut self, sched_ctx: SchedulerCtx, comp_ctx: &mut ComponentCtx) -> ConnectorScheduling {
         if let Some(scheduling) = self.handle_new_messages(comp_ctx) {
             return scheduling;
+        }
         match self.mode {
             Mode::Sync => {
                 // Run in sync mode
                 let scheduling = self.run_in_sync_mode(sched_ctx, comp_ctx);
                 // Handle any new finished branches
                 let mut iter_id = self.last_finished_handled.or(self.tree.get_queue_first(QueueKind::FinishedSync));
                 while let Some(branch_id) = iter_id {
                     iter_id = self.tree.get_queue_next(branch_id);
                     self.last_finished_handled = Some(branch_id);
                     if let Some(round_conclusion) = self.consensus.handle_new_finished_sync_branch(branch_id, comp_ctx) {
                         // Actually found a solution
                         return self.enter_non_sync_mode(round_conclusion, comp_ctx);
+                    }
                     self.last_finished_handled = Some(branch_id);
+                }
                 return scheduling;
             },
             Mode::NonSync => {
                 let scheduling = self.run_in_deterministic_mode(sched_ctx, comp_ctx);
                 return scheduling;
             },
             Mode::SyncError => {
                 let scheduling = self.run_in_sync_mode(sched_ctx, comp_ctx);
                 return scheduling;
             },
             Mode::Error => {
@@ @@ -341,108 +341,114 @@ impl ConnectorPDL { @@
                     if self.consensus.branch_can_receive(branch_id, &message) {
                         // This branch can receive the message, so we do the
                         // fork-and-receive dance
                         let receiving_branch_id = self.tree.fork_branch(branch_id);
                         let branch = &mut self.tree[receiving_branch_id];
                         branch.awaiting_port = PortIdLocal::new_invalid();
                         branch.prepared = PreparedStatement::PerformedGet(message.content.clone());
                         self.consensus.notify_of_new_branch(branch_id, receiving_branch_id);
                         self.consensus.notify_of_received_message(receiving_branch_id, &message, comp_ctx);
                         self.tree.push_into_queue(QueueKind::Runnable, receiving_branch_id);
                         any_message_received = true;
+                    }
+                }
                 if any_message_received {
                     return ConnectorScheduling::Immediate;
+                }
+            }
             EvalContinuation::SyncBlockEnd => {
                 let consistency = self.consensus.notify_of_finished_branch(branch_id);
                 if consistency == Consistency::Valid {
                     branch.sync_state = SpeculativeState::ReachedSyncEnd;
                     self.tree.push_into_queue(QueueKind::FinishedSync, branch_id);
                 } else {
                     branch.sync_state = SpeculativeState::Inconsistent;
+                }
             },
             EvalContinuation::NewFork => {
                 // Like the `NewChannel` result. This means we're setting up
                 // a branch and putting a marker inside the RunContext for the
                 // next time we run the PDL code
                 let left_id = branch_id;
                 let right_id = self.tree.fork_branch(left_id);
                 self.consensus.notify_of_new_branch(left_id, right_id);
                 self.tree.push_into_queue(QueueKind::Runnable, left_id);
                 self.tree.push_into_queue(QueueKind::Runnable, right_id);
                 let left_branch = &mut self.tree[left_id];
                 left_branch.prepared = PreparedStatement::ForkedExecution(true);
                 let right_branch = &mut self.tree[right_id];
                 right_branch.prepared = PreparedStatement::ForkedExecution(false);
+            }
             EvalContinuation::Put(port_id, content) => {
                 // Branch is attempting to send data
                 let port_id = PortIdLocal::new(port_id.0.u32_suffix);
                 let (sync_header, data_header) = self.consensus.handle_message_to_send(branch_id, port_id, &content, comp_ctx);
                 if let Err(_) = comp_ctx.submit_message(Message::Data(DataMessage {
                     sync_header, data_header,
                     content,
                 })) {
                     // We don't own the port
                     let pd = &sched_ctx.runtime.protocol_description;
                     let eval_error = branch.code_state.new_error_at_expr(
                         &pd.modules, &pd.heap,
                         String::from("attempted to 'put' on port that is no longer owned")
                     );
                     self.eval_error = Some(eval_error);
                     self.mode = Mode::SyncError;
                 let message = DataMessage{ sync_header, data_header, content };
                 match comp_ctx.submit_message(Message::Data(message)) {
                     Ok(_) => {
                         // Message is underway
                         branch.prepared = PreparedStatement::PerformedPut;
                         self.tree.push_into_queue(QueueKind::Runnable, branch_id);
                         return ConnectorScheduling::Immediate;
                     },
                     Err(_) => {
                         // We don't own the port
                         let pd = &sched_ctx.runtime.protocol_description;
                         let eval_error = branch.code_state.new_error_at_expr(
                             &pd.modules, &pd.heap,
                             String::from("attempted to 'put' on port that is no longer owned")
                         );
                         self.eval_error = Some(eval_error);
                         self.mode = Mode::SyncError;
+                    }
+                }
                 branch.prepared = PreparedStatement::PerformedPut;
                 self.tree.push_into_queue(QueueKind::Runnable, branch_id);
                 return ConnectorScheduling::Immediate;
             },
             _ => unreachable!("unexpected run result {:?} in sync mode", run_result),
+        }
         // If here then the run result did not require a particular action. We
         // return whether we have more active branches to run or not.
         if self.tree.queue_is_empty(QueueKind::Runnable) {
             return ConnectorScheduling::NotNow;
         } else {
             return ConnectorScheduling::Later;
+        }
+    }
     pub fn run_in_deterministic_mode(&mut self, sched_ctx: SchedulerCtx, comp_ctx: &mut ComponentCtx) -> ConnectorScheduling {
         debug_assert!(!self.tree.is_in_sync() && !self.consensus.is_in_sync());
         let branch = self.tree.base_branch_mut();
         debug_assert!(branch.sync_state == SpeculativeState::RunningNonSync);
         let mut run_context = ConnectorRunContext{
             branch_id: branch.id,
             consensus: &self.consensus,
             prepared: branch.prepared.take(),
         };
         let run_result = Self::run_prompt(&mut branch.code_state, &sched_ctx.runtime.protocol_description, &mut run_context);
         if let Err(eval_error) = run_result {
             comp_ctx.push_error(eval_error);
             return ConnectorScheduling::Exit
+        }
         let run_result = run_result.unwrap();
         match run_result {
             EvalContinuation::ComponentTerminated => {
                 branch.sync_state = SpeculativeState::Finished;
                 return ConnectorScheduling::Exit;
             },
             EvalContinuation::SyncBlockStart => {
                 comp_ctx.notify_sync_start();
                 let sync_branch_id = self.tree.start_sync();
                 debug_assert!(self.last_finished_handled.is_none());
                 self.consensus.start_sync(comp_ctx);
                 self.consensus.notify_of_new_branch(BranchId::new_invalid(), sync_branch_id);
                 self.tree.push_into_queue(QueueKind::Runnable, sync_branch_id);

src/runtime2/consensus.rs

➞

Show inline comments

@@ @@ -141,186 +141,200 @@ impl Consensus { @@
         self.highest_connector_id = ctx.id;
+    }
     /// Notifies the consensus algorithm that a new branch has appeared. Must be
     /// called for each forked branch in the execution tree.
     pub fn notify_of_new_branch(&mut self, parent_branch_id: BranchId, new_branch_id: BranchId) {
         // If called correctly. Then each time we are notified the new branch's
         // index is the length in `branch_annotations`.
         debug_assert!(self.branch_annotations.len() == new_branch_id.index as usize);
         let parent_branch_annotations = &self.branch_annotations[parent_branch_id.index as usize];
         let new_marker = BranchMarker::new(self.branch_markers.len() as u32);
         let new_branch_annotations = BranchAnnotation{
             channel_mapping: parent_branch_annotations.channel_mapping.clone(),
             cur_marker: new_marker,
         };
         self.branch_annotations.push(new_branch_annotations);
         self.branch_markers.push(new_branch_id);
+    }
     /// Notifies the consensus algorithm that a particular branch has
     /// encountered an unrecoverable error.
     pub fn notify_of_fatal_branch(&mut self, failed_branch_id: BranchId, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
         debug_assert!(self.is_in_sync());
         // Check for trivial case, where branch has not yet communicated within
         // the consensus algorithm
         let branch = &self.branch_annotations[failed_branch_id.index as usize];
         if branch.channel_mapping.iter().all(|v| v.registered_id.is_none()) {
             return Some(RoundConclusion::Failure);
+        }
         // We need to go through the hassle of notifying all participants in the
         // sync round that we've encountered an error.
         // --- notify leader
         let maybe_conclusion = self.send_to_leader_or_handle_as_leader(SyncCompContent::LocalFailure, ctx);
         // --- initiate discovery wave (to let leader know about all components)
         self.handled_wave = true;
         for mapping in &self.branch_annotations[0].channel_mapping {
             let channel_id = mapping.channel_id;
             let port_info = ctx.get_port_by_channel_id(channel_id).unwrap();
             let message = SyncPortMessage{
                 sync_header: self.create_sync_header(ctx),
                 source_port: port_info.self_id,
                 target_port: port_info.peer_id,
                 content: SyncPortContent::NotificationWave,
             };
             ctx.submit_message(Message::SyncPort(message));
             // Note: submitting the message might fail. But we're attempting to
             // handle the error anyway.
             // TODO: Think about this a second time: how do we make sure the
             //  entire network will fail if we reach this condition
             let _unused = ctx.submit_message(Message::SyncPort(message));
+        }
         return maybe_conclusion;
+    }
     /// Notifies the consensus algorithm that a branch has reached the end of
     /// the sync block. A final check for consistency will be performed that the
     /// caller has to handle. Note that
     pub fn notify_of_finished_branch(&self, branch_id: BranchId) -> Consistency {
         debug_assert!(self.is_in_sync());
         let branch = &self.branch_annotations[branch_id.index as usize];
         for mapping in &branch.channel_mapping {
             match mapping.expected_firing {
                 Some(expected) => {
                     if expected != mapping.registered_id.is_some() {
                         // Inconsistent speculative state and actual state
                         debug_assert!(mapping.registered_id.is_none()); // because if we did fire on a silent port, we should've caught that earlier
                         return Consistency::Inconsistent;
+                    }
                 },
                 None => {},
+            }
+        }
         return Consistency::Valid;
+    }
     /// Notifies the consensus algorithm that a particular branch has assumed
     /// a speculative value for its port mapping.
     pub fn notify_of_speculative_mapping(&mut self, branch_id: BranchId, port_id: PortIdLocal, does_fire: bool, ctx: &ComponentCtx) -> Consistency {
         debug_assert!(self.is_in_sync());
         let port_desc = ctx.get_port_by_id(port_id).unwrap();
         let channel_id = port_desc.channel_id;
         let branch = &mut self.branch_annotations[branch_id.index as usize];
         for mapping in &mut branch.channel_mapping {
             if mapping.channel_id == channel_id {
                 match mapping.expected_firing {
                     None => {
                         // Not yet mapped, perform speculative mapping
                         mapping.expected_firing = Some(does_fire);
                         return Consistency::Valid;
                     },
                     Some(current) => {
                         // Already mapped
                         if current == does_fire {
                             return Consistency::Valid;
                         } else {
                             return Consistency::Inconsistent;
+                        }
+                    }
+                }
+            }
+        }
         unreachable!("notify_of_speculative_mapping called with unowned port");
+    }
     /// Generates a new local solution from a finished branch. If the component
     /// is not the leader of the sync region then it will be sent to the
     /// appropriate component. If it is the leader then there is a chance that
     /// this solution completes a global solution. In that case the solution
     /// branch ID will be returned.
     pub(crate) fn handle_new_finished_sync_branch(&mut self, branch_id: BranchId, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
         // Turn the port mapping into a local solution
         let source_mapping = &self.branch_annotations[branch_id.index as usize].channel_mapping;
         let mut target_mapping = Vec::with_capacity(source_mapping.len());
         for port in source_mapping {
             // Note: if the port is silent, and we've never communicated
             // over the port, then we need to do so now, to let the peer
             // component know about our sync leader state.
             let port_desc = ctx.get_port_by_channel_id(port.channel_id).unwrap();
             let self_port_id = port_desc.self_id;
             let peer_port_id = port_desc.peer_id;
             let channel_id = port_desc.channel_id;
             if !self.encountered_ports.contains(&self_port_id) {
-                ctx.submit_message(Message::SyncPort(SyncPortMessage {
+                let message = SyncPortMessage {
                     sync_header: SyncHeader{
                         sending_component_id: ctx.id,
                         highest_component_id: self.highest_connector_id,
                         sync_round: self.sync_round
                     },
                     source_port: self_port_id,
                     target_port: peer_port_id,
                     content: SyncPortContent::SilentPortNotification,
                 }));
                 self.encountered_ports.push(self_port_id);
                 };
                 match ctx.submit_message(Message::SyncPort(message)) {
                     Ok(_) => {
                         self.encountered_ports.push(self_port_id);
                     },
                     Err(_) => {
                         // Seems like we were done with this branch, but one of
                         // the silent ports (in scope) is actually closed
                         return self.notify_of_fatal_branch(branch_id, ctx);
+                    }
+                }
+            }
             target_mapping.push((
                 channel_id,
                 port.registered_id.unwrap_or(BranchMarker::new_invalid())
             ));
+        }
         let local_solution = LocalSolution{
             component: ctx.id,
             final_branch_id: branch_id,
             port_mapping: target_mapping,
         };
         let maybe_conclusion = self.send_to_leader_or_handle_as_leader(SyncCompContent::LocalSolution(local_solution), ctx);
         return maybe_conclusion;
+    }
     /// Notifies the consensus algorithm about the chosen branch to commit to
     /// memory (may be the invalid "start" branch)
     pub fn end_sync(&mut self, branch_id: BranchId, final_ports: &mut Vec<ComponentPortChange>) {
         debug_assert!(self.is_in_sync());
         // TODO: Handle sending and receiving ports
         // Set final ports
         let branch = &self.branch_annotations[branch_id.index as usize];
         // Clear out internal storage to defaults
         self.highest_connector_id = ConnectorId::new_invalid();
         self.branch_annotations.clear();
         self.branch_markers.clear();
         self.encountered_ports.clear();
         self.solution_combiner.clear();
         self.handled_wave = false;
         self.conclusion = None;
         self.ack_remaining = 0;
         // And modify persistent storage
         self.sync_round += 1;
         for peer in self.peers.iter_mut() {
             peer.encountered_this_round = false;
             peer.expected_sync_round += 1;
+        }
+    }
     // --- Handling messages
     /// Prepares a message for sending. Caller should have made sure that
@@ @@ -424,376 +438,378 @@ impl Consensus { @@
                 return Some(RoundConclusion::Failure);
             },
             SyncCompContent::Notification => {
                 // We were just interested in the sync header we handled above
                 return None;
+            }
+        }
+    }
     pub fn handle_new_sync_port_message(&mut self, message: SyncPortMessage, ctx: &mut ComponentCtx) {
         if !self.handle_received_sync_header(&message.sync_header, ctx) {
             return;
+        }
         debug_assert!(self.is_in_sync());
         debug_assert!(ctx.get_port_by_id(message.target_port).is_some());
         match message.content {
             SyncPortContent::SilentPortNotification => {
                 // The point here is to let us become part of the sync round and
                 // take note of the leader in case all of our ports are silent.
                 self.encountered_ports.push(message.target_port);
+            }
             SyncPortContent::NotificationWave => {
                 // Wave to discover everyone in the network, handling sync
                 // header takes care of leader discovery, here we need to make
                 // sure we propagate the wave
                 if self.handled_wave {
                     return;
+                }
                 self.handled_wave = true;
                 // Propagate wave to all peers except the one that has sent us
                 // the wave.
                 for mapping in &self.branch_annotations[0].channel_mapping {
                     let channel_id = mapping.channel_id;
                     let port_desc = ctx.get_port_by_channel_id(channel_id).unwrap();
                     if port_desc.self_id == message.target_port {
                         // Wave came from this port, no need to send one back
                         continue;
+                    }
                     let message = SyncPortMessage{
                         sync_header: self.create_sync_header(ctx),
                         source_port: port_desc.self_id,
                         target_port: port_desc.peer_id,
                         content: SyncPortContent::NotificationWave,
                     };
                     ctx.submit_message(Message::SyncPort(message)).unwrap();
                     // As with the other SyncPort where we throw away the
                     // result: we're dealing with an error here anyway
                     let _unused = ctx.submit_message(Message::SyncPort(message));
+                }
+            }
+        }
+    }
     pub fn handle_new_sync_control_message(&mut self, message: SyncControlMessage, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
         if message.in_response_to_sync_round < self.sync_round {
             // Old message
             return None
+        }
         match message.content {
             SyncControlContent::ChannelIsClosed(_) => {
                 return Some(RoundConclusion::Failure);
+            }
+        }
+    }
     pub fn notify_of_received_message(&mut self, branch_id: BranchId, message: &DataMessage, ctx: &ComponentCtx) {
         debug_assert!(self.branch_can_receive(branch_id, message));
         let target_port = ctx.get_port_by_id(message.data_header.target_port).unwrap();
         let branch = &mut self.branch_annotations[branch_id.index as usize];
         for mapping in &mut branch.channel_mapping {
             if mapping.channel_id == target_port.channel_id {
                 // Found the port in which the message should be inserted
                 mapping.registered_id = Some(message.data_header.new_mapping);
                 // Check for sent ports
                 debug_assert!(self.workspace_ports.is_empty());
                 find_ports_in_value_group(&message.content, &mut self.workspace_ports);
                 if !self.workspace_ports.is_empty() {
                     todo!("handle received ports");
                     self.workspace_ports.clear();
+                }
                 return;
+            }
+        }
         // If here, then the branch didn't actually own the port? Means the
         // caller made a mistake
         unreachable!("incorrect notify_of_received_message");
+    }
     /// Matches the mapping between the branch and the data message. If they
     /// match then the branch can receive the message.
     pub fn branch_can_receive(&self, branch_id: BranchId, message: &DataMessage) -> bool {
         if let Some(peer) = self.peers.iter().find(|v| v.id == message.sync_header.sending_component_id) {
             if message.sync_header.sync_round < peer.expected_sync_round {
                 return false;
+            }
+        }
         let annotation = &self.branch_annotations[branch_id.index as usize];
         for expected in &message.data_header.expected_mapping {
             // If we own the port, then we have an entry in the
             // annotation, check if the current mapping matches
             for current in &annotation.channel_mapping {
                 if expected.channel_id == current.channel_id {
                     if expected.registered_id != current.registered_id {
                         // IDs do not match, we cannot receive the
                         // message in this branch
                         return false;
+                    }
+                }
+            }
+        }
         return true;
+    }
     // --- Internal helpers
     fn handle_received_sync_header(&mut self, sync_header: &SyncHeader, ctx: &mut ComponentCtx) -> bool {
         debug_assert!(sync_header.sending_component_id != ctx.id); // not sending to ourselves
         if !self.handle_peer(sync_header) {
             // We can drop this package
             return false;
+        }
         if sync_header.highest_component_id > self.highest_connector_id {
             // Sender has higher component ID. So should be the target of our
             // messages. We should also let all of our peers know
             self.highest_connector_id = sync_header.highest_component_id;
             for peer in self.peers.iter() {
                 if peer.id == sync_header.sending_component_id || !peer.encountered_this_round {
                     // Don't need to send it to this one
                     continue
+                }
                 let message = SyncCompMessage {
                     sync_header: self.create_sync_header(ctx),
                     target_component_id: peer.id,
                     content: SyncCompContent::Notification,
                 };
                 ctx.submit_message(Message::SyncComp(message));
+                ctx.submit_message(Message::SyncComp(message)).unwrap(); // unwrap: sending to component instead of through channel
+            }
             // But also send our locally combined solution
             self.forward_local_data_to_new_leader(ctx);
         } else if sync_header.highest_component_id < self.highest_connector_id {
             // Sender has lower leader ID, so it should know about our higher
             // one.
             let message = SyncCompMessage {
                 sync_header: self.create_sync_header(ctx),
                 target_component_id: sync_header.sending_component_id,
                 content: SyncCompContent::Notification
             };
             ctx.submit_message(Message::SyncComp(message));
+            ctx.submit_message(Message::SyncComp(message)).unwrap(); // unwrap: sending to component instead of through channel
         } // else: exactly equal, so do nothing
         return true;
+    }
     /// Handles a (potentially new) peer. Returns `false` if the provided sync
     /// number is different then the expected one.
     fn handle_peer(&mut self, sync_header: &SyncHeader) -> bool {
         let position = self.peers.iter().position(|v| v.id == sync_header.sending_component_id);
         match position {
             Some(index) => {
                 let entry = &mut self.peers[index];
                 entry.encountered_this_round = true;
                 // TODO: Proper handling of potential overflow
                 if sync_header.sync_round >= entry.expected_sync_round {
                     entry.expected_sync_round = sync_header.sync_round;
                     return true;
                 } else {
                     return false;
+                }
             },
             None => {
                 self.peers.push(Peer{
                     id: sync_header.sending_component_id,
                     encountered_this_round: true,
                     expected_sync_round: sync_header.sync_round,
                 });
                 return true;
+            }
+        }
+    }
     /// Sends a message towards the leader, if already the leader then the
     /// message will be handled immediately.
     fn send_to_leader_or_handle_as_leader(&mut self, content: SyncCompContent, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
         if self.highest_connector_id == ctx.id {
             // We are the leader
             match content {
                 SyncCompContent::LocalFailure => {
                     if self.solution_combiner.mark_failure_and_check_for_global_failure() {
                         return self.handle_global_failure_as_leader(ctx);
+                    }
                 },
                 SyncCompContent::LocalSolution(local_solution) => {
                     if let Some(global_solution) = self.solution_combiner.add_solution_and_check_for_global_solution(local_solution) {
                         return self.handle_global_solution_as_leader(global_solution, ctx);
+                    }
                 },
                 SyncCompContent::PartialSolution(partial_solution) => {
                     if let Some(conclusion) = self.solution_combiner.combine(partial_solution) {
                         match conclusion {
                             LeaderConclusion::Solution(global_solution) => {
                                 return self.handle_global_solution_as_leader(global_solution, ctx);
                             },
                             LeaderConclusion::Failure => {
                                 return self.handle_global_failure_as_leader(ctx);
+                            }
+                        }
+                    }
                 },
                 SyncCompContent::Presence(component_id, presence) => {
                     if self.solution_combiner.add_presence_and_check_for_global_failure(component_id, &presence) {
                         return self.handle_global_failure_as_leader(ctx);
+                    }
                 },
                 SyncCompContent::AckFailure => {
                     debug_assert_eq!(Some(RoundConclusion::Failure), self.conclusion);
                     debug_assert!(self.ack_remaining > 0);
                     self.ack_remaining -= 1;
                     if self.ack_remaining == 0 {
                         return Some(RoundConclusion::Failure);
+                    }
+                }
                 SyncCompContent::Notification | SyncCompContent::GlobalSolution(_) |
                 SyncCompContent::GlobalFailure => {
                     unreachable!("unexpected message content for leader");
                 },
+            }
         } else {
             // Someone else is the leader
             let message = SyncCompMessage {
                 sync_header: self.create_sync_header(ctx),
                 target_component_id: self.highest_connector_id,
                 content,
             };
             ctx.submit_message(Message::SyncComp(message));
+            ctx.submit_message(Message::SyncComp(message)).unwrap(); // unwrap: sending to component instead of through channel
+        }
         return None;
+    }
     fn handle_global_solution_as_leader(&mut self, global_solution: GlobalSolution, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
         if self.conclusion.is_some() {
             return None;
+        }
         // Handle the global solution
         let mut my_final_branch_id = BranchId::new_invalid();
         for (connector_id, branch_id) in global_solution.component_branches.iter().copied() {
             if connector_id == ctx.id {
                 // This is our solution branch
                 my_final_branch_id = branch_id;
                 continue;
+            }
             let message = SyncCompMessage {
                 sync_header: self.create_sync_header(ctx),
                 target_component_id: connector_id,
                 content: SyncCompContent::GlobalSolution(global_solution.clone()),
             };
             ctx.submit_message(Message::SyncComp(message));
+            ctx.submit_message(Message::SyncComp(message)).unwrap(); // unwrap: sending to component instead of through channel
+        }
         debug_assert!(my_final_branch_id.is_valid());
         self.conclusion = Some(RoundConclusion::Success(my_final_branch_id));
         return Some(RoundConclusion::Success(my_final_branch_id));
+    }
     fn handle_global_failure_as_leader(&mut self, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
         debug_assert!(self.solution_combiner.failure_reported && self.solution_combiner.check_for_global_failure());
         if self.conclusion.is_some() {
             return None;
+        }
         // TODO: Performance
         let mut encountered = VecSet::new();
         for presence in &self.solution_combiner.presence {
             if presence.added_by != ctx.id {
                 // Did not add it ourselves
                 if encountered.push(presence.added_by) {
                     // Not yet sent a message
                     let message = SyncCompMessage{
                         sync_header: self.create_sync_header(ctx),
                         target_component_id: presence.added_by,
                         content: SyncCompContent::GlobalFailure,
                     };
                     ctx.submit_message(Message::SyncComp(message));
+                    ctx.submit_message(Message::SyncComp(message)).unwrap(); // unwrap: sending to component instead of through channel
+                }
+            }
+        }
         self.conclusion = Some(RoundConclusion::Failure);
         if encountered.is_empty() {
             // We don't have to wait on Acks
             return Some(RoundConclusion::Failure);
         } else {
             return None;
+        }
+    }
     #[inline]
     fn create_sync_header(&self, ctx: &ComponentCtx) -> SyncHeader {
         return SyncHeader{
             sending_component_id: ctx.id,
             highest_component_id: self.highest_connector_id,
             sync_round: self.sync_round,
+        }
+    }
     fn forward_local_data_to_new_leader(&mut self, ctx: &mut ComponentCtx) {
         debug_assert_ne!(self.highest_connector_id, ctx.id);
         if let Some(partial_solution) = self.solution_combiner.drain() {
             let message = SyncCompMessage {
                 sync_header: self.create_sync_header(ctx),
                 target_component_id: self.highest_connector_id,
                 content: SyncCompContent::PartialSolution(partial_solution),
             };
             ctx.submit_message(Message::SyncComp(message));
+            ctx.submit_message(Message::SyncComp(message)).unwrap(); // unwrap: sending to component instead of through channel
+        }
+    }
+}
 // -----------------------------------------------------------------------------
 // Solution storage and algorithms
 // -----------------------------------------------------------------------------
 // TODO: Remove all debug derives
 #[derive(Debug, Clone)]
 struct MatchedLocalSolution {
     final_branch_id: BranchId,
     channel_mapping: Vec<(ChannelId, BranchMarker)>,
     matches: Vec<ComponentMatches>,
+}
 #[derive(Debug, Clone)]
 struct ComponentMatches {
     target_id: ConnectorId,
     target_index: usize,
     match_indices: Vec<usize>, // of local solution in connector
+}
 #[derive(Debug, Clone)]
 struct ComponentPeer {
     target_id: ConnectorId,
     target_index: usize, // in array of global solution components
     involved_channels: Vec<ChannelId>,
+}
 #[derive(Debug, Clone)]
 struct ComponentLocalSolutions {
     component: ConnectorId,
     peers: Vec<ComponentPeer>,
     solutions: Vec<MatchedLocalSolution>,
     all_peers_present: bool,
+}
 #[derive(Debug, Clone)]
 struct ChannelPresence {
     added_by: ConnectorId,
     channel: ChannelId,
     both_sides_present: bool,
+}
 // TODO: Flatten? Flatten. Flatten everything.
 #[derive(Debug)]

src/runtime2/mod.rs

➞

Show inline comments

@@ @@ -386,93 +386,93 @@ impl ConnectorStore { @@
+        }
+    }
     /// Retrieves private part of connector - accessible by one thread at a
     /// time.
     fn get_private(&self, key: &ConnectorKey) -> &'static mut ScheduledConnector {
         unsafe {
             debug_assert!(!self.free.contains(&(key.index as usize)));
             let connector = self.connectors.get_mut(key.index as usize);
             debug_assert!(!connector.is_null());
             return &mut (**connector);
+        }
+    }
     /// Creates a new connector. Caller should ensure ports are set up correctly
     /// and the connector is queued for execution if needed.
     fn create(&mut self, connector: ConnectorVariant, initially_sleeping: bool) -> ConnectorKey {
         let mut connector = ScheduledConnector {
             connector,
             ctx: ComponentCtx::new_empty(),
             public: ConnectorPublic::new(initially_sleeping),
             router: ControlMessageHandler::new(),
             shutting_down: false,
         };
         let index;
         let key;
         if self.free.is_empty() {
             // No free entries, allocate new entry
             index = self.connectors.len();
             key = ConnectorKey{ index: index as u32 };
             connector.ctx.id = key.downcast();
             let connector = Box::into_raw(Box::new(connector));
             self.connectors.push(connector);
         } else {
             // Free spot available
             index = self.free.pop().unwrap();
             key = ConnectorKey{ index: index as u32 };
             connector.ctx.id = key.downcast();
             unsafe {
                 let target = self.connectors.get_mut(index);
                 std::ptr::write(*target, connector);
+            }
+        }
-        // println!("DEBUG [ global store  ] Created component at {}", key.index);
         println!("DEBUG [ global store  ] Created component at {}", key.index);
         return key;
+    }
     /// Destroys a connector. Caller should make sure it is not scheduled for
     /// execution. Otherwise one experiences "bad stuff" (tm).
     fn destroy(&mut self, key: ConnectorKey) {
         unsafe {
             let target = self.connectors.get_mut(key.index as usize);
             std::ptr::drop_in_place(*target);
             // Note: but not deallocating!
+        }
-        // println!("DEBUG [ global store  ] Destroyed component at {}", key.index);
         println!("DEBUG [ global store  ] Destroyed component at {}", key.index);
         self.free.push(key.index as usize);
+    }
+}
 impl Drop for ConnectorStore {
     fn drop(&mut self) {
         // Everything in the freelist already had its destructor called, so only
         // has to be deallocated
         for free_idx in self.free.iter().copied() {
             unsafe {
                 let memory = self.connectors.get_mut(free_idx);
                 let layout = std::alloc::Layout::for_value(&**memory);
                 std::alloc::dealloc(*memory as *mut u8, layout);
                 // mark as null for the remainder
                 *memory = std::ptr::null_mut();
+            }
+        }
         // With the deallocated stuff marked as null, clear the remainder that
         // is not null
         for idx in 0..self.connectors.len() {
             unsafe {
                 let memory = *self.connectors.get_mut(idx);
                 if !memory.is_null() {
                     let _ = Box::from_raw(memory); // take care of deallocation, bit dirty, but meh
+                }
+            }
+        }
+    }
+}
@@ \ No newline at end of file @@

src/runtime2/scheduler.rs

➞

Show inline comments

@@ @@ -360,101 +360,101 @@ impl Scheduler { @@
             scheduled.ctx.changed_in_sync = false; // reset flag
+        }
+    }
     fn try_go_to_sleep(&self, connector_key: ConnectorKey, connector: &mut ScheduledConnector) {
         debug_assert_eq!(connector_key.index, connector.ctx.id.0);
         debug_assert_eq!(connector.public.sleeping.load(Ordering::Acquire), false);
         // This is the running connector, and only the running connector may
         // decide it wants to sleep again.
         connector.public.sleeping.store(true, Ordering::Release);
         // But due to reordering we might have received messages from peers who
         // did not consider us sleeping. If so, then we wake ourselves again.
         if !connector.public.inbox.is_empty() {
             // Try to wake ourselves up (needed because someone might be trying
             // the exact same atomic compare-and-swap at this point in time)
             let should_wake_up_again = connector.public.sleeping
                 .compare_exchange(true, false, Ordering::SeqCst, Ordering::Acquire)
                 .is_ok();
             if should_wake_up_again {
                 self.runtime.push_work(connector_key)
+            }
+        }
+    }
     #[inline]
     fn get_message_target_port(message: &Message) -> Option<PortIdLocal> {
         match message {
             Message::Data(data) => return Some(data.data_header.target_port),
             Message::SyncComp(_) => {},
             Message::SyncPort(content) => return Some(content.target_port),
             Message::SyncControl(_) => return None,
             Message::Control(control) => {
                 match &control.content {
                     ControlContent::PortPeerChanged(port_id, _) => return Some(*port_id),
                     ControlContent::CloseChannel(port_id) => return Some(*port_id),
                     ControlContent::Ping | ControlContent::Ack => {},
+                }
             },
+        }
         return None
+    }
     // TODO: Remove, this is debugging stuff
     fn debug(&self, message: &str) {
-        // println!("DEBUG [thrd:{:02} conn:  ]: {}", self.scheduler_id, message);
         println!("DEBUG [thrd:{:02} conn:  ]: {}", self.scheduler_id, message);
+    }
     fn debug_conn(&self, conn: ConnectorId, message: &str) {
-        // println!("DEBUG [thrd:{:02} conn:{:02}]: {}", self.scheduler_id, conn.0, message);
         println!("DEBUG [thrd:{:02} conn:{:02}]: {}", self.scheduler_id, conn.0, message);
+    }
+}
 // -----------------------------------------------------------------------------
 // ComponentCtx
 // -----------------------------------------------------------------------------
 enum ComponentStateChange {
     CreatedComponent(ConnectorPDL, Vec<PortIdLocal>),
     CreatedPort(Port),
     ChangedPort(ComponentPortChange),
+}
 #[derive(Clone)]
 pub(crate) struct ComponentPortChange {
     pub is_acquired: bool, // otherwise: released
     pub port: Port,
+}
 /// The component context (better name may be invented). This was created
 /// because part of the component's state is managed by the scheduler, and part
 /// of it by the component itself. When the component starts a sync block or
 /// exits a sync block the partially managed state by both component and
 /// scheduler need to be exchanged.
 pub(crate) struct ComponentCtx {
     // Mostly managed by the scheduler
     pub(crate) id: ConnectorId,
     ports: Vec<Port>,
     inbox_messages: Vec<Message>,
     inbox_len_read: usize,
     // Submitted by the component
     is_in_sync: bool,
     changed_in_sync: bool,
     outbox: VecDeque<Message>,
     state_changes: VecDeque<ComponentStateChange>,
     // Workspaces that may be used by components to (generally) prevent
     // allocations. Be a good scout and leave it empty after you've used it.
     // TODO: Move to scheduler ctx, this is the wrong place
     pub workspace_ports: Vec<PortIdLocal>,
     pub workspace_branches: Vec<BranchId>,
+}
 impl ComponentCtx {
     pub(crate) fn new_empty() -> Self {
         return Self{
             id: ConnectorId::new_invalid(),
             ports: Vec::new(),

src/runtime2/tests/mod.rs

➞

Show inline comments

 mod network_shapes;
 mod api_component;
 mod speculation;
 mod data_transmission;
 mod sync_failure;
 use super::*;
 use crate::{PortId, ProtocolDescription};
 use crate::common::Id;
 use crate::protocol::eval::*;
 use crate::runtime2::native::{ApplicationSyncAction};
 // Generic testing constants, use when appropriate to simplify stress-testing
 // pub(crate) const NUM_THREADS: u32 = 3;     // number of threads in runtime
 // pub(crate) const NUM_INSTANCES: u32 = 7;   // number of test instances constructed
 // pub(crate) const NUM_LOOPS: u32 = 8;       // number of loops within a single test (not used by all tests)
 pub(crate) const NUM_THREADS: u32 = 1;
 pub(crate) const NUM_INSTANCES: u32 = 5;
 pub(crate) const NUM_LOOPS: u32 = 5;
 pub(crate) const NUM_THREADS: u32 = 2;
 pub(crate) const NUM_INSTANCES: u32 = 1;
 pub(crate) const NUM_LOOPS: u32 = 1;
 fn create_runtime(pdl: &str) -> Runtime {
     let protocol = ProtocolDescription::parse(pdl.as_bytes()).expect("parse pdl");
     let runtime = Runtime::new(NUM_THREADS, protocol);
     return runtime;
+}
 fn run_test_in_runtime<F: Fn(&mut ApplicationInterface)>(pdl: &str, constructor: F) {
     let protocol = ProtocolDescription::parse(pdl.as_bytes())
         .expect("parse PDL");
     let runtime = Runtime::new(NUM_THREADS, protocol);
     let mut api = runtime.create_interface();
     for _ in 0..NUM_INSTANCES {
         constructor(&mut api);
+    }
+}
 pub(crate) struct TestTimer {
     name: &'static str,
     started: std::time::Instant
+}
 impl TestTimer {
     pub(crate) fn new(name: &'static str) -> Self {
         Self{ name, started: std::time::Instant::now() }
+    }
+}
 impl Drop for TestTimer {
     fn drop(&mut self) {
         let delta = std::time::Instant::now() - self.started;
         let nanos = (delta.as_secs_f64() * 1_000_000.0) as u64;
         let millis = nanos / 1000;
         let nanos = nanos % 1000;
         println!("[{}] Took {:>4}.{:03} ms", self.name, millis, nanos);
+    }
+}

src/runtime2/tests/sync_failure.rs

➞

Show inline comments

@@ @@ -20,56 +20,56 @@ fn test_local_sync_failure() { @@
         u32[] only_allows_index_0 = { 1 };
         auto never_gonna_get = only_allows_index_0[1];
         while (true) sync {}
+    }
     ";
     // let thing = TestTimer::new("local_sync_failure");
     run_test_in_runtime(CODE, |api| {
         api.create_connector("", "immediate_failure_outside_sync", ValueGroup::new_stack(Vec::new()))
             .expect("create component");
         api.create_connector("", "immediate_failure_inside_sync", ValueGroup::new_stack(Vec::new()))
             .expect("create component");
     })
+}
 #[test]
 fn test_shared_sync_failure() {
     // Same as above. One of the components should fail, the other should follow
     // suit because it cannot complete a sync round. We intentionally have an
     // infinite loop in the while condition because we need at least two loops
     // for the last error to get picked up.
     const CODE: &'static str = "
     enum Location { BeforeSync, AfterPut, AfterGet, AfterSync, Never }
     primitive failing_at_location(in<bool> input, out<bool> output, Location loc) {
         u32[] failure_array = {};
         while (true) {
             if (loc == Location::BeforeSync) failure_array[0];
             sync {
                 put(output, true);
                 if (loc == Location::AfterPut) failure_array[0];
                 auto received = get(input);
                 assert(received);
                 if (loc == Location::AfterGet) failure_array[0];
+            }
             if (loc == Location::AfterSync) failure_array[0];
+        }
+    }
     composite constructor(Location loc) {
         channel output_a -> input_a;
         channel output_b -> input_b;
         new failing_at_location(input_a, output_b, Location::Never);
         new failing_at_location(input_b, output_a, loc);
+    }
     ";
     run_test_in_runtime(CODE, |api| {
-        for variant in 0..4 { // all `Location` enum variants, except `Never`.
+        for variant in 3..4 { // all `Location` enum variants, except `Never`.
             // Create the channels
             api.create_connector("", "constructor", ValueGroup::new_stack(vec![
                 Value::Enum(variant)
             ])).expect("create connector");
+        }
     })
+}
@@ \ No newline at end of file @@

0 comments (0 inline, 0 general)