Currently, BE will print fail to get master client from cache. host=xxxxx, port=9228, code=THRIFT_RPC_ERROR but we did not know which step generate this error. So that I refactor error status in be and add error stack for RPC_ERROR.
W1122 10:19:21.130796 30405 utils.cpp:89] fail to get master client from cache. host=xxxx, port=9228, code=RPC error(error -1): Couldn't open transport for xxxx:9228 (open() timed out)/n @ 0x559af8f774ea doris::Status::ConstructErrorStatus()
@ 0x559af9aacbee _ZN5doris16ThriftClientImpl4openEv.cold
@ 0x559af97f563a doris::ClientCacheHelper::_create_client()
@ 0x559af97f78cd doris::ClientCacheHelper::get_client()
@ 0x559af934f38b doris::MasterServerClient::report()
@ 0x559af932e7a7 doris::TaskWorkerPool::_handle_report()
@ 0x559af932f07c doris::TaskWorkerPool::_report_task_worker_thread_callback()
@ 0x559af9b223c5 doris::ThreadPool::dispatch_thread()
@ 0x559af9b187af doris::Thread::supervise_thread()
@ 0x7f661bd8bea5 start_thread
@ 0x7f661c09eb0d __clone
Co-authored-by: yiguolei <yiguolei@gmail.com>
When some columns specified are not in the parquet or orc file in broker load, _batch->num_columns() will less than _num_of_columns_from_file. It will lead to be core dump.
To prevent be core dump, just return an error in this case.
It is a painful work to read profile, especially there are multi-parallel instances.
This tool helps us to grasp the main information of profile in a graphical view.
The profile is represented by a tree.
Sql operation nodes contains operation type(join, scan...), its node id, its fragment id. The number on the arrow edge means how many rows output by child node. This tool will sum the output rows of the same node in multi-parallel instances, that is if there are 4 parallel instance, and each ScanNode on lineitem table output 10 rows, the label on the arrow beginning with ScanNode(lineitem) is 40.
Here is a demo for tpch Q2
tpch q2 profile viewer
Issue Number: close #xxx
Fix Thread pool token was shut down error.
This is because when there are more than 1 fragment of a query on one BE, the thread token maybe
reset incorrectly, causing thread token shutdown earlier.
cherry-pick from master
Introduced from #13021
* [enhancemeng](sql) Remove unused wide common factors to improve scan performance in ExtractCommonFactorsRule
* fix regression test
Co-authored-by: caiconghui1 <caiconghui1@jd.com>
Originally, Order By Limit returned a maximum of 65535 rows of data by default during the query,
but now many businesses do not apply this limit.
It is necessary to add larger data after the query statement to complete the full data query,
which is extremely inconvenient, so adjustments have been made.
At the same time, I added the variable DEFAULT_ORDER_BY_LIMIT to the SessionVariable,
the default value is -1, if the user does not use the LIMIT keyword or the LIMIT value is a negative integer,
the default query return value is Long.MAX_VALUE. If the corresponding maximum query value is set,
the number of data items is returned according to the maximum query value or the value followed by the
LIMIT keyword.
For the table name check, the regular match error and the length exceeds the limit, both of which display the message "Incorrect table name 'xxx'. Table name regex is 'xxx'".
Obviously, the message cannot clearly point out what kind of error it is.
So it is a better way to separate the two error messages.
