You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@thrift.apache.org by Julien Greard <jg...@e-vitech.com> on 2022/04/04 07:54:08 UTC
Using a processor factory with TMultiplexedProcessor in thrift
Hello,
I have the following architecture in my code:
* 1 thrift Server
* 1 thrift MultiplexedProcessor
* several Services & associated processors/handlers
* N thrift clients (several by processor/handler)
* 1 queue which contains the tasks asked by the clients
What I*really*want is to be able cancel a task when a client crashes.
Let me explain:
Let's say I have the following service handler method :
|// inside thrift handler // this method is currently instanciated once
and called by every client void do_stuff(const std::string& parameter) {
auto task = make_task(parameter); auto future_result = task
.get_future(); add_to_queue(task); auto status =
resultFuture.wait_for(timeout); // wait here until the task is over if
(status != std::future_status::timeout) { return future_result.get(); } } |
If the client crashes/disconnects, there is no need to process with the
task anymore, so I would like to cancel it (remove from my queue) asap.
To do so, I had the following ideas:
1/ Using the*setServerEventHandler*method from my Thrift Server to be
notified when a client disconnects (deleteContext method). It works very
well but I am not able to know which request was made by which specific
client: within my handler I do not have access to the client infos, and
I can't use (or didnt managed to) the void * context created by the
method creatContext in my ServerEventHandler
2/ Using the contructor & destructor of my handler object to link 1
client to 1 handler. Then when the client disconnects, I only have to
cancel every task he asked in the handler destructor. Currently I have N
client and 1 handler, so it doesn't work. I figured I could use a
TMultiplexedProcessorFactory but there is no such class in thrift. The
TMultiplexedProcessor has to register Processor (with single handler)
when I'd like to register a Factory which will create one handler
instance by client. I could implement my own
TMultiplexedProcessorFactory but I think there might be a good reason
this doesn't exist yet.
Any idea of what I should do ?
I can post it inhttps://issues.apache.org/but it's not exactly an issue
with thrift (great lib btw), rather than a question.
Thanks in advance, don't hesitate to ask for more details
This post is a duplicate of a StackOverflow question I asked last Friday
<https://stackoverflow.com/questions/71707726/using-a-processor-factory-with-tmultiplexedprocessor-in-thrift>
Thanks in advance,
Julien Greard
Dev at Evitech - France
Re: Using a processor factory with TMultiplexedProcessor in thrift
Posted by Yuxuan Wang <yu...@reddit.com.INVALID>.
Yea that would be a big breaking change. We did the same breaking change in
go library a few years ago, but also the request context thing is a
language level feature (it's in go's stdlib) so the breaking change was
easier to reason about.
I'm just providing a data point on how we solved the same problem in
another language library :) I would certainly defer to people more familiar
with the cpp library to discuss what's the best way for cpp library.
It's also probably a good idea to create an issue on jira to discuss the
solution. I'm not sure which one between the jira system and the maillist
is more active, I'd say that's probably the jira system.
On Mon, Apr 4, 2022 at 11:56 AM Julien Greard <jg...@e-vitech.com> wrote:
> Hi,
>
> First, thank you very much for the answer !
>
> Yes, it's in C++, sorry for the lack of info.
>
> Yes, indeed, using a context passed into the handler should do the trick
> but I dont have any context passed in my contexts already: I would have
> to add it, which means forking thrift, make the change in thrift github
> (which I could do, but it would be a huge amount of work) or modify the
> generated code afterwards, right ?
>
> Julien
>
>
> Le 04/04/2022 à 19:23, Yuxuan Wang a écrit :
> > From your code snippet I assume you use cpp. I'm not familiar with
> thrift
> > cpp library but I can share how we handled this in go library:
> >
> > 1. in TSocket.IsOpen and TSSLSocket.IsOpen, we added a connectivity check
> > on supported platforms (basically all unix) to return false if we
> detected
> > that the other end already closed the connection (src
> > <
> https://github.com/apache/thrift/blob/15cc0c4da218a375cadc67e541a99fdc6c5083f2/lib/go/thrift/socket_unix_conn.go#L41
> >
> > )
> > 2. in compiler generated handler process function, we add a background
> > goroutine (= lightweight thread in other languages) that checks for
> > connectivity periodically, and propagate the result via a context passed
> > into handler functions (src in thrift compiler
> > <
> https://github.com/apache/thrift/blob/15cc0c4da218a375cadc67e541a99fdc6c5083f2/compiler/cpp/src/thrift/generate/t_go_generator.cc#L2992-L3041
> >
> > )
> >
> > this way the service handler function (and everything downstream it
> calls)
> > can check for the context passed in and abandon the request if it's no
> > longer needed.
> >
> > On Mon, Apr 4, 2022 at 12:54 AM Julien Greard <jg...@e-vitech.com>
> wrote:
> >
> >> Hello,
> >>
> >> I have the following architecture in my code:
> >>
> >> * 1 thrift Server
> >> * 1 thrift MultiplexedProcessor
> >> * several Services & associated processors/handlers
> >> * N thrift clients (several by processor/handler)
> >> * 1 queue which contains the tasks asked by the clients
> >>
> >> What I*really*want is to be able cancel a task when a client crashes.
> >>
> >> Let me explain:
> >>
> >> Let's say I have the following service handler method :
> >>
> >> |// inside thrift handler // this method is currently instanciated once
> >> and called by every client void do_stuff(const std::string& parameter) {
> >> auto task = make_task(parameter); auto future_result = task
> >> .get_future(); add_to_queue(task); auto status =
> >> resultFuture.wait_for(timeout); // wait here until the task is over if
> >> (status != std::future_status::timeout) { return future_result.get(); }
> } |
> >>
> >> If the client crashes/disconnects, there is no need to process with the
> >> task anymore, so I would like to cancel it (remove from my queue) asap.
> >>
> >> To do so, I had the following ideas:
> >>
> >> 1/ Using the*setServerEventHandler*method from my Thrift Server to be
> >> notified when a client disconnects (deleteContext method). It works very
> >> well but I am not able to know which request was made by which specific
> >> client: within my handler I do not have access to the client infos, and
> >> I can't use (or didnt managed to) the void * context created by the
> >> method creatContext in my ServerEventHandler
> >>
> >> 2/ Using the contructor & destructor of my handler object to link 1
> >> client to 1 handler. Then when the client disconnects, I only have to
> >> cancel every task he asked in the handler destructor. Currently I have N
> >> client and 1 handler, so it doesn't work. I figured I could use a
> >> TMultiplexedProcessorFactory but there is no such class in thrift. The
> >> TMultiplexedProcessor has to register Processor (with single handler)
> >> when I'd like to register a Factory which will create one handler
> >> instance by client. I could implement my own
> >> TMultiplexedProcessorFactory but I think there might be a good reason
> >> this doesn't exist yet.
> >>
> >> Any idea of what I should do ?
> >>
> >> I can post it inhttps://issues.apache.org/but it's not exactly an issue
> >> with thrift (great lib btw), rather than a question.
> >>
> >> Thanks in advance, don't hesitate to ask for more details
> >>
> >>
> >> This post is a duplicate of a StackOverflow question I asked last Friday
> >> <
> >>
> https://stackoverflow.com/questions/71707726/using-a-processor-factory-with-tmultiplexedprocessor-in-thrift
> >>
> >> Thanks in advance,
> >>
> >>
> >> Julien Greard
> >>
> >> Dev at Evitech - France
> >>
>
Re: Using a processor factory with TMultiplexedProcessor in thrift
Posted by Julien Greard <jg...@e-vitech.com>.
Hi,
First, thank you very much for the answer !
Yes, it's in C++, sorry for the lack of info.
Yes, indeed, using a context passed into the handler should do the trick
but I dont have any context passed in my contexts already: I would have
to add it, which means forking thrift, make the change in thrift github
(which I could do, but it would be a huge amount of work) or modify the
generated code afterwards, right ?
Julien
Le 04/04/2022 à 19:23, Yuxuan Wang a écrit :
> From your code snippet I assume you use cpp. I'm not familiar with thrift
> cpp library but I can share how we handled this in go library:
>
> 1. in TSocket.IsOpen and TSSLSocket.IsOpen, we added a connectivity check
> on supported platforms (basically all unix) to return false if we detected
> that the other end already closed the connection (src
> <https://github.com/apache/thrift/blob/15cc0c4da218a375cadc67e541a99fdc6c5083f2/lib/go/thrift/socket_unix_conn.go#L41>
> )
> 2. in compiler generated handler process function, we add a background
> goroutine (= lightweight thread in other languages) that checks for
> connectivity periodically, and propagate the result via a context passed
> into handler functions (src in thrift compiler
> <https://github.com/apache/thrift/blob/15cc0c4da218a375cadc67e541a99fdc6c5083f2/compiler/cpp/src/thrift/generate/t_go_generator.cc#L2992-L3041>
> )
>
> this way the service handler function (and everything downstream it calls)
> can check for the context passed in and abandon the request if it's no
> longer needed.
>
> On Mon, Apr 4, 2022 at 12:54 AM Julien Greard <jg...@e-vitech.com> wrote:
>
>> Hello,
>>
>> I have the following architecture in my code:
>>
>> * 1 thrift Server
>> * 1 thrift MultiplexedProcessor
>> * several Services & associated processors/handlers
>> * N thrift clients (several by processor/handler)
>> * 1 queue which contains the tasks asked by the clients
>>
>> What I*really*want is to be able cancel a task when a client crashes.
>>
>> Let me explain:
>>
>> Let's say I have the following service handler method :
>>
>> |// inside thrift handler // this method is currently instanciated once
>> and called by every client void do_stuff(const std::string& parameter) {
>> auto task = make_task(parameter); auto future_result = task
>> .get_future(); add_to_queue(task); auto status =
>> resultFuture.wait_for(timeout); // wait here until the task is over if
>> (status != std::future_status::timeout) { return future_result.get(); } } |
>>
>> If the client crashes/disconnects, there is no need to process with the
>> task anymore, so I would like to cancel it (remove from my queue) asap.
>>
>> To do so, I had the following ideas:
>>
>> 1/ Using the*setServerEventHandler*method from my Thrift Server to be
>> notified when a client disconnects (deleteContext method). It works very
>> well but I am not able to know which request was made by which specific
>> client: within my handler I do not have access to the client infos, and
>> I can't use (or didnt managed to) the void * context created by the
>> method creatContext in my ServerEventHandler
>>
>> 2/ Using the contructor & destructor of my handler object to link 1
>> client to 1 handler. Then when the client disconnects, I only have to
>> cancel every task he asked in the handler destructor. Currently I have N
>> client and 1 handler, so it doesn't work. I figured I could use a
>> TMultiplexedProcessorFactory but there is no such class in thrift. The
>> TMultiplexedProcessor has to register Processor (with single handler)
>> when I'd like to register a Factory which will create one handler
>> instance by client. I could implement my own
>> TMultiplexedProcessorFactory but I think there might be a good reason
>> this doesn't exist yet.
>>
>> Any idea of what I should do ?
>>
>> I can post it inhttps://issues.apache.org/but it's not exactly an issue
>> with thrift (great lib btw), rather than a question.
>>
>> Thanks in advance, don't hesitate to ask for more details
>>
>>
>> This post is a duplicate of a StackOverflow question I asked last Friday
>> <
>> https://stackoverflow.com/questions/71707726/using-a-processor-factory-with-tmultiplexedprocessor-in-thrift
>>
>> Thanks in advance,
>>
>>
>> Julien Greard
>>
>> Dev at Evitech - France
>>
Re: Using a processor factory with TMultiplexedProcessor in thrift
Posted by Yuxuan Wang <yu...@reddit.com.INVALID>.
From your code snippet I assume you use cpp. I'm not familiar with thrift
cpp library but I can share how we handled this in go library:
1. in TSocket.IsOpen and TSSLSocket.IsOpen, we added a connectivity check
on supported platforms (basically all unix) to return false if we detected
that the other end already closed the connection (src
<https://github.com/apache/thrift/blob/15cc0c4da218a375cadc67e541a99fdc6c5083f2/lib/go/thrift/socket_unix_conn.go#L41>
)
2. in compiler generated handler process function, we add a background
goroutine (= lightweight thread in other languages) that checks for
connectivity periodically, and propagate the result via a context passed
into handler functions (src in thrift compiler
<https://github.com/apache/thrift/blob/15cc0c4da218a375cadc67e541a99fdc6c5083f2/compiler/cpp/src/thrift/generate/t_go_generator.cc#L2992-L3041>
)
this way the service handler function (and everything downstream it calls)
can check for the context passed in and abandon the request if it's no
longer needed.
On Mon, Apr 4, 2022 at 12:54 AM Julien Greard <jg...@e-vitech.com> wrote:
> Hello,
>
> I have the following architecture in my code:
>
> * 1 thrift Server
> * 1 thrift MultiplexedProcessor
> * several Services & associated processors/handlers
> * N thrift clients (several by processor/handler)
> * 1 queue which contains the tasks asked by the clients
>
> What I*really*want is to be able cancel a task when a client crashes.
>
> Let me explain:
>
> Let's say I have the following service handler method :
>
> |// inside thrift handler // this method is currently instanciated once
> and called by every client void do_stuff(const std::string& parameter) {
> auto task = make_task(parameter); auto future_result = task
> .get_future(); add_to_queue(task); auto status =
> resultFuture.wait_for(timeout); // wait here until the task is over if
> (status != std::future_status::timeout) { return future_result.get(); } } |
>
> If the client crashes/disconnects, there is no need to process with the
> task anymore, so I would like to cancel it (remove from my queue) asap.
>
> To do so, I had the following ideas:
>
> 1/ Using the*setServerEventHandler*method from my Thrift Server to be
> notified when a client disconnects (deleteContext method). It works very
> well but I am not able to know which request was made by which specific
> client: within my handler I do not have access to the client infos, and
> I can't use (or didnt managed to) the void * context created by the
> method creatContext in my ServerEventHandler
>
> 2/ Using the contructor & destructor of my handler object to link 1
> client to 1 handler. Then when the client disconnects, I only have to
> cancel every task he asked in the handler destructor. Currently I have N
> client and 1 handler, so it doesn't work. I figured I could use a
> TMultiplexedProcessorFactory but there is no such class in thrift. The
> TMultiplexedProcessor has to register Processor (with single handler)
> when I'd like to register a Factory which will create one handler
> instance by client. I could implement my own
> TMultiplexedProcessorFactory but I think there might be a good reason
> this doesn't exist yet.
>
> Any idea of what I should do ?
>
> I can post it inhttps://issues.apache.org/but it's not exactly an issue
> with thrift (great lib btw), rather than a question.
>
> Thanks in advance, don't hesitate to ask for more details
>
>
> This post is a duplicate of a StackOverflow question I asked last Friday
> <
> https://stackoverflow.com/questions/71707726/using-a-processor-factory-with-tmultiplexedprocessor-in-thrift
> >
>
>
> Thanks in advance,
>
>
> Julien Greard
>
> Dev at Evitech - France
>