Simplify process(Script|Python)(Remote)?Call #57857
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
lw
requested review from
H-Huang,
mingzhe09088,
mrshenli,
pritamdamania87,
rohan-varma,
wayi1 and
zhaojuanmao
as code owners
facebook-github-bot
added
oncall: distributed
💊 CI failures summary and remediationsAs of commit 3f9ed81 (more details on the Dr. CI page):
🕵️ 1 new failure recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
This was referenced May 7, 2021
Closed
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
This was referenced May 17, 2021
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
mrshenli
approved these changes
May 18, 2021
| const std::function<void(Message)>& markComplete, | ||
| const c10::intrusive_ptr<JitFuture>& responseFuture) const { | ||
| c10::intrusive_ptr<JitFuture> RequestCallbackImpl::processPythonCall( | ||
| RpcCommandBase& rpc) const { | ||
| auto& upc = static_cast<UnpickledPythonCall&>(rpc); | ||
| auto future = runPythonFunction(upc.pythonUdf(), upc.isAsyncExecution()); |
There was a problem hiding this comment.
I assume, later, we will need to pass agent devices to this Future?
There was a problem hiding this comment.
Correct, I'm doing it in #58426.
| responseFuture->markCompleted(c10::make_intrusive<Message>( | ||
| PythonResp(std::move(serializedPyObj)).toMessage())); | ||
| }); | ||
| return future->then( |
There was a problem hiding this comment.
Each of the futures in the chain will introduce another data ptrs extraction for CUDA (when CUDA is enabled here). Do we need to worry about the perf issue here?
There was a problem hiding this comment.
I have been asking myself the same thing. I'm somewhat worried about DataPtr extraction (although not too much, because for the most part here we're inspecting JIT IValues, which should be much cheaper than inspecting Python objects) but I'm also a bit worried about creating/recording/waiting on events each time, getting new streams, ...
I have some ideas on how we can mitigate some of these issues. For example in #58424 I'll allow callbacks to provide pre-extracted DataPtrs, which should allow us to skip extraction in some cases. I also believe we can add some fastpaths in Future so that for example we reuse the same streams from the "parent" when running callbacks.
Honestly, though, I'm also fine for now to take a perf hit if necessary, if that means we can get CUDA support to work correctly and reliably, and then look later into optimizing it.
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
This was referenced May 20, 2021
There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/) [ghstack-poisoned]
dgl-intel
pushed a commit
to dgl-intel/pytorch
that referenced
this pull request
May 21, 2021
Pull Request resolved: pytorch#57857 There used to be a whole lot of methods: `processPythonCall`, `processScriptCall`, `processScriptRemoteCall`, `processPythonRemoteCall`, `processScriptCallOp`, `processBaseScriptRemoteCall` and `processScriptRemoteCallOp`. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future. ghstack-source-id: 129567070 Differential Revision: [D28253848](https://our.internmc.facebook.com/intern/diff/D28253848/)
|
This pull request has been merged in cd9dbbd. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack:
There used to be a whole lot of methods:
processPythonCall,processScriptCall,processScriptRemoteCall,processPythonRemoteCall,processScriptCallOp,processBaseScriptRemoteCallandprocessScriptRemoteCallOp. Thanks to the previous simplification, we can now drop all but the first four, which map nicely 1:1 to the four message types we need to handle. Also their signatures become much simpler: they take an RPC command and return a future.Differential Revision: D28253848