I'm writing a simple networking framework for Clojure using Java's new I/O package. It manages a pool of "selector agents", each of which holds a Selector.
I defined a dispatch action to for the selector agent. This action blocks on a call to selector.select(). When that returns, the selector agent iterates over the selected keys and performs I/O. When I/O is completed, the selector agent send's itself the dispatch action using send-off, effectively looping on calls to selector.select().
When I want to add a new channel or change a channel's interest ops, I send the selector agent the appropriate action and then unblock the selector (it's blocked on select(), remember?). This ensures that (send-off selector-agent dispatch) in the selector agent is executed after (send selector-agent add-channel channel).
I thought this would be bullet-proof since the call to send-off is performed before the selector waking up, and thus, before the selector agent send itself the dispatch action. However, this yields inconsistent behavior. Sometimes, the dispatch action occurs first and sometimes it doesn't.
My understanding is that it's not guaranteed that agents execute actions in the exact order they were sent when they come from multiple threads (i.e. send and send-off are not synchronous as far as queuing actions is concerned).
Is this correct?
send and send-off guarantee that actions will be placed on the Agent's queue in the order they are sent, within a single thread. Updating the Agent's queue happens synchronously.
I expect you have a simple race condition, although I can't identify it from the description.
You are absolutely right. Actions coming from the same thread will be executed in the same order, as they were submitted. But you cannot make any assumptions about execution order of actions, that come from different threads.
send and send off are built for asynchronous state changes.
if you need synchronous updates then atoms are likely your best tool.
since you need to preserve the order of requests you may have to use another data structure within a concurrency object (atom) can be syncrounoustly updated. It may work to put a persistent-queue inside an atom and have all your threads synchronousness add to that queue while your consumers synchronously pull entries from it.
here is the super brief decission chart:
more than one and synchronous: use a ref
asynchronous one: use an agent
asynchronous more than one: agents within a dosync
synchronous and only one: use an agent.
Related
I'm creating an async gRPC server in C++. One of the methods streams data from the server to clients - it's used to send data updates to clients. The frequency of the data updates isn't predictable. They could be nearly continuous or as infrequent as once per hour. The model used in the gRPC example with the "CallData" class and the CREATE/PROCESS/FINISH states doesn't seem like it would work very well for that. I've seen an example that shows how to create a 'polling' loop that sleeps for some time and then wakes up to check for new data, but that doesn't seem very efficient.
Is there another way to do this? If I use the "CallData" method can it block in the 'PROCESS' state until there's data (which probably wouldn't be my first choice)? Or better, can I structure my code so I can notify a gRPC handler when data is available?
Any ideas or examples would be appreciated.
In a server-side streaming example, you probably need more states, because you need to track whether there is currently a write already in progress. I would add two states, one called WRITE_PENDING that is used when a write is in progress, and another called WRITABLE that is used when a new message can be sent immediately. When a new message is produced, if you are in state WRITABLE, you can send immediately and go into state WRITE_PENDING, but if you are in state WRITE_PENDING, then the newly produced message needs to go into a queue to be sent after the current write finishes. When a write finishes, if the queue is non-empty, you can grab the next message from the queue and immediately start a write for it; otherwise, you can just go into state WRITABLE and wait for another message to be produced.
There should be no need to block here, and you probably don't want to do that anyway, because it would tie up a thread that should otherwise be polling the completion queue. If all of your threads wind up blocked that way, you will be blind to new events (such as new calls coming in).
An alternative here would be to use the C++ sync API, which is much easier to use. In that case, you can simply write straight-line blocking code. But the cost is that it creates one thread on the server for each in-progress call, so it may not be feasible, depending on the amount of traffic you're handling.
I hope this information is helpful!
My application is futures-based with async/await, and has the following structure within one of its components:
a "manager", which is responsible for starting/stopping/restarting "workers", based both on external input and on the current state of "workers";
a dynamic set of "workers", which perform some continuous work, but may fail or be stopped externally.
A worker is just a spawned task which does some I/O work. Internally it is a loop which is intended to be infinite, but it may exit early due to errors or other reasons, and in this case the worker must be restarted from scratch by the manager.
The manager is implemented as a loop which awaits on several channels, including one returned by async_std::stream::interval, which essentially makes the manager into a poller - and indeed, I need this because I do need to poll some Mutex-protected external state. Based on this state, the manager, among everything else, creates or destroys its workers.
Additionally, the manager stores a set of async_std::task::JoinHandles representing live workers, and it uses these handles to check whether any workers has exited, restarting them if so. (BTW, I do this currently using select(handle, future::ready()), which is totally suboptimal because it relies on the select implementation detail, specifically that it polls the left future first. I couldn't find a better way of doing it; something like race() would make more sense, but race() consumes both futures, which won't work for me because I don't want to lose the JoinHandle if it is not ready. This is a matter for another question, though.)
You can see that in this design workers can only be restarted when the next poll "tick" in the manager occurs. However, I don't want to use a too small interval for polling, because in most cases polling just wastes CPU cycles. Large intervals, however, can delay restarting a failed/canceled worker by too much, leading to undesired latencies. Therefore, I though I'd set up another channel of ()s back from each worker to the manager, which I'd add to the main manager loop, so when a worker stops due to an error or otherwise, it will first send a message to its channel, resulting in the manager being woken up earlier than the next poll in order to restart the worker right away.
Unfortunately, with any kinds of channels this might result in more polls than needed, in case two or more workers stop at approximately the same time (which due to the nature of my application, is somewhat likely to happen). In such case it would make sense to only run the manager loop once, handling all of the stopped workers, but with channels it will necessarily result in the number of polls equal to the number of stopped workers, even if additional polls don't do anything.
Therefore, my question is: how do I notify the manager from its workers that they are finished, without resulting in extra polls in the manager? I've tried the following things:
As explained above, regular unbounded channels just won't work.
I thought that maybe bounded channels could work - if I used a channel with capacity 0, and there was a way to try and send a message into it but just drop the message if the channel is full (like the offer() method on Java's BlockingQueue), this seemingly would solve the problem. Unfortunately, the channels API, while providing such a method (try_send() seems to be like it), also has this property of having capacity larger than or equal to the number of senders, which means it can't really be used for such notifications.
Some kind of atomic or a mutex-protected boolean flag also look as if it could work, but there is no atomic or mutex API which would provide a future to wait on, and would also require polling.
Restructure the manager implementation to include JoinHandles into the main select somehow. It might do the trick, but it would result in large refactoring which I'm unwilling to make at this point. If there is a way to do what I want without this refactoring, I'd like to use that first.
I guess some kind of combination of atomics and channels might work, something like setting an atomic flag and sending a message, and then skipping any extra notifications in the manager based on the flag (which is flipped back to off after processing one notification), but this also seems like a complex approach, and I wonder if anything simpler is possible.
I recommend using the FuturesUnordered type from the futures crate. This collection allows you to push many futures of the same type into a collection and wait for any one of them to complete at once.
It implements Stream, so if you import StreamExt, you can use unordered.next() to obtain a future that completes once any future in the collection completes.
If you also need to wait for a timeout or mutex etc., you can use select to create a future that completes once either the timeout or one of the join handles completes. The future returned by next() implements Unpin, so it is usable with select without problems.
Let's say that I have a switch statement in my thread function that evaluates for triggered events. Each case is a different event. Is it better to put the call to ResetEvent at the end of the case, or at the beginning? It seems to me that it should go at the end, so that the event cannot be triggered again, until the thread has finished processing the previous event. IF it is placed at the beginning, the event could be triggered again, while being processed.
Yes. think that is the way to go. Create a manual reset event (second parameter of CreateEvent API) so that event is not automatically reset after setting it.
If you handle incoming traffic using a single Event object (implying you have no inbound queue), you will miss events. Is this really what you want?
If you want to catch all events, a full-blown producer-consumer queue wouold be a better bet. Reference implementation for Boost.Thread here.
One problem that comes up time and
again with multi-threaded code is how
to transfer data from one thread to
another. For example, one common way
to parallelize a serial algorithm is
to split it into independent chunks
and make a pipeline — each stage in
the pipeline can be run on a separate
thread, and each stage adds the data
to the input queue for the next stage
when it's done. For this to work
properly, the input queue needs to be
written so that data can safely be
added by one thread and removed by
another thread without corrupting the
data structure.
Ours is a huge project. I need to call certain functions in my code asynchronously to avoid some circular function calls. Upon receiving a specific input, I can call my function asynchronously either by using Event or Timer.
Which way is preferable considering Performance ?
Sending events to Event manager and handling them with an Event handler ? Or
Starting a timer and provide timeout handler ?
For pure performance, event-driven model will be better. Use timers only if you cannot rely on one or more of your events to get set in a timely way by the worker code, and so need a backup means by which to continue processing. This may be the case if your worker code makes external calls to a database or other remote service whose reliability or performance is unproven.
I'm using Qt framework which has by default non-blocking I/O to develop an application navigating through several web pages (online stores) and carrying out different actions on these pages. I'm "mapping" specific web page to a state machine which I use to navigate through this page.
This state machine has these transitions;
Connect, LogIn, Query, LogOut, Disconnect
and these states;
Start, Connecting, Connected, LoggingIn, LoggedIn, Querying, QueryDone, LoggingOut, LoggedOut, Disconnecting, Disconnected
Transitions from *ing to *ed states (Connecting->Connected), are due to LoadFinished asynchronous network events received from network object when currently requested url is loaded. Transitions from *ed to *ing states (Connected->LoggingIn) are due to events send by me.
I want to be able to send several events (commands) to this machine (like Connect, LogIn, Query("productA"), Query("productB"), LogOut, LogIn, Query("productC"), LogOut, Disconnect) at once and have it process them. I don't want to block waiting for the machine to finish processing all events I sent to it. The problem is they have to be interleaved with the above mentioned network events informing machine about the url being downloaded. Without interleaving machine can't advance its state (and process my events) because advancing from *ing to *ed occurs only after receiving network type of event.
How can I achieve my design goal?
EDIT
The state machine I'm using has its own event loop and events are not queued in it so could be missed by machine if they come when the machine is busy.
Network I/O events are not posted directly to neither the state machine nor the event queue I'm using. They are posted to my code (handler) and I have to handle them. I can forward them as I wish but please have in mind remark no. 1.
Take a look at my answer to this question where I described my current design in details. The question is if and how can I improve this design by making it
More robust
Simpler
Sounds like you want the state machine to have an event queue. Queue up the events, start processing the first one, and when that completes pull the next event off the queue and start on that. So instead of the state machine being driven by the client code directly, it's driven by the queue.
This means that any logic which involves using the result of one transition in the next one has to be in the machine. For example, if the "login complete" page tells you where to go next. If that's not possible, then the event could perhaps include a callback which the machine can call, to return whatever it needs to know.
Asking this question I already had a working design which I didn't want to write about not to skew answers in any direction :) I'm going to describe in this pseudo answer what the design I have is.
In addition to the state machine I have a queue of events. Instead of posting events directly to the machine I'm placing them in the queue. There is however problem with network events which are asynchronous and come in any moment. If the queue is not empty and a network event comes I can't place it in the queue because the machine will be stuck waiting for it before processing events already in the queue. And the machine will wait forever because this network event is waiting behind all events placed in the queue earlier.
To overcome this problem I have two types of messages; normal and priority ones. Normal ones are those send by me and priority ones are all network ones. When I get network event I don't place it in the queue but instead I send it directly to the machine. This way it can finish its current task and progress to the next state before pulling the next event from the queue of events.
It works designed this way only because there is exactly 1:1 interleave of my events and network events. Because of this when the machine is waiting for a network event it's not busy doing anything (so it's ready to accept it and does not miss it) and vice versa - when the machine waits for my task it's only waiting for my task and not another network one.
I asked this question in hope for some more simple design than what I have now.
Strictly speaking, you can't. Because you only have state "Connecting", you don't know whether you need top login afterwards. You'd have to introduce a state "ConnectingWithIntentToLogin" to represent the result of a "Connect, then Login" event from the Start state.
Naturally there will be a lot of overlap between the "Connecting" and the "ConnectingWithIntentToLogin" states. This is most easily achieved by a state machine architecture that supports state hierarchies.
--- edit ---
Reading your later reactions, it's now clear what your actual problem is.
You do need extra state, obviously, whether that's ingrained in the FSM or outside it in a separate queue. Let's follow the model you prefer, with extra events in a queue. The rick here is that you're wondering how to "interleave" those queued events vis-a-vis the realtime events. You don't - events from the queue are actively extracted when entering specific states. In your case, those would be the "*ed" states like "Connected". Only when the queue is empty would you stay in the "Connected" state.
If you don't want to block, that means you don't care about the network replies. If on the other hand the replies interest you, you have to block waiting for them. Trying to design your FSM otherwise will quickly lead to your automaton's size reaching infinity.
How about moving the state machine to a different thread, i. e. QThread. I would implent a input queue in the state machine so I could send queries non blocking and a output queue to read the results of the queries. You could even call back a slotted function in your main thread via connect(...) if a result of a query arrives, Qt is thread safe in this regard.
This way your state machine could block as long as it needs without blocking your main program.
Sounds like you just want to do a list of blocking I/O in the background.
So have a thread execute:
while( !commands.empty() )
{
command = command.pop_back();
switch( command )
{
Connect:
DoBlockingConnect();
break;
...
}
}
NotifySenderDone();