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
- Rename generic file and identifier names in library-bridge to
something more dictionary-specific. This is needed because later on,
catboost will be integrated into library-bridge.
- Also: Some smaller fixes like typos and un-inlining non-performance
critical code.
- The logic remains unchanged in this commit.
- Add status log message
- Add it to clickhouse-bundle in shared build
- Move clickhouse-su.cpp into su.cpp, since executable does not have
include directories of linked libraries (dbms here), only
clickhouse-lib-su does, hence it cannot find includes
CI: https://github.com/ClickHouse/ClickHouse/runs/7566319416?check_suite_focus=true
Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
A simple HelloWorld program with zero includes except iostream triggers
a build of ca. 2000 source files. The reason is that ClickHouse's
top-level CMakeLists.txt overrides "add_executable()" to link all
binaries against "clickhouse_new_delete". This links against
"clickhouse_common_io", which in turn has lots of 3rd party library
dependencies ... Without linking "clickhouse_new_delete", the number of
compiled files for "HelloWorld" goes down to ca. 70.
As an example, the self-extracting-executable needs none of its current
dependencies but other programs may also benefit.
In order to restore access to the original "add_executable()", the
overriding version is now prefixed. There is precedence for a
"clickhouse_" prefix (as opposed to "ch_"), for example
"clickhouse_split_debug_symbols". In general prefixing makes sense also
because overriding CMake commands relies on undocumented behavior and is
considered not-so-great practice (*).
(*) https://crascit.com/2018/09/14/do-not-redefine-cmake-commands/
* initial commit: add setting and stub
* typo
* added test stub
* fix
* wip merging new integration test and code proto
* adding steps interpreters
* adding firstly proposed solution (moving parts etc)
* added checking zookeeper path existence
* fixing the include
* fixing and sorting includes
* fixing outdated struct
* fix the name
* added ast ptr as level of indirection
* fix ref
* updating the changes
* working on test stub
* fix iterator -> reference
* revert rocksdb submodule update
* fixed show privileges test
* updated the test stub
* replaced rand() with thread_local_rng(), updated the tests
updated the test
fixed test config path
test fix
removed error messages
fixed the test
updated the test
fixed string literal
fixed literal
typo: =
* fixed the empty replica error message
* updated the test and the code with logs
* updated the possible test cases, updated
* added the code/test milestone comments
* updated the test (added more testcases)
* replaced native assert with CH one
* individual replicas recursive delete fix
* updated the AS db.name AST
* two small logging fixes
* manually generated AST fixes
* Updated the test, added the possible algo change
* Some thoughts about optimizing the solution:
ALTER MOVE PARTITION .. TO TABLE -> move to detached/ + ALTER ... ATTACH
* fix
* Removed the replica sync in test as it's invalid
* Some test tweaks
* tmp
* Rewrote the algo by using the executeQuery instead of
hand-crafting the ASTPtr.
Two questions still active.
* tr: logging active parts
* Extracted the parts moving algo into a separate helper function
* Fixed the test data and the queries slightly
* Replaced query to system.parts to direct invocation,
started building the test that breaks on various parts.
* Added the case for tables when at least one replica is alive
* Updated the test to test replicas restoration by detaching/attaching
* Altered the test to check restoration without replica restart
* Added the tables swap in the start if the server failed last time
* Hotfix when only /replicas/replica... path was deleted
* Restore ZK paths while creating a replicated MergeTree table
* Updated the docs, fixed the algo for individual replicas restoration case
* Initial parts table storage fix, tests sync fix
* Reverted individual replica restoration to general algo
* Slightly optimised getDataParts
* Trying another solution with parts detaching
* Rewrote algo without any steps, added ON CLUSTER support
* Attaching parts from other replica on restoration
* Getting part checksums from ZK
* Removed ON CLUSTER, finished working solution
* Multiple small changes after review
* Fixing parallel test
* Supporting rewritten form on cluster
* Test fix
* Moar logging
* Using source replica as checksum provider
* improve test, remove some code from parser
* Trying solution with move to detached + forget
* Moving all parts (not only Committed) to detached
* Edited docs for RESTORE REPLICA
* Re-merging
* minor fixes
Co-authored-by: Alexander Tokmakov <avtokmakov@yandex-team.ru>
