[WIP] per-RPC device mapping#64901
Closed
pbelevich wants to merge 26 commits intogh/pbelevich/155/basefrom
Closed
Conversation
[ghstack-poisoned]
Contributor
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit a018e7d (more details on the Dr. CI page):
🕵️ 8 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
| Job | Step | Action |
|---|---|---|
| Test | 🔁 rerun | |
| Fail if there were any warnings | 🔁 rerun | |
| Run mypy | 🔁 rerun |
This comment was automatically generated by Dr. CI (expand for details).
Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions to the (internal) Dr. CI Users group.
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
mrshenli
reviewed
Sep 20, 2021
Contributor
mrshenli
left a comment
There was a problem hiding this comment.
Just a minor comment that, it might be easier for review and test if we have incremental PRs that targets one API at a time.
mrshenli
reviewed
Sep 20, 2021
| py::arg("device_map") = DeviceMap(), | ||
| py::arg("timeout") = py::cast(kUnsetRpcTimeout), | ||
| py::call_guard<py::gil_scoped_release>(), | ||
| R"( |
Contributor
There was a problem hiding this comment.
add this new argument to docstring?
| .def( | ||
| "to_here", | ||
| &PyRRef::toHere, | ||
| py::arg("device_map") = DeviceMap(), |
Contributor
There was a problem hiding this comment.
What's the default behavior here? Will it inherit the global device map or fallback to CPU RPC?
| : payload_(std::move(payload)), | ||
| tensors_(std::move(tensors)), | ||
| type_(type), | ||
| deviceMap_(deviceMap) {} |
Contributor
There was a problem hiding this comment.
do we need std::move here?
| } | ||
|
|
||
| void Message::setDeviceMap(DeviceMap&& deviceMap) { | ||
| deviceMap_ = deviceMap; |
| pc.serializedPyObj(), pc.isAsyncExecution()); | ||
| pc.serializedPyObj(), | ||
| std::move(pc).moveDeviceMap(), | ||
| pc.isAsyncExecution()); |
| namespace distributed { | ||
| namespace rpc { | ||
|
|
||
| // TODO(pbelevich) |
Contributor
There was a problem hiding this comment.
I assume the TODO here means consolidation? :)
Comment on lines
+4
to
+6
| #include <torch/csrc/distributed/rpc/rpc_agent.h> | ||
| #include <torch/csrc/distributed/rpc/types.h> | ||
| #include <torch/csrc/jit/serialization/pickle.h> |
Contributor
There was a problem hiding this comment.
I saw these two headers are included in the .cpp files in the derived classes. Do we still need them here? Will this create circular dependency?
|
|
||
| c10::intrusive_ptr<Message> ScriptRRefFetchRet::toMessageImpl() && { | ||
| auto res = fromIValues(values_, type_); | ||
| res->setDeviceMap(std::move(deviceMap_)); |
Contributor
There was a problem hiding this comment.
curious, why do we set after creating the message instead of letting fromIValues handle deviceMap_ as well?
| RRefFetchRet( | ||
| std::vector<at::IValue> values, | ||
| MessageType type, | ||
| DeviceMap&& deviceMap) |
Contributor
There was a problem hiding this comment.
does the RRefFetchRet need a deviceMap?
Contributor
Author
There was a problem hiding this comment.
rref.to_here(device_map) propagates device_map to ScriptRRefFetchRet/PythonRRefFetchRet to put the result to the target device
| public: | ||
| RRefMessageBase(const RRefId& rrefId, MessageType type) | ||
| : rrefId_(rrefId), type_(type) {} | ||
| RRefMessageBase(const RRefId& rrefId, MessageType type, DeviceMap&& deviceMap) |
Contributor
There was a problem hiding this comment.
not 100% sure, but looks like only Script/PythonRRefFetchCall needs this device map. Does it make sense to only add this device map to those two messages?
mrshenli
reviewed
Sep 20, 2021
| # C++ (see python_rpc_handler.cpp). | ||
| class RRefProxy: | ||
| def __init__(self, rref, rpc_api, timeout=UNSET_RPC_TIMEOUT): | ||
| def __init__(self, rref, rpc_api, device_map, timeout=UNSET_RPC_TIMEOUT): |
Contributor
There was a problem hiding this comment.
curious, why do we need device_map in RRefProxy instead of letting the RPCs (_invoke_rpc above) on this proxy handling that?
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
CI Flow Status⚛️ CI FlowRuleset - Version:
You can add a comment to the PR and tag @pytorchbot with the following commands: # ciflow rerun, "ciflow/default" will always be added automatically
@pytorchbot ciflow rerun
# ciflow rerun with additional labels "-l <ciflow/label_name>", which is equivalent to adding these labels manually and trigger the rerun
@pytorchbot ciflow rerun -l ciflow/scheduled -l ciflow/slowFor more information, please take a look at the CI Flow Wiki. |
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
cc pietern mrshenli pritamdamania87 zhaojuanmao satgera rohan-varma gqchen aazzolini osalpekar jiayisuse @SciPioneer @H-Huang cbalioglu gcramer23 [ghstack-poisoned]
Contributor
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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:
cc @pietern @mrshenli @pritamdamania87 @zhaojuanmao @satgera @rohan-varma @gqchen @aazzolini @osalpekar @jiayisuse @SciPioneer @H-Huang @cbalioglu @gcramer23