2021-03-05 09:38:00 +00:00
|
|
|
#pragma once
|
|
|
|
|
feat: implement catboost in library-bridge
This commit moves the catboost model evaluation out of the server
process into the library-bridge binary. This serves two goals: On the
one hand, crashes / memory corruptions of the catboost library no longer
affect the server. On the other hand, we can forbid loading dynamic
libraries in the server (catboost was the last consumer of this
functionality), thus improving security.
SQL syntax:
SELECT
catboostEvaluate('/path/to/model.bin', FEAT_1, ..., FEAT_N) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
Required configuration:
<catboost_lib_path>/path/to/libcatboostmodel.so</catboost_lib_path>
*** Implementation Details ***
The internal protocol between the server and the library-bridge is
simple:
- HTTP GET on path "/extdict_ping":
A ping, used during the handshake to check if the library-bridge runs.
- HTTP POST on path "extdict_request"
(1) Send a "catboost_GetTreeCount" request from the server to the
bridge, containing a library path (e.g /home/user/libcatboost.so) and
a model path (e.g. /home/user/model.bin). Rirst, this unloads the
catboost library handler associated to the model path (if it was
loaded), then loads the catboost library handler associated to the
model path, then executes GetTreeCount() on the library handler and
finally sends the result back to the server. Step (1) is called once
by the server from FunctionCatBoostEvaluate::getReturnTypeImpl(). The
library path handler is unloaded in the beginning because it contains
state which may no longer be valid if the user runs
catboost("/path/to/model.bin", ...) more than once and if "model.bin"
was updated in between.
(2) Send "catboost_Evaluate" from the server to the bridge, containing
the model path and the features to run the interference on. Step (2)
is called multiple times (once per chunk) by the server from function
FunctionCatBoostEvaluate::executeImpl(). The library handler for the
given model path is expected to be already loaded by Step (1).
Fixes #27870
2022-08-05 07:53:06 +00:00
|
|
|
#include <Common/logger_useful.h>
|
2021-03-05 09:38:00 +00:00
|
|
|
#include <Interpreters/Context.h>
|
|
|
|
#include <Server/HTTP/HTTPRequestHandler.h>
|
|
|
|
|
|
|
|
|
|
|
|
namespace DB
|
|
|
|
{
|
|
|
|
|
2021-04-05 13:13:07 +00:00
|
|
|
/// Handler for requests to Library Dictionary Source, returns response in RowBinary format.
|
2022-08-03 11:19:13 +00:00
|
|
|
/// When a library dictionary source is created, it sends 'extDict_libNew' request to library bridge (which is started on first
|
2021-04-05 13:13:07 +00:00
|
|
|
/// request to it, if it was not yet started). On this request a new sharedLibrayHandler is added to a
|
2022-08-03 11:19:13 +00:00
|
|
|
/// sharedLibraryHandlerFactory by a dictionary uuid. With 'extDict_libNew' request come: library_path, library_settings,
|
2021-04-05 13:13:07 +00:00
|
|
|
/// names of dictionary attributes, sample block to parse block of null values, block of null values. Everything is
|
|
|
|
/// passed in binary format and is urlencoded. When dictionary is cloned, a new handler is created.
|
|
|
|
/// Each handler is unique to dictionary.
|
2022-08-05 07:57:47 +00:00
|
|
|
class ExternalDictionaryLibraryBridgeRequestHandler : public HTTPRequestHandler, WithContext
|
2021-03-05 09:38:00 +00:00
|
|
|
{
|
|
|
|
public:
|
2022-08-05 07:57:47 +00:00
|
|
|
ExternalDictionaryLibraryBridgeRequestHandler(size_t keep_alive_timeout_, ContextPtr context_);
|
2021-03-05 09:38:00 +00:00
|
|
|
|
2024-01-03 16:47:15 +00:00
|
|
|
void handleRequest(HTTPServerRequest & request, HTTPServerResponse & response, const ProfileEvents::Event & write_event) override;
|
2021-03-05 09:38:00 +00:00
|
|
|
|
|
|
|
private:
|
2021-03-10 18:02:43 +00:00
|
|
|
static constexpr inline auto FORMAT = "RowBinary";
|
|
|
|
|
feat: implement catboost in library-bridge
This commit moves the catboost model evaluation out of the server
process into the library-bridge binary. This serves two goals: On the
one hand, crashes / memory corruptions of the catboost library no longer
affect the server. On the other hand, we can forbid loading dynamic
libraries in the server (catboost was the last consumer of this
functionality), thus improving security.
SQL syntax:
SELECT
catboostEvaluate('/path/to/model.bin', FEAT_1, ..., FEAT_N) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
Required configuration:
<catboost_lib_path>/path/to/libcatboostmodel.so</catboost_lib_path>
*** Implementation Details ***
The internal protocol between the server and the library-bridge is
simple:
- HTTP GET on path "/extdict_ping":
A ping, used during the handshake to check if the library-bridge runs.
- HTTP POST on path "extdict_request"
(1) Send a "catboost_GetTreeCount" request from the server to the
bridge, containing a library path (e.g /home/user/libcatboost.so) and
a model path (e.g. /home/user/model.bin). Rirst, this unloads the
catboost library handler associated to the model path (if it was
loaded), then loads the catboost library handler associated to the
model path, then executes GetTreeCount() on the library handler and
finally sends the result back to the server. Step (1) is called once
by the server from FunctionCatBoostEvaluate::getReturnTypeImpl(). The
library path handler is unloaded in the beginning because it contains
state which may no longer be valid if the user runs
catboost("/path/to/model.bin", ...) more than once and if "model.bin"
was updated in between.
(2) Send "catboost_Evaluate" from the server to the bridge, containing
the model path and the features to run the interference on. Step (2)
is called multiple times (once per chunk) by the server from function
FunctionCatBoostEvaluate::executeImpl(). The library handler for the
given model path is expected to be already loaded by Step (1).
Fixes #27870
2022-08-05 07:53:06 +00:00
|
|
|
const size_t keep_alive_timeout;
|
2021-03-05 09:38:00 +00:00
|
|
|
Poco::Logger * log;
|
|
|
|
};
|
|
|
|
|
|
|
|
|
feat: implement catboost in library-bridge
This commit moves the catboost model evaluation out of the server
process into the library-bridge binary. This serves two goals: On the
one hand, crashes / memory corruptions of the catboost library no longer
affect the server. On the other hand, we can forbid loading dynamic
libraries in the server (catboost was the last consumer of this
functionality), thus improving security.
SQL syntax:
SELECT
catboostEvaluate('/path/to/model.bin', FEAT_1, ..., FEAT_N) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
Required configuration:
<catboost_lib_path>/path/to/libcatboostmodel.so</catboost_lib_path>
*** Implementation Details ***
The internal protocol between the server and the library-bridge is
simple:
- HTTP GET on path "/extdict_ping":
A ping, used during the handshake to check if the library-bridge runs.
- HTTP POST on path "extdict_request"
(1) Send a "catboost_GetTreeCount" request from the server to the
bridge, containing a library path (e.g /home/user/libcatboost.so) and
a model path (e.g. /home/user/model.bin). Rirst, this unloads the
catboost library handler associated to the model path (if it was
loaded), then loads the catboost library handler associated to the
model path, then executes GetTreeCount() on the library handler and
finally sends the result back to the server. Step (1) is called once
by the server from FunctionCatBoostEvaluate::getReturnTypeImpl(). The
library path handler is unloaded in the beginning because it contains
state which may no longer be valid if the user runs
catboost("/path/to/model.bin", ...) more than once and if "model.bin"
was updated in between.
(2) Send "catboost_Evaluate" from the server to the bridge, containing
the model path and the features to run the interference on. Step (2)
is called multiple times (once per chunk) by the server from function
FunctionCatBoostEvaluate::executeImpl(). The library handler for the
given model path is expected to be already loaded by Step (1).
Fixes #27870
2022-08-05 07:53:06 +00:00
|
|
|
// Handler for checking if the external dictionary library is loaded (used for handshake)
|
2022-08-05 07:57:47 +00:00
|
|
|
class ExternalDictionaryLibraryBridgeExistsHandler : public HTTPRequestHandler, WithContext
|
2021-03-05 09:38:00 +00:00
|
|
|
{
|
|
|
|
public:
|
2022-08-05 07:57:47 +00:00
|
|
|
ExternalDictionaryLibraryBridgeExistsHandler(size_t keep_alive_timeout_, ContextPtr context_);
|
2021-03-05 09:38:00 +00:00
|
|
|
|
2024-01-03 16:47:15 +00:00
|
|
|
void handleRequest(HTTPServerRequest & request, HTTPServerResponse & response, const ProfileEvents::Event & write_event) override;
|
2021-03-05 09:38:00 +00:00
|
|
|
|
|
|
|
private:
|
2021-04-05 14:15:10 +00:00
|
|
|
const size_t keep_alive_timeout;
|
2021-07-30 14:20:57 +00:00
|
|
|
Poco::Logger * log;
|
2021-03-05 09:38:00 +00:00
|
|
|
};
|
|
|
|
|
feat: implement catboost in library-bridge
This commit moves the catboost model evaluation out of the server
process into the library-bridge binary. This serves two goals: On the
one hand, crashes / memory corruptions of the catboost library no longer
affect the server. On the other hand, we can forbid loading dynamic
libraries in the server (catboost was the last consumer of this
functionality), thus improving security.
SQL syntax:
SELECT
catboostEvaluate('/path/to/model.bin', FEAT_1, ..., FEAT_N) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
Required configuration:
<catboost_lib_path>/path/to/libcatboostmodel.so</catboost_lib_path>
*** Implementation Details ***
The internal protocol between the server and the library-bridge is
simple:
- HTTP GET on path "/extdict_ping":
A ping, used during the handshake to check if the library-bridge runs.
- HTTP POST on path "extdict_request"
(1) Send a "catboost_GetTreeCount" request from the server to the
bridge, containing a library path (e.g /home/user/libcatboost.so) and
a model path (e.g. /home/user/model.bin). Rirst, this unloads the
catboost library handler associated to the model path (if it was
loaded), then loads the catboost library handler associated to the
model path, then executes GetTreeCount() on the library handler and
finally sends the result back to the server. Step (1) is called once
by the server from FunctionCatBoostEvaluate::getReturnTypeImpl(). The
library path handler is unloaded in the beginning because it contains
state which may no longer be valid if the user runs
catboost("/path/to/model.bin", ...) more than once and if "model.bin"
was updated in between.
(2) Send "catboost_Evaluate" from the server to the bridge, containing
the model path and the features to run the interference on. Step (2)
is called multiple times (once per chunk) by the server from function
FunctionCatBoostEvaluate::executeImpl(). The library handler for the
given model path is expected to be already loaded by Step (1).
Fixes #27870
2022-08-05 07:53:06 +00:00
|
|
|
|
|
|
|
/// Handler for requests to catboost library. The call protocol is as follows:
|
2022-09-11 15:50:36 +00:00
|
|
|
/// (1) Send a "catboost_GetTreeCount" request from the server to the bridge. It contains a library path (e.g /home/user/libcatboost.so) and
|
|
|
|
/// a model path (e.g. /home/user/model.bin). This loads the catboost library handler associated with the model path, then executes
|
|
|
|
/// GetTreeCount() on the library handler and sends the result back to the server.
|
|
|
|
/// (2) Send "catboost_Evaluate" from the server to the bridge. It contains a model path and the features to run the interference on. Step
|
|
|
|
/// (2) is called multiple times (once per chunk) by the server.
|
|
|
|
///
|
|
|
|
/// We would ideally like to have steps (1) and (2) in one atomic handler but can't because the evaluation on the server side is divided
|
|
|
|
/// into two dependent phases: FunctionCatBoostEvaluate::getReturnTypeImpl() and ::executeImpl(). So the model may in principle be unloaded
|
|
|
|
/// from the library-bridge between steps (1) and (2). Step (2) checks if that is the case and fails gracefully. This is okay because that
|
|
|
|
/// situation considered exceptional and rare.
|
|
|
|
///
|
|
|
|
/// An update of a model is performed by unloading it. The first call to "catboost_GetTreeCount" brings it into memory again.
|
|
|
|
///
|
|
|
|
/// Further handlers are provided for unloading a specific model, for unloading all models or for retrieving information about the loaded
|
|
|
|
/// models for display in a system view.
|
feat: implement catboost in library-bridge
This commit moves the catboost model evaluation out of the server
process into the library-bridge binary. This serves two goals: On the
one hand, crashes / memory corruptions of the catboost library no longer
affect the server. On the other hand, we can forbid loading dynamic
libraries in the server (catboost was the last consumer of this
functionality), thus improving security.
SQL syntax:
SELECT
catboostEvaluate('/path/to/model.bin', FEAT_1, ..., FEAT_N) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
Required configuration:
<catboost_lib_path>/path/to/libcatboostmodel.so</catboost_lib_path>
*** Implementation Details ***
The internal protocol between the server and the library-bridge is
simple:
- HTTP GET on path "/extdict_ping":
A ping, used during the handshake to check if the library-bridge runs.
- HTTP POST on path "extdict_request"
(1) Send a "catboost_GetTreeCount" request from the server to the
bridge, containing a library path (e.g /home/user/libcatboost.so) and
a model path (e.g. /home/user/model.bin). Rirst, this unloads the
catboost library handler associated to the model path (if it was
loaded), then loads the catboost library handler associated to the
model path, then executes GetTreeCount() on the library handler and
finally sends the result back to the server. Step (1) is called once
by the server from FunctionCatBoostEvaluate::getReturnTypeImpl(). The
library path handler is unloaded in the beginning because it contains
state which may no longer be valid if the user runs
catboost("/path/to/model.bin", ...) more than once and if "model.bin"
was updated in between.
(2) Send "catboost_Evaluate" from the server to the bridge, containing
the model path and the features to run the interference on. Step (2)
is called multiple times (once per chunk) by the server from function
FunctionCatBoostEvaluate::executeImpl(). The library handler for the
given model path is expected to be already loaded by Step (1).
Fixes #27870
2022-08-05 07:53:06 +00:00
|
|
|
class CatBoostLibraryBridgeRequestHandler : public HTTPRequestHandler, WithContext
|
|
|
|
{
|
|
|
|
public:
|
|
|
|
CatBoostLibraryBridgeRequestHandler(size_t keep_alive_timeout_, ContextPtr context_);
|
|
|
|
|
2024-01-03 16:47:15 +00:00
|
|
|
void handleRequest(HTTPServerRequest & request, HTTPServerResponse & response, const ProfileEvents::Event & write_event) override;
|
feat: implement catboost in library-bridge
This commit moves the catboost model evaluation out of the server
process into the library-bridge binary. This serves two goals: On the
one hand, crashes / memory corruptions of the catboost library no longer
affect the server. On the other hand, we can forbid loading dynamic
libraries in the server (catboost was the last consumer of this
functionality), thus improving security.
SQL syntax:
SELECT
catboostEvaluate('/path/to/model.bin', FEAT_1, ..., FEAT_N) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
Required configuration:
<catboost_lib_path>/path/to/libcatboostmodel.so</catboost_lib_path>
*** Implementation Details ***
The internal protocol between the server and the library-bridge is
simple:
- HTTP GET on path "/extdict_ping":
A ping, used during the handshake to check if the library-bridge runs.
- HTTP POST on path "extdict_request"
(1) Send a "catboost_GetTreeCount" request from the server to the
bridge, containing a library path (e.g /home/user/libcatboost.so) and
a model path (e.g. /home/user/model.bin). Rirst, this unloads the
catboost library handler associated to the model path (if it was
loaded), then loads the catboost library handler associated to the
model path, then executes GetTreeCount() on the library handler and
finally sends the result back to the server. Step (1) is called once
by the server from FunctionCatBoostEvaluate::getReturnTypeImpl(). The
library path handler is unloaded in the beginning because it contains
state which may no longer be valid if the user runs
catboost("/path/to/model.bin", ...) more than once and if "model.bin"
was updated in between.
(2) Send "catboost_Evaluate" from the server to the bridge, containing
the model path and the features to run the interference on. Step (2)
is called multiple times (once per chunk) by the server from function
FunctionCatBoostEvaluate::executeImpl(). The library handler for the
given model path is expected to be already loaded by Step (1).
Fixes #27870
2022-08-05 07:53:06 +00:00
|
|
|
|
|
|
|
private:
|
|
|
|
const size_t keep_alive_timeout;
|
|
|
|
Poco::Logger * log;
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
// Handler for pinging the library-bridge for catboost access (used for handshake)
|
|
|
|
class CatBoostLibraryBridgeExistsHandler : public HTTPRequestHandler, WithContext
|
|
|
|
{
|
|
|
|
public:
|
|
|
|
CatBoostLibraryBridgeExistsHandler(size_t keep_alive_timeout_, ContextPtr context_);
|
|
|
|
|
2024-01-03 16:47:15 +00:00
|
|
|
void handleRequest(HTTPServerRequest & request, HTTPServerResponse & response, const ProfileEvents::Event & write_event) override;
|
feat: implement catboost in library-bridge
This commit moves the catboost model evaluation out of the server
process into the library-bridge binary. This serves two goals: On the
one hand, crashes / memory corruptions of the catboost library no longer
affect the server. On the other hand, we can forbid loading dynamic
libraries in the server (catboost was the last consumer of this
functionality), thus improving security.
SQL syntax:
SELECT
catboostEvaluate('/path/to/model.bin', FEAT_1, ..., FEAT_N) > 0 AS prediction,
ACTION AS target
FROM amazon_train
LIMIT 10
Required configuration:
<catboost_lib_path>/path/to/libcatboostmodel.so</catboost_lib_path>
*** Implementation Details ***
The internal protocol between the server and the library-bridge is
simple:
- HTTP GET on path "/extdict_ping":
A ping, used during the handshake to check if the library-bridge runs.
- HTTP POST on path "extdict_request"
(1) Send a "catboost_GetTreeCount" request from the server to the
bridge, containing a library path (e.g /home/user/libcatboost.so) and
a model path (e.g. /home/user/model.bin). Rirst, this unloads the
catboost library handler associated to the model path (if it was
loaded), then loads the catboost library handler associated to the
model path, then executes GetTreeCount() on the library handler and
finally sends the result back to the server. Step (1) is called once
by the server from FunctionCatBoostEvaluate::getReturnTypeImpl(). The
library path handler is unloaded in the beginning because it contains
state which may no longer be valid if the user runs
catboost("/path/to/model.bin", ...) more than once and if "model.bin"
was updated in between.
(2) Send "catboost_Evaluate" from the server to the bridge, containing
the model path and the features to run the interference on. Step (2)
is called multiple times (once per chunk) by the server from function
FunctionCatBoostEvaluate::executeImpl(). The library handler for the
given model path is expected to be already loaded by Step (1).
Fixes #27870
2022-08-05 07:53:06 +00:00
|
|
|
|
|
|
|
private:
|
|
|
|
const size_t keep_alive_timeout;
|
|
|
|
Poco::Logger * log;
|
|
|
|
};
|
|
|
|
|
2021-03-05 09:38:00 +00:00
|
|
|
}
|