Support implement UDF through GRPC protocol. 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 cause doris coredump, 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 implement `check_fn` and `fn_call` methods
Note:
THIS IS AN EXPERIMENTAL FEATURE, THE INTERFACE AND DATA STRUCTURE MAY BE CHANGED IN FUTURE !!!
This PR mainly changes:
1. Fix bug when enable `transfer_data_by_brpc_attachment`
In `data_stream_sender`, we will send a serialized PRowBatch data to multiple Channels.
And if `transfer_data_by_brpc_attachment` is enabled, we will mistakenly clear the data in PRowBatch
after sending PRowBatch to the first Channel.
As a result, the following Channel cannot receive the correct data, causing an error.
So I use a separate buffer instead of `tuple_data` in PRowBatch to store the serialized data
and reuse it in multiple channels.
2. Fix bug that the the offset in serialized row batch may overflow
Use int64 to replace int32 offset. And for compatibility, add a new field `new_tuple_offsets` in PRowBatch.
The tuple ids of the empty set node must be exactly the same as the tuple ids of the origin root node.
In the issue, we found that once the tree where the root node is located has a window function,
the tuple ids of the empty set node cannot be calculated correctly.
This pr mostly fixes the problem.
In order to calculate the correct tuple ids,
the tuple ids obtained from the SelectStmt.getMaterializedTupleIds() function in the past
are changed to directly use the tuple ids of the origin root node.
Although we tried to fix#7929 by modifying the SelectStmt.getMaterializedTupleIds() function,
this method can't get the tuple of the last correct window function.
So we use other ways to construct tupleids of empty nodes.
Currently, if we encounter a problem with a replica of a tablet during the load process,
such as a write error, rpc error, -235, etc., it will cause the entire load job to fail,
which results in a significant reduction in Doris' fault tolerance.
This PR mainly changes:
1. refined the judgment of failed replicas in the load process, so that the failure of a few replicas will not affect the normal completion of the load job.
2. fix a bug introduced from #7754 that may cause BE coredump
This PR mainly changes:
1. Help to Cancel the load job ASAP when encounter unqualified data.
Solution is described in #6318 .
Also replace some std::stringstream with fmt::memory_buffer to avoid performance issues.
2. fix a NPE bug when create user with empty host
3. fix compile warning after rebasing the master(vectorization)
# Proposed changes
Issue Number: close#6238
Co-authored-by: HappenLee <happenlee@hotmail.com>
Co-authored-by: stdpain <34912776+stdpain@users.noreply.github.com>
Co-authored-by: Zhengguo Yang <yangzhgg@gmail.com>
Co-authored-by: wangbo <506340561@qq.com>
Co-authored-by: emmymiao87 <522274284@qq.com>
Co-authored-by: Pxl <952130278@qq.com>
Co-authored-by: zhangstar333 <87313068+zhangstar333@users.noreply.github.com>
Co-authored-by: thinker <zchw100@qq.com>
Co-authored-by: Zeno Yang <1521564989@qq.com>
Co-authored-by: Wang Shuo <wangshuo128@gmail.com>
Co-authored-by: zhoubintao <35688959+zbtzbtzbt@users.noreply.github.com>
Co-authored-by: Gabriel <gabrielleebuaa@gmail.com>
Co-authored-by: xinghuayu007 <1450306854@qq.com>
Co-authored-by: weizuo93 <weizuo@apache.org>
Co-authored-by: yiguolei <guoleiyi@tencent.com>
Co-authored-by: anneji-dev <85534151+anneji-dev@users.noreply.github.com>
Co-authored-by: awakeljw <993007281@qq.com>
Co-authored-by: taberylyang <95272637+taberylyang@users.noreply.github.com>
Co-authored-by: Cui Kaifeng <48012748+azurenake@users.noreply.github.com>
## Problem Summary:
### 1. Some code from clickhouse
**ClickHouse is an excellent implementation of the vectorized execution engine database,
so here we have referenced and learned a lot from its excellent implementation in terms of
data structure and function implementation.
We are based on ClickHouse v19.16.2.2 and would like to thank the ClickHouse community and developers.**
The following comment has been added to the code from Clickhouse, eg:
// This file is copied from
// https://github.com/ClickHouse/ClickHouse/blob/master/src/Interpreters/AggregationCommon.h
// and modified by Doris
### 2. Support exec node and query:
* vaggregation_node
* vanalytic_eval_node
* vassert_num_rows_node
* vblocking_join_node
* vcross_join_node
* vempty_set_node
* ves_http_scan_node
* vexcept_node
* vexchange_node
* vintersect_node
* vmysql_scan_node
* vodbc_scan_node
* volap_scan_node
* vrepeat_node
* vschema_scan_node
* vselect_node
* vset_operation_node
* vsort_node
* vunion_node
* vhash_join_node
You can run exec engine of SSB/TPCH and 70% TPCDS stand query test set.
### 3. Data Model
Vec Exec Engine Support **Dup/Agg/Unq** table, Support Block Reader Vectorized.
Segment Vec is working in process.
### 4. How to use
1. Set the environment variable `set enable_vectorized_engine = true; `(required)
2. Set the environment variable `set batch_size = 4096; ` (recommended)
### 5. Some diff from origin exec engine
https://github.com/doris-vectorized/doris-vectorized/issues/294
## Checklist(Required)
1. Does it affect the original behavior: (No)
2. Has unit tests been added: (Yes)
3. Has document been added or modified: (No)
4. Does it need to update dependencies: (No)
5. Are there any changes that cannot be rolled back: (Yes)
If an load task has a relatively short timeout, then we need to ensure that
each RPC of this task does not get blocked for a long time.
And an RPC is usually blocked for two reasons.
1. handling "memory exceeds limit" in the RPC
If the system finds that the memory occupied by the load exceeds the threshold,
it will select the load channel that occupies the most memory and flush the memtable in it.
this operation is done in the RPC, which may be more time consuming.
2. close the load channel
When the load channel receives the last batch, it will end the task.
It will wait for all memtables flushes to finish synchronously. This process is also time consuming.
Therefore, this PR solves this problem by.
1. Use timeout to determine whether it is a high-priority load task
If the timeout of an load task is relatively short, then we mark it as a high-priority task.
2. not processing "memory exceeds limit" for high priority tasks
3. use a separate flush thread to flush memtable for high priority tasks.
Support merge IN predicate when exist remote target(e.g. shuffle hash join).
Remote the code that IN predicate implicit conversion to Bloom filter then exist remote target.
Close related #7546
If the load result set is empty, or the load data is all filtered by the `where` condition,
it will not return failed with msg `all partitions have no load data`, but will return success directly.
For the first, we need to make a parameter to discribe the data is local or remote.
At then, we need to support some basic function to support the operation for remote storage.
1. Refactor the scheduling logic of broker load. Details see #7367
2. Fix bug that loadedBytes in SHOW LOAD result is wrong.
3. Cancel the thread of LoadTimeoutChecker
Now for PENDING load jobs, there will be no timeout. And the timeout of a load job
start when pending load task is scheduled.
4. Fix a bug that the loading task is never submitted to the pool.
The logic of BlockedPolicy is wrong. We should make sure the task is submitted to the pool,
or the RejectedExecutionException should be thrown.
5. Now the transaction of a load job will begin in pending task, instead of when submitting the job.
1. Delete useless variables
2. Add const modifier for read-only function
3. Delete the empty destructor, the compiler will automatically generate it, refer to the 3/5/0 rule:
[https://en.cppreference.com/w/cpp/language/rule_of_three]
4. It is recommended to add the override keyword (instead of the virtual keyword) to the subclass virtual function.
Override will let the compiler help check and improve security. This is also the reason why C++11 introduces override
1. Fix some memory leaks
2. Remove redundant and invalid code
3. Fix some buggy writes to reduce extra memory copies and return null pointers to string
4. Reframing the naming to make the structure clearer
code refactor: improve code's readability, avoid const_cast
1. make loop simpler and clearer by using range-based loop grammar, it's safer than old loop style
2. iteration for _row_desc.tuple_descriptors() use index replace index and iterator mixed
3. add new function To cast_to(From from), use this union-based casting between two types to replace reinterpret_cast, this new cast is more readable
4. avoid using the same variable name for nested loop, it's dangerous
5. add const keyword for member functions followed CppCoreGuidelines
the tuple String Slot's ptr and len are not assigned appropriately on send side, the receive side may crash in some situation.
detail description:
on send side, when we call RowBatch::serialize(PRowBatch* output_batch) to pack RowBatch, the Tuple::deep_copy()
will be called, for each String Slot, only String Slots that is not null will set ptr and len with proper value, the null String
Slots will keep original status, the ptr member will point randomly and the len member may unexpect.
on recv side, unpack is processed by RowBatch::RowBatch(const RowDescriptor&, const PRowBatch&...), in this
function, each String Slot will transfer offset to valid string_val->ptr whether the String Slot is null or not.
but some business logic depends on string_val->len=0, such as AggregateFuncTraits::init(), HyperLogLog::deserialize()
will return correctly if slice.size<=0. so if string_val->len is set to 0 in send side, everything will be ok, otherwise server
may crash.
by netcomm viewpoint, we should make sure transfer correct data, it's sender's responsibility to set data with proper
value, and do not make any presume which way the recv side will use it.
Transfer RowBatch in Protobuf Request to Controller Attachment,
when the maximum length of the RowBatch in the Protobuf Request is exceeded.
This can avoid reaching the upper limit of the Protobuf Request length (2G),
and it is expected that performance can be improved.
This is beacuse of an const MAX_PHYSICAL_PACKET_LENGTH in fe should be 2^24 -1,
but it is set as 2^24 -2 by mistake.
2. Fix bitmap_to_string may failed when the result is large than 2G
1. setting _report_thread_active to false is not necessary protected by _report_thread_lock, because
_report_thread_active's type is bool, writing data is multi-threadly safety if size <= marchine word length
2. report_profile thread terminates early is possiable, in the function report_profile(), while (_report_thread_active) may
break if _report_thread_active is false, the thread of calling open() may be scheduled out between
_report_thread_started_cv.wait(l) and _report_thread_active = true, we should not assume that how long time elapsed
between a thread be scheduled twice
Now minidump file will be created when BE crashes.
And user can manually trigger a minidump by sending SIGUSR1 to BE process.
More details can be found in minidump.md documents
BitShufflePageDecoder reuses the memory for storing decoder results, allocate memory directly from the
`ChunkAllocator`, the performance is improved to a certain extent.
In the case of #6285, the total time consumption is reduced by 13.5%, and the time consumption ratio of `~Reader()`
has also been reduced from 17.65% to 1.53%, and the memory allocation is unified to `ChunkAllocator` for centralized
management , Which is conducive to subsequent memory optimization.
which can avoid the memory waste caused by `Mempool`, because the chunk can be free at any time, but the
performance is lower than the allocation from `Mempool`. The guess is that there is no `Mempool` after secondary
allocation of large chunks , Will directly apply for a large number of small chunks from `ChunkAllocator`, and it takes
longer to lock in `pop_free_chunk` and `push_free_chunk` (but this is not proven from the flame graphs of BE's cpu and
contention).
1. replace all boost::shared_ptr to std::shared_ptr
2. replace all boost::scopted_ptr to std::unique_ptr
3. replace all boost::scoped_array to std::unique<T[]>
4. replace all boost:thread to std::thread
