Related
When I have code like this:
receive do
{:hello, msg} -> msg
end
And let's say I have N messages in my mailbox. Is the performance of finding this particular message O(1), O(N), or something in between?
Receive perform a linear scan of the message box and then returns the first one which matches. There is one exception (Since R14A)
OTP-8623 == compiler erts hipe stdlib ==
Receive statements that can only read out a newly created
reference are now specially optimized so that it will execute
in constant time regardless of the number of messages in the
receive queue for the process. That optimization will benefit
calls to gen_server:call(). (See gen:do_call/4 for an example
of a receive statement that will be optimized.)
So in your case it is O(N) operation.
Messaging in Erlang, hence Elixir, is "first in, first out". You browse them one by one and the first that's meet any clause in receive is handled. In worst case scenario you can choke up your messagebox.
The performance will grow linearly and in direct proportion to the number of elements in the mailbox, thus being O(N).
I'm still in the learning fase of Erlang, so I might be wrong, but this is how I understood a process' message queue.
A process could be in it's main receive loop, receiving certain types of messages, while later it could be put in a waiting loop to deal with a different kind of message in the second loop. If the process would receive messages intended for the first loop in the second loop, it would just put them in the queue, ignore them for the time being and only process those message that it can match against in the current loop it is in. Now if it would enter the first receive loop again, it would start from the beginning and again process the messages that it can match against.
Now my question would be, if this is how Erlang works and I understood this correctly, then what happens when a malicious process would send a bunch of messages that the process will never process. Will the queue eventually overflow, resulting in a crash for the process or how should I deal with this? I'll type out an example to illustrate what I mean.
Now if a malicious program would get a hold of the Pid and would go Pid ! {malicioudata, LotsOfData} repeatedly, would those messages be filtered out since they will never possibly be processed or would they just stack up in the queue?
startproc() -> firstloop(InitValues).
firstloop(Values) ->
receive
retrieveinformation ->
WaitingList=askforinformation(),
retrieveloop(WaitingList);
dostuff ->
NewValues=doingstuff(),
firstloop(NewValues);
sendmeyourdata ->
sendingdata(Values),
firstloop(Values)
end.
retrieveloop([],Values) -> firstloop(Values).
retrieveloop(WaitingList,Values) ->
receive
{hereismyinformation,Id,MyInfo} ->
NewValues=dosomethingwithinfo(Id,MyInfo),
retrieveloop(lists:remove(Id,1,WaitingList),NewValues);
end.
There is not a hard limit on message counts, and there is not a fixed amount of memory you are limited to, but you can certainly run out of memory if you have billions of messages (or a few super huge ones, maybe).
Long before you OOM because of a huge mailbox you will notice either selective receives taking a long time (not that "selective receive" is a good pattern to follow much of the time...) or innocently peek into a process mail queue and realized you've opened Pandora's Box in your terminal.
This is usually treated as a throttling and monitoring issue in the Erlang world. If you aren't able to keep up and your problem is parallelizable then you need more workers. If you are maxing out your hardware then you need more efficiency. If you are still maxing out your hardware, can't get any more, and you're still overwhelmed then you need to decide how to implement pushback or load shedding.
Unfortunately there is no "message queue overflow" and it's going to grow until VM crashes due to memory allocation error.
Solution is to drop any invalid messages in main loop, because you are not suppose to receive any of {hereismyinformation, _,_} nor one you get in askforinformation() due to blocking nature of your process.
startproc() -> firstloop(InitValues).
firstloop(Values) ->
receive
retrieveinformation ->
WaitingList=askforinformation(),
retrieveloop(WaitingList, Values); % i assume you meant that
dostuff ->
NewValues=doingstuff(),
firstloop(NewValues);
sendmeyourdata ->
sendingdata(Values),
firstloop(Values);
_ ->
firstloop(Values) % you can't get {hereismyinformation, _,_} here so we can drop any invalid message
end.
retrieveloop([],Values) -> firstloop(Values).
retrieveloop(WaitingList,Values) ->
receive
{hereismyinformation,Id,MyInfo} ->
NewValues=dosomethingwithinfo(Id,MyInfo),
retrieveloop(lists:remove(Id,1,WaitingList),NewValues);
end.
It's not really a problem with unexpected messages because it's easily avoidable but when process queue is growing faster than it's processed. For this specific problem there is a nice jobs framework for production systems.
Is there an Erlang/OTP pattern/library for the following problem(before I hack my own):
At the highest level, imagine there are three components(or processes?) such that A->B->C where -> means sends a message to.
B in terms of architecture is a composite process. It is composed of many unit processes(shown in khaki green below). Sometimes, the message chain goes from B1->B2->B3->C and sometimes it goes from B1->B4->B5->B6->B3->C.
What I would like to do is:
B can only accept the next message when all it's children processes are done i.e B receives a message I1 and depending on the message, it will choose one flow and finally C gets a message O1. Until that happens, B should not accept the message I2. This is to ensure ordering of messages so that O2 of I2 does not reach C before O1 of I1.
This has a few names. One is "dataflow" (as in "reactive programming" -- which is sort of an overblown ball of buzzwords if you look it up) and another is "signal simulation" (as in simulation of electrical signal switches). I am not aware of a framework for this in Erlang, because it is very straightforward to implement directly.
The issue of message ordering can be made to take care of itself, depending on how you want to write things. Erlang guarantees the ordering of message between two processes, so as long as messages travel in well-defined channels, this system-wide promise can be made to work for you. If you need some more interesting signal paths than straight lines you can force synch communication; though all Erlang message are asynchronous, you can introduce synchronous blocking on receive wherever you want.
If you want the "B constellation" to pass a message to C but only after its signal processing has completely run its route through the B's, you can make a signal manager which sends a message to B1, and blocks until it receives the output from B3, whence it passes the completed message on to C and checks its box for the next thing from A:
a_loop(B) ->
receive {in, Data} -> B ! Data end,
a_loop(B).
% Note the two receives here -- we are blocking for the end of processing based
% on the known Ref we send out and expect to receive back in a message match.
b_manager(B1, C) ->
Ref = make_ref(),
receive Data -> B1 ! {Ref, Data} end,
receive {Ref, Result} -> C ! Result end,
b_manager(B1, C).
b_1(B2) ->
receive
{Ref, Data} ->
Mod1 = do_processing(Data),
B2 ! {Ref, Mod1}
end,
b_1(B2).
% Here you have as many "b_#" processes as you need...
b_2(B) ->
receive
{Ref, Data} ->
Result = do_other_processing(Data),
B ! {Ref, Result}
end,
b_2(B).
c_loop() ->
receive Result -> stuff(Result) end,
c_loop().
Obviously I drastically simplified things -- as in this obviously doesn't include any concept of supervision -- I didn't even address how you would want to link these together (and with this little checking for liveness, you would need to spawn_link them so if anything dies they all die -- which is probably exactly what you want with the B subset anyway, so you can treat it as a single unit). Also, you may wind up needing a throttle in there somewhere (like at/before A, or in B). But basically speaking, this is a way of passing messages through in a way that makes B block until its segment of processing is finished.
There are other ways, like gen_event, but I find them to be less flexible than writing a actual simulation of a processing pipeline. As far as how to implement this -- I would make it a combination of OTP supervisors and gen_fsm, as these two components represent a nearly perfect parallel to signal processing components,which your system seems to be aimed at mimicking.
To discover what states you need in your gen_fsms and how you want to clump them together I would probably prototype in a very simplistic fashion in pure Erlang for a few hours, just to make sure I actually understand the problem, and then write my proper OTP supervisors and gen_fsms. This makes sure I don't get invested in some temple of gen_foo behaviors instead of getting invested in actually solving my problem (you're going to have to write it at least twice before its right anyway...).
Hopefully this gives you at least a place to start tackling your problem. In any case, this is a very natural sort of thing to do in Erlang -- and is close enough to the way the language and the problem work that it should be pretty fun to work on.
I have seen people use dict, ordict, record for maintaining state in many blogs that I have read. I find it as very vital concept.
Generally I understand the meaning of maintaining state and recursions but when it comes to Erlang..I am a little vague about how it is handled.
Any help?
State is the present arrangement of data. It is sometimes hard to remember this for two reasons:
State means both the data in the program and the program's current point of execution and "mode".
We build this up to be some magical thing unnecessarily.
Consider this:
"What is the process's state?" is asking about the present value of variables.
"What state is the process in?" usually refers to the mode, options, flags or present location of execution.
If you are a Turing machine then these are the same question; we have separated the ideas to give us handy abstractions to build on (like everything else in programming).
Let's think about state variables for a moment...
In many older languages you can alter state variables from whatever context you like, whether the modification of state is appropriate or not, because you manage this directly. In more modern languages this is a bit more restricted by imposing type declarations, scoping rules and public/private context to variables. This is really a rules arms-race, each language finding more ways to limit when assignment is permitted. If scheduling is the Prince of Frustration in concurrent programming, assignment is the Devil Himself. Hence the various cages built to manage him.
Erlang restricts the situations that assignment is permitted in a different way by setting the basic rule that assignment is only once per entry to a function, and functions are themselves the sole definition of procedural scope, and that all state is purely encapsulated by the executing process. (Think about the statement on scope to understand why many people feel that Erlang macros are a bad thing.)
These rules on assignment (use of state variables) encourage you to think of state as discreet slices of time. Every entry to a function starts with a clean slate, whether the function is recursive or not. This is a fundamentally different situation than the ongoing chaos of in-place modifications made from anywhere to anywhere in most other languages. In Erlang you never ask "what is the value of X right now?" because it can only ever be what it was initially assigned to be in the context of the current run of the current function. This significantly limits the chaos of state changes within functions and processes.
The details of those state variables and how they are assigned is incidental to Erlang. You already know about lists, tuples, ETS, DETS, mnesia, db connections, etc. Whatever. The core idea to understand about Erlang's style is how assignment is managed, not the incidental details of this or that particular data type.
What about "modes" and execution state?
If we write something like:
has_cheeseburger(BurgerName) ->
receive
{From, ask, burger_name} ->
From ! {ok, BurgerName},
has_cheeseburger(BurgerName);
{From, new_burger, _SomeBurger} ->
From ! {error, already_have_a_burger},
has_cheeseburger(BurgerName);
{From, eat_burger} ->
From ! {ok, {ate, BurgerName}},
lacks_cheeseburger()
end.
lacks_cheeseburger() ->
receive
{From, ask, burger_name} ->
From ! {error, no_burger},
lacks_cheeseburger();
{From, new_burger, BurgerName} ->
From ! {ok, thanks},
has_cheeseburger(BurgerName);
{From, eat_burger} ->
From ! {error, no_burger},
lacks_cheeseburger()
end.
What are we looking at? A loop. Conceptually its just one loop. Quite often a programmer would choose to write just one loop in code and add an argument like IsHoldingBurger to the loop and check it after each message in the receive clause to determine what action to take.
Above, though, the idea of two operating modes is both more explicit (its baked into the structure, not arbitrary procedural tests) and less verbose. We have separated the context of execution by writing basically the same loop twice, once for each condition we might be in, either having a burger or lacking one. This is at the heart of how Erlang deals with a concept called "finite state machines" and its really useful. OTP includes a tool build around this idea in the gen_fsm module. You can write your own FSMs by hand as I did above or use gen_fsm -- either way, when you identify you have a situation like this writing code in this style makes reasoning much easier. (For anything but the most trivial FSM you will really appreciate gen_fsm.)
Conclusion
That's it for state handling in Erlang. The chaos of untamed assignment is rendered impotent by the basic rules of single-assignment and absolute data encapsulation within each process (this implies that you shouldn't write gigantic processes, by the way). The supremely useful concept of a limited set of operating modes is abstracted by the OTP module gen_fsm or can be rather easily written by hand.
Since Erlang does such a good job limiting the chaos of state within a single process and makes the nightmare of concurrent scheduling among processes entirely invisible, that only leaves one complexity monster: the chaos of interactions among loosely coupled actors. In the mind of an Erlanger this is where the complexity belongs. The hard stuff should generally wind up manifesting there, in the no-man's-land of messages, not within functions or processes themselves. Your functions should be tiny, your needs for procedural checking relatively rare (compared to C or Python), your need for mode flags and switches almost nonexistant.
Edit
To reiterate Pascal's answer, in a super limited way:
loop(State) ->
receive
{async, Message} ->
NewState = do_something_with(Message),
loop(NewState);
{sync, From, Message} ->
NewState = do_something_with(Message),
Response = process_some_response_on(NewState),
From ! {ok, Response},
loop(NewState);
shutdown ->
exit(shutdown);
Any ->
io:format("~p: Received: ~tp~n", [self(), Any]),
loop(State)
end.
Re-read tkowal's response for the most minimal version of this. Re-read Pascal's for an expansion of the same idea to include servicing messages. Re-read the above for a slightly different style of the same pattern of state handling with the addition of ouputting unexpected messages. Finally, re-read the two-state loop I wrote above and you'll see its actually just another expansion on this same idea.
Remember, you can't re-assign a variable within the same iteration of a function but the next call can have different state. That is the extent of state handling in Erlang.
These are all variations on the same thing. I think you're expecting there to be something more, a more expansive mechanism or something. There is not. Restricting assignment eliminates all the stuff you're probably used to seeing in other languages. In Python you do somelist.append(NewElement) and the list you had now has changed. In Erlang you do NewList = lists:append(NewElement, SomeList) and SomeList is sill exactly the same as it used to be, and a new list has been returned that includes the new element. Whether this actually involves copying in the background is not your problem. You don't handle those details, so don't think about them. This is how Erlang is designed, and that leaves single assignment and making fresh function calls to enter a fresh slice of time where the slate has been wiped clean again.
The easiest way to maintain state is using gen_server behaviour. You can read more on Learn you some Erlang and in the docs.
gen_server is process, that can be:
initialised with given state,
can have defined synchronous and asynchronous callbacks (synchronous for querying the data in "request-response style" and asynchronous for changing the state with "fire and forget" style)
It also has couple of nice OTP mechanisms:
it can be supervised
it gives you basic logging
its code can be upgraded while the server is running without loosing the state
and so on...
Conceptually gen_server is an endless loop, that looks like this:
loop(State) ->
NewState = handle_requests(State),
loop(NewState).
where handle requests receives messages. This way all requests are serialised, so there are no race conditions. Of course it is a little bit more complicated to give you all the goodies, that I described.
You can choose what data structure you want to use for State. It is common to use records, because they have named fields, but since Erlang 17 maps can come in handy. This one depends on, what you want to store.
Variable are not mutable, so when you want to have an evolution of state, you create a new variable, and later recall the same function with this new state as parameter.
This structure is meant for processes like server, there is no base condition as in the factorial usual example, generally there is a specific message to stop the server smoothly.
loop(State) ->
receive
{add,Item} -> NewState = [Item|State], % create a new variable
loop(NewState); % recall loop with the new variable
{remove,Item} -> NewState = lists:filter(fun(X) -> X /= Item end,State) , % create a new variable
loop(NewState); % recall loop with the new variable
{items,Pid} -> Pid ! {items,State},
loop(State);
stop -> stopped; % this will be the stop condition
_ -> loop(State) % ignoring other message may be interesting in a never ending loop
end
HOW I WISH I HAD PHRASED MY QUESTION TO BEGIN WITH
Take a table with 26 keys, a-z and let them have integer values.
Create a process, Ouch, that does two things over and over again
In one transaction, write random values for a, b, and c such that those values always sum to 10
In another transaction, read the values for a, b and c and complain if their values do not sum to 10
If you spin-up even a few of these processes you will see that very quickly a, b and c are in a state where their values do not sum to 10. I believe there is no way to ask mnesia to "lock these 3 records before starting the writes (or reads)", one can only have mnesia lock the records as it gets to them (so to speak) which allows for the set of records' values to violate my "must sum to 10" constraint.
If I am right, solutions to this problem include
lock the entire table before writing (or reading) the set of 3 records -- I hate to lock whole table for 3 recs,
Create a process that keeps track of who is reading or writing which keys and protects bulk operations from anyone else writing or reading until the operation is completed. Of course I would have to make sure that all processes made use of this... crap, I guess this means writing my own AccessMod as the fourth parameter to activity/4 which seems like a non-trivial exercise
Some other thing that I am not smart enough to figure out.
thoughts?
Ok, I'm an ambitious Erlang newbee, so sorry if this is a dumb question, but
I am building an application-specific, in-memory distributed cache and I need to be able to write sets of Key, Value pairs in one transaction and also retrieve sets of values in one transaction. In other words I need to
1) Write 40 key,value pairs into the cache and ensure that no one else can read or write any of these 40 keys during this multi-key write operation; and,
2) Read 40 keys in one operation and get back 40 values knowing that all 40 values have been unchanged from the moment that this read operation started until it ended.
The only way I can think of doing this is to lock the entire table at the beginning of the fetch_keylist([ListOfKeys]) or at the beginning of the write_keylist([KeyValuePairs], but I don't want to do this because I have many processes simultaneously doing their own multi_key reads and writes and I don't want to lock the entire table any time any process needs to read/write a relatively small subset of records.
Help?
Trying to be more clear: I do not think this is just about using vanilla transactions
I think I am asking a more subtle question than this. Imagine that I have a process that, within a transaction, iterates through 10 records, locking them as it goes. Now imagine this process starts but before it iterates to the 3rd record ANOTHER process updates the 3rd record. This will be just fine as far as transactions go because the first process hadn't locked the 3rd record (yet) and the OTHER process modified it and released it before the first process got to it. What I want is to be guaranteed that once my first process starts that no other process can touch the 10 records until the first process is done with them.
PROBLEM SOLVED - I'M AN IDIOT... I guess...
Thank you all for your patients, especially Hynek -Pichi- Vychodil!
I prepared my test code to show the problem, and I could in fact reproduce the problem. I then simplified the code for readability and the problem went away. I was not able to again reproduce the problem. This is both embarrassing and mysterious to me since I had this problem for days. Also mnesia never complained that I was executing operations outside of a transaction and I have no dirty transactions anywhere in my code, I have no idea how I was able to introduce this bug into my code!
I have pounded the notion of Isolation into my head and will not doubt that it exists again.
Thanks for the education.
Actually, turns out the problem was using try/catch around mnesia operations within a transaction. See here for more.
Mnesia transaction will do exactly this thing for you. It is what is transaction for unless you do dirty operations. So just place your write and read operations to one transaction a mnesia will do rest. All operations in one transaction is done as one atomic operation. Mnesia transaction isolation level is what is sometimes known as "serializable" i.e. strongest isolation level.
Edit:
It seems you missed one important point about concurrent processes in Erlang. (To be fair it is not only true in Erlang but in any truly concurrent environment and when someone arguing else it is not really concurrent environment.) You can't distinguish which action happen first and which happen second unless you do some synchronization. Only way you can do this synchronization is using message passing. You have guaranteed only one thing about messages in Erlang, ordering of messages sent from one process to other process. It means when you send two messages M1 and M2 from process A to process B they arrives in same order. But if you send message M1 from A to B and message M2 from C to B they can arrive in any order. Simply because how you can even tell which message you sent first? It is even worse if you send message M1 from A to B and then M2 from A to C and when M2 arrives to C send M3 from C to B you don't have guarantied that M1 arrives to B before M3. Even it will happen in one VM in current implementation. But you can't rely on it because it is not guaranteed and can change even in next version of VM just due message passing implementation between different schedulers.
It illustrates problems of event ordering in concurrent processes. Now back to the mnesia transaction. Mnesia transaction have to be side effect free fun. It means there may not be any message sending outside from transaction. So you can't tell which transaction starts first and when starts. Only thing you can tell if transaction succeed and they order you can only determine by its effect. When you consider this your subtle clarification makes no sense. One transaction will read all keys in atomic operation even it is implemented as reading one key by one in transaction implementation and your write operation will be also performed as atomic operation. You can't tell if write to 4th key in second transaction was happen after you read 1st key in first transaction because there it is not observable from outside. Both transaction will be performed in particular order as separate atomic operation. From outside point of view all keys will be read in same point of time and it is work of mnesia to force it. If you send message from inside of transaction you violate mnesia transaction property and you can't be surprised it will behave strange. To be concrete, this message can be send many times.
Edit2:
If you spin-up even a few of these processes you will see that very
quickly a, b and c are in a state where their values do not sum to 10.
I'm curious why you think it would happen or you tested it? Show me your test case and I will show mine:
-module(transactions).
-export([start/2, sum/0, write/0]).
start(W, R) ->
mnesia:start(),
{atomic, ok} = mnesia:create_table(test, [{ram_copies,[node()]}]),
F = fun() ->
ok = mnesia:write({test, a, 10}),
[ ok = mnesia:write({test, X, 0}) || X <-
[b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z]],
ok
end,
{atomic, ok} = mnesia:transaction(F),
F2 = fun() ->
S = self(),
erlang:send_after(1000, S, show),
[ spawn_link(fun() -> writer(S) end) || _ <- lists:seq(1,W) ],
[ spawn_link(fun() -> reader(S) end) || _ <- lists:seq(1,R) ],
collect(0,0)
end,
spawn(F2).
collect(R, W) ->
receive
read -> collect(R+1, W);
write -> collect(R, W+1);
show ->
erlang:send_after(1000, self(), show),
io:format("R: ~p, W: ~p~n", [R,W]),
collect(R, W)
end.
keys() ->
element(random:uniform(6),
{[a,b,c],[a,c,b],[b,a,c],[b,c,a],[c,a,b],[c,b,a]}).
sum() ->
F = fun() ->
lists:sum([X || K<-keys(), {test, _, X} <- mnesia:read(test, K)])
end,
{atomic, S} = mnesia:transaction(F),
S.
write() ->
F = fun() ->
[A, B ] = L = [ random:uniform(10) || _ <- [1,2] ],
[ok = mnesia:write({test, K, V}) || {K, V} <- lists:zip(keys(),
[10-A-B|L])],
ok
end,
{atomic, ok} = mnesia:transaction(F),
ok.
reader(P) ->
case sum() of
10 ->
P ! read,
reader(P);
_ ->
io:format("ERROR!!!~n",[]),
exit(error)
end.
writer(P) ->
ok = write(),
P ! write,
writer(P).
If it would not work it would be really serious problem. There are serious applications including payment systems which rely on it. If you have test case which shows it is broken, please report bug at erlang-bugs#erlang.org
Have you tried mnesia Events ? You can have the reader subscribe to mnesia's Table Events especially write events so as not to interrupt the process doing the writing. In this way, mnesia just keeps sending a copy of what has been written in real-time to the other process which checks what the values are at any one time. take a look at this:
subscriber()->
mnesia:subscribe({table,YOUR_TABLE_NAME,simple}),
%% OR mnesia:subscribe({table,YOUR_TABLE_NAME,detailed}),
wait_events().
wait_events()->
receive
%% For simple events
{mnesia_table_event,{write, NewRecord, ActivityId}} ->
%% Analyse the written record as you wish
wait_events();
%% For detailed events
{mnesia_table_event,{write, YOUR_TABLE, NewRecord, [OldRecords], ActivityId}} ->
%% Analyse the written record as you wish
wait_events();
_Any -> wait_events()
end.
Now you spawn your analyser as a process like this:
spawn(?MODULE,subscriber,[]).
This makes the whole process to run without any process being blocked, mnesia needs not lock any tabel or record because now what you have is a writer process and an analyser process. The whole thing will run in real-time. Remember that there are many other events that you can make use of if you wish by pattern matching them in the subscriber wait_events() receive body.
Its possible to build a heavy duty gen_server or complete application intended for reception and analysis of all your mnesia events. Its usually better to have one capable subscriber than many failing event subscribers. If i have understood you question well, this unblocking solution fits your requirements.
mnesia:read/3 with write locks seems to be suffient.
Mnesia's transaction is implemented by read-write lock and locks are well-formed (holding lock untill the end of transaction). So the isolation level is serializable.
The granularity of locks are per record as long as you access by primary key.