[feature] Support udf through RPC #7519

yangzhg · 2021-12-29T02:55:45Z

Proposed changes

Support UDF is implemented through GRPC protocol. This brings several benefits:

The udf implementation language is not limited to c++, users can use any familiar language to implement udf
UDF is decoupled from Doris, udf will not be doris coredump together, udf computing resources are separated from doris, and doris services are not affected

But RPC's UDF has a fixed overhead, so its performance is much slower than C++ UDF, especially when the amount of data is large.

Create function like

CREATE FUNCTION rpc_add(INT, INT) RETURNS INT PROPERTIES (
  "SYMBOL"="add_int",
  "OBJECT_FILE"="127.0.0.1:9999",
  "TYPE"="RPC"
);

function service need to implements check_fn and fn_call methods
#7502 #7578

the rpc udf call is a little slower than the native function call in vectorized mode:

But it will much slow in non-vectorized mode, and if using rpc udf in vectorized mode the query time is equal to native function call in non-vectorized mode

Types of changes

What types of changes does your code introduce to Doris?
Put an x in the boxes that apply

Bugfix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to not work as expected)
Documentation Update (if none of the other choices apply)
Code refactor (Modify the code structure, format the code, etc...)
Optimization. Including functional usability improvements and performance improvements.
Dependency. Such as changes related to third-party components.
Other.

Checklist

Put an x in the boxes that apply. You can also fill these out after creating the PR. If you're unsure about any of them, don't hesitate to ask. We're here to help! This is simply a reminder of what we are going to look for before merging your code.

I have created an issue on (Fix [Feature] Support UDF through RPC #7520) and described the bug/feature there in detail
Compiling and unit tests pass locally with my changes
I have added tests that prove my fix is effective or that my feature works
If these changes need document changes, I have updated the document
Any dependent changes have been merged

Further comments

If this is a relatively large or complex change, kick off the discussion at dev@doris.apache.org by explaining why you chose the solution you did and what alternatives you considered, etc...

stalary · 2022-01-05T15:54:32Z

Let me see

stalary · 2022-01-08T14:10:11Z

Whether RPC addresses support load balance?

stalary · 2022-01-08T14:18:18Z

_brpc_stub_cache split _internal_client_cache& _function_client_cache Whether compatibility problems exist

stalary · 2022-01-08T14:03:36Z

be/src/exec/tablet_sink.cpp

-                                                                  _node_info.brpc_port)) {
-            ExecEnv::GetInstance()->brpc_stub_cache()->erase(_open_closure->cntl.remote_side());
-        }
+        ExecEnv::GetInstance()->brpc_internal_client_cache()->erase(


Why is it no longer necessary to judge unavailabe

it is not cost so much when create a new stub

stalary · 2022-01-08T14:06:44Z

be/src/exprs/rpc_fn_call.cpp

+            ptype->set_type(PParameterType::NULL_TYPE);
+            continue;
+        }
+        switch (_children[i]->type().type) {


Whether need to support TYPE_STRING?

Yes! I will add later

morningman · 2022-02-05T14:02:07Z

be/src/exprs/rpc_fn_call.cpp

+
+    if (_client == nullptr) {
+        return Status::InternalError(
+                fmt::format("rpc env init error: {}/{}", _fn.hdfs_location, _rpc_function_symbol));


hdfs_location?

be/src/exprs/rpc_fn_call.cpp

morningman · 2022-02-05T14:09:20Z

be/src/exprs/rpc_fn_call.cpp

+    Status st = _eval_children(context, row, &response);
+    if (!st.ok() || response.status().status_code() != 0 ||
+        (response.result().has_null() && response.result().null_map(0))) {
+        res_val.is_null = true;


Is it ok to return null if RPC failed?

morningman · 2022-02-05T14:14:51Z

be/src/vec/functions/function_rpc.cpp

+    PFunctionCallResponse response;
+    request.set_function_name(_symbol);
+    int64_t name_hash = 0;
+    murmur_hash3_x64_64(_symbol.c_str(), _symbol.size(), 21217891, &name_hash);


Does murmur_hash3_x64_64 work for ARM64?

morningman · 2022-02-05T14:28:22Z

gensrc/proto/types.proto

@@ -63,3 +67,150 @@ message PUniqueId {
    required int64 lo = 2;
 };

+message PGenericType {


This is a same datatype define in data.proto, how about unify them?
I will do this in PR #7939

That's a good idea, but I think the datatype define in data.proto can not describe complex types very well, PGenericType + PList + PMap + PField + PStruct + PDecimal may be better. And BTW, datatype should not to complex, it should be used more easier between languages

gensrc/proto/types.proto

…ocol. This brings several benefits: 1. The udf implementation language is not limited to c++, users can use any familiar language to implement udf 2. UDF is decoupled from Doris, udf will not be doris coredump together, udf computing resources are separated from doris, and doris services are not affected But RPC's UDF has a fixed overhead, so its performance is much slower than C++ UDF, especially when the amount of data is large. Create function like ``` CREATE FUNCTION rpc_add(INT, INT) RETURNS INT PROPERTIES ( "SYMBOL"="add_int", "OBJECT_FILE"="127.0.0.1:9999", "TYPE"="RPC" ); ``` function service need to implements `check_fn` and `fn_call` methods Note: THIS IS AN EXPERIMENTAL FEATRUE, THE INTERFACE AND DATA STRUCTURE MAY CHANGED IN FUTURE !!

morningman

LGTM

github-actions · 2022-02-07T07:28:20Z

PR approved by at least one committer and no changes requested.

github-actions · 2022-02-07T07:28:23Z

PR approved by anyone and no changes requested.

morningman · 2022-02-07T07:30:07Z

Document need to be added later

yangzhg changed the title … [feature] Support udf through RPC Dec 29, 2021

yangzhg force-pushed the rpc_udf branch from a366347 to afffd13 Compare December 29, 2021 03:02

morningman self-assigned this Jan 5, 2022

morningman assigned morningman and stalary and unassigned morningman Jan 5, 2022

morningman added area/udf Issues or PRs related to the UDF kind/feature Categorizes issue or PR as related to a new feature. labels Jan 5, 2022

stalary reviewed Jan 8, 2022

View reviewed changes

yangzhg force-pushed the rpc_udf branch from afffd13 to 4d97cc2 Compare January 14, 2022 05:22

yangzhg force-pushed the rpc_udf branch from 4d97cc2 to 16c3dd6 Compare January 24, 2022 06:18

yangzhg marked this pull request as draft January 24, 2022 06:21

yangzhg force-pushed the rpc_udf branch 2 times, most recently from bd0c3bc to c35abc1 Compare January 26, 2022 10:24

yangzhg marked this pull request as ready for review January 26, 2022 10:25

yangzhg force-pushed the rpc_udf branch 2 times, most recently from d6ad395 to f1cd42a Compare January 31, 2022 03:15

morningman reviewed Feb 5, 2022

View reviewed changes

yangzhg force-pushed the rpc_udf branch 6 times, most recently from 452dd4c to 8754929 Compare February 7, 2022 07:03

yangzhg force-pushed the rpc_udf branch from 8754929 to b6de6d6 Compare February 7, 2022 07:23

morningman approved these changes Feb 7, 2022

View reviewed changes

github-actions bot added the approved Indicates a PR has been approved by one committer. label Feb 7, 2022

github-actions bot added the reviewed label Feb 7, 2022

yangzhg merged commit f8d086d into apache:master Feb 8, 2022

yangzhg deleted the rpc_udf branch February 14, 2022 01:48

yangzhg mentioned this pull request Mar 3, 2022

Doris Roadmap 2022 #7502

Closed

morningman mentioned this pull request Mar 8, 2022

[Feature] Support Java UDF #8389

Closed

3 tasks

wzymumon mentioned this pull request May 31, 2023

[Feature] Support Wasm UDF #20267

Open

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[feature] Support udf through RPC #7519

[feature] Support udf through RPC #7519

yangzhg commented Dec 29, 2021 •

edited

Loading

stalary commented Jan 5, 2022

stalary commented Jan 8, 2022

stalary commented Jan 8, 2022

stalary Jan 8, 2022

yangzhg Jan 14, 2022

stalary Jan 8, 2022

yangzhg Jan 14, 2022

morningman Feb 5, 2022

morningman Feb 5, 2022

morningman Feb 5, 2022

morningman Feb 5, 2022

yangzhg Feb 7, 2022

morningman left a comment

github-actions bot commented Feb 7, 2022

github-actions bot commented Feb 7, 2022

morningman commented Feb 7, 2022

[feature] Support udf through RPC #7519

[feature] Support udf through RPC #7519

Conversation

yangzhg commented Dec 29, 2021 • edited Loading

Proposed changes

Types of changes

Checklist

Further comments

stalary commented Jan 5, 2022

stalary commented Jan 8, 2022

stalary commented Jan 8, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

morningman left a comment

Choose a reason for hiding this comment

github-actions bot commented Feb 7, 2022

github-actions bot commented Feb 7, 2022

morningman commented Feb 7, 2022

yangzhg commented Dec 29, 2021 •

edited

Loading