[improve](move-memtable) remove heavy work pool from load stream #28745
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
|
TeamCity be ut coverage result: |
|
TPC-H test result on machine: 'aliyun_ecs.c7a.8xlarge_32C64G' |
|
(From new machine)TeamCity pipeline, clickbench performance test result: |
|
PR approved by at least one committer and no changes requested. |
github-actions
bot
added
approved
|
PR approved by anyone and no changes requested. |
github-actions
bot
removed
the
approved
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
1 similar comment
|
clang-tidy review says "All clean, LGTM! 👍" |
|
TeamCity be ut coverage result: |
|
TPC-H test result on machine: 'aliyun_ecs.c7a.8xlarge_32C64G' |
|
(From new machine)TeamCity pipeline, clickbench performance test result: |
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
|
TPC-H test result on machine: 'aliyun_ecs.c7a.8xlarge_32C64G' |
|
TeamCity be ut coverage result: |
|
(From new machine)TeamCity pipeline, clickbench performance test result: |
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
|
TeamCity be ut coverage result: |
|
TPC-H test result on machine: 'aliyun_ecs.c7a.8xlarge_32C64G' |
|
(From new machine)TeamCity pipeline, clickbench performance test result: |
|
clang-tidy review says "All clean, LGTM! 👍" |
kaijchen
force-pushed
the
load-stream-no-heavy
branch
from
234b5b0
to
22bc123
Compare
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
|
TeamCity be ut coverage result: |
|
(From new machine)TeamCity pipeline, clickbench performance test result: |
|
TPC-H test result on machine: 'aliyun_ecs.c7a.8xlarge_32C64G' |
kaijchen
force-pushed
the
load-stream-no-heavy
branch
from
22bc123
to
d627659
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
kaijchen
force-pushed
the
load-stream-no-heavy
branch
from
d627659
to
c87d372
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
github-actions
bot
added
the
approved
|
PR approved by at least one committer and no changes requested. |
kaijchen
force-pushed
the
load-stream-no-heavy
branch
from
c87d372
to
11c69e8
Compare
|
run buildall |
github-actions
bot
removed
the
approved
|
clang-tidy review says "All clean, LGTM! 👍" |
|
TeamCity be ut coverage result: |
|
(From new machine)TeamCity pipeline, clickbench performance test result: |
github-actions
bot
added
the
approved
|
PR approved by at least one committer and no changes requested. |
|
TPC-H test result on machine: 'aliyun_ecs.c7a.8xlarge_32C64G' |
CalvinKirs
added a commit
to CalvinKirs/incubator-doris
that referenced
this pull request
Dec 27, 2023
* [Performance](point query)Opimize partition prune for point query (apache#28150) * [Performance](point query)Opimize partition prune for point query * [fix](stacktrace) ignore stacktrace for error code INVALID_ARGUMENT INVERTED_INDEX_NOT_IMPLEMENTED (apache#27898) (apache#28598) * ignore stacktrace for error INVALID_ARGUMENT INVERTED_INDEX_NOT_IMPLEMENTED * AndBlockColumnPredicate::evaluate * [bugfix](topn) fix coredump in copy_column_data_to_block when nullable mismatch (apache#28597) * [bugfix](topn) fix coredump in copy_column_data_to_block when nullable mismatch return RuntimeError if copy_column_data_to_block nullable mismatch to avoid coredump in input_col_ptr->filter_by_selector(sel_rowid_idx, select_size, raw_res_ptr) . The problem is reported by a doris user but I can not reproduce it, so there is no testcase added currently. * clang format * [feature](Nereids): eliminate semi join (apache#28588) Eliminate Semi/Anti Join which is FALSE or TRUE. * [feature](Nereids) support datev1 and datetimev1 (apache#28581) * [improvement](http) add show_table_data http api (apache#28380) In some cases, users need to get the data size of single replica of a table, and evaluate certain actions based on this, such as estimating the precise backup size. Signed-off-by: nextdreamblue <[email protected]> * [Enhance](regression)enhance jdbc case to adapt to use case concurrency (apache#28565) enhance jdbc case to adapt to use case concurrency * [opt](Nereids)when both Nereids and old parsers report errors, prompt error messages for the Nereids (apache#28580) * [Improvement](regression) change compound predicate regression case name to make it more clear (apache#28612) * [regression](memtable) add case for memtable flush error handle (apache#28285) Co-authored-by: ziyang zhang <[email protected]> * [regression-test](memtable) test memtable flush is high priority for vtable writerV1 (apache#28502) * enhance performance for broken tablet checking under multi-core scenario with a coarse-grained read lock (apache#28552) * [fix](load) fix memtracking orphan too large (apache#28600) * [fix](memtable-limiter) do not block write if load mem usage is low (apache#28602) Co-authored-by: Yongqiang YANG <[email protected]> * [refactor](renamefile) rename some files according to the class names (apache#28606) * [feature](mtmv)create mtmv support refresh_partition_num (apache#28566) - create/alter mtmv support refresh_partition_num - mtmv task according to refresh_partition_num executes refresh tasks in batches - `tasks` tvf add column `CompletedPartitions` and `progress` - fix mtmv can not `show temp partition` and `drop temp partition` - fix task can not get error msg when insert overwrite error - fix when the partition field is capitalized, the verification of creating a mtmv does not pass * [fix](Nereids) stats estimation of lessThan apache#28444 * [fix](regression)Change analyze_timeout to global. (apache#28587) Fix hive statistics regression case. analyze_timeout is a global session variable. * [chore] Add bvar for meta operations of BE (apache#28374) * [Bug](cooldown) Fix problem that followers may never completely cooldown (apache#28561) * [Improve](tvf)jni-avro support split file (apache#27933) * [feature](mtmv)after creating a materialized view, if other operations fail, roll back (apache#28621) after create mtmv, if create job failed,need drop mtmv * [fix](meta) update killed query state (#) (apache#25917) * [refactor](profile&names) using dst_id in pipelinex profile to be same as non pipeline; rename some function names (apache#28626) Co-authored-by: yiguolei <[email protected]> * [feature][executor]support workload schedule policy (apache#28443) * [feature](expr) Support kill query by query_id (apache#28530) Issue Number: open apache#28517 * [chore](user) Add user property `parallel_fragment_exec_instance_num` (apache#28447) * [feature](inverted index) add ignore_above property to prevent long s… (apache#28585) When string is too long, clucene will throw an error. And the string is too long to analyze. So we ignore the string in index process when the string is longer than 256 bytes by default. We add an poperty `ignore_above` for user to customize. * [Enhance](broker) add inputstream expire scheduled checker to avoid memory leak for broker scan (apache#28589) This pr introduces 2 broker conf: 1. enable_input_stream_expire_check: which indicates whether enable inputStream expire check. 2. input_stream_expire_seconds: which indicates the timeout seconds for inputStream since last update. * [fix](hive) add support for `quoteChar` and `seperatorChar` for hive (apache#28613) add support for quoteChar and seperatorChar . * [Fix](transactional-hive) Fix hive transactional table return empty result. (apache#28518) * [Fix](memtable) fix `shrink_memtable_by_agg` without duplicated keys (apache#28660) remove duplicated logic: ``` vectorized::Block in_block = _input_mutable_block.to_block(); _put_into_output(in_block); ``` `_input_mutable_block.to_block()` will move `_input_mutable_block`, and lead to `flush` with empty block * [enhance](partition_id) check partition id before store meta (apache#28055) * [opt](task-assignment) use consistent hash as default task assigner and cache the consistent hash ring (apache#28522) 1. Use consistent hash algo as the default assigner for file query scan node A consistent assignment can better utilize the page cache of BE node. 2. Cache the consistent hash ring Init a consistent hash ring is time-consuming because there a thousands of virtual node need to be added. So cache it for better performance * [Fix](Job)Fixed job scheduling missing certain time window schedules (apache#28659) Since scheduling itself consumes a certain amount of time, the start time of the time window should not be the current time, but the end time of the last schedule. * [fix](test) fix ccr test cases (apache#28664) * [fix](regression) fix test_set_replica_status due to force_olap_table_replication_num=3 (apache#28573) * [fix](regression) restore reserve num replicas (apache#28541) * [fix](regression) fix test_alter_colocate_table due to force_olap_table_replication_num=3 (apache#28575) * [improvement](transaction) reduce publish txn log (apache#28277) * [fix](stream_load)fix bug for stream (apache#27752) 1. forbid thed stream_load without content-length or chunked Transfer Encoding 2. forbid thed stream_load both with content-length and chunked Transfer Encoding Co-authored-by: xingying01 <[email protected]> * [Revert](partial update) Revert "Fix missing rowsets during doing alignment when flushing memtable due to compaction (apache#28062)" (apache#28674) This reverts commit 485d7db. * [feat](Nereids) support outer join and aggregate bitmap rewrite by mv (apache#28596) - Support left outer join rewrite by materialized view - Support bitmap_union roll up to imp count(distinct) - Support partition materialized view rewrite * [refactor](runtimefilter) do not use QueryContext in runtime filter (apache#28559) * [fix](planner)should bind expr using no slot to correct tuple (apache#28656) * [fix](planner) ctas update datev1 to datev2 should use equals (apache#28641) * [fix](Nereids) should only do bind relation in view analyzer (apache#28637) * [code](pipelineX) refine some pipelineX code (apache#28570) * [fix](partial update) report error directly if missing rowsets during doing alignment when flushing memtable due to compaction (apache#28677) * [opt](inverted index) Add null document interface to optimize empty string indexing (apache#28661) * [feature](invert index) match_regexp feature added (apache#28257) * [Feature](datatype) update be ut codes and fix bugs for IPv4/v6 (apache#28670) * [test](Nereids): remove shape plan project and distribute in eager test (apache#28701) * [fix](statistics)Fix drop stats fail silently bug. (apache#28635) Drop stats use IN predicate to filter the column stats to delete. The default length of IN predicate is 1024, drop table stats with more than 1024 columns may fail. This pr is to split the delete sql based on the IN predicate length. * [fix](mtmv) fix insert overwrite getExecTimeout error (apache#28700) should use InsertTimeout but QueryTimeout * [Enhancement](auto-partition) change the behaviour when insert overwrite an auto partition table apache#28683 If we specific target partition(s) when inserting overwrite an auto partition table, before: could create new partition now: behalf just like non-auto partition table * [FIX](regresstest) fix case with double value apache#28668 double value in case has 27 figures after the point, which will make not standable output * [test](Nereids): add test for scalar agg (apache#28712) * [fix](meta_scanner) fix meta_scanner process ColumnNullable (apache#28711) * [fix](nereids) Fix data wrong using mv rewrite and ignore case when getting mv related partition table (apache#28699) 1. Fix data wrong using mv rewrite 2. Ignore case when getting mv related partition table 3. Enable infer expression column name without alias when create mv * [test](regression-test) order by decs should only make effect on its nearest column apache#28728 * [refine](pipelineX)Make the 'set ready' logic of SenderQueue in pipelineX the same as that in the pipeline (apache#28488) * [fix](Nereids) delete partition failed (apache#28717) 1. parser's partitionSpec changed unexpectly by PR apache#26492 2. delete without using should support un-equals expression * [fix](tvf)Fixed the avro-scanner projection pushdown failing to query on multiple BEs (apache#28709) * [fix](block) fix nullptr in MutableBlock::allocated_bytes (apache#28738) * [fix](mtmv)fix thread local reference to checkpoint's Env, causing Env to be unable to be reclaimed, resulting in excessive memory usage by FE (apache#28723) when replay addTaskResult log,will create one ConnectContext,and set Env.getCurrentEnv,then store this ctx in ConnectContext.threadLocalInfo,threadLocalInfo is static,so this ctx can not be recycling，Env of replay thread also can not be recycling * [Fix](json reader) fix json reader crash due to `fmt::format_to` (apache#28737) ``` 4# __gnu_cxx::__verbose_terminate_handler() [clone .cold] at ../../../../libstdc++-v3/libsupc++/vterminate.cc:75 5# __cxxabiv1::__terminate(void (*)()) at ../../../../libstdc++-v3/libsupc++/eh_terminate.cc:48 6# 0x00005622F33D22B1 in /mnt/ssd01/pipline/OpenSourceDoris/clusterEnv/P0/Cluster0/be/lib/doris_be 7# 0x00005622F33D2404 in /mnt/ssd01/pipline/OpenSourceDoris/clusterEnv/P0/Cluster0/be/lib/doris_be 8# fmt::v7::detail::error_handler::on_error(char const*) in /mnt/ssd01/pipline/OpenSourceDoris/clusterEnv/P0/Cluster0/be/lib/doris_be 9# char const* fmt::v7::detail::parse_replacement_field<char, fmt::v7::detail::format_handler<fmt::v7::detail::buffer_appender<char>, char, fmt::v7::basic_format_context<fmt::v7::detail::buffer_appender<char>, char> >&>(char const*, char const*, fmt::v7::detail::format_handler<fmt::v7::detail::buffer_appender<char>, char, fmt::v7::basic_format_context<fmt::v7::detail::buffer_appender<char>, char> >&) in /mnt/ssd01/pipline/OpenSourceDoris/clusterEnv/P0/Cluster0/be/lib/doris_be 10# void fmt::v7::detail::vformat_to<char>(fmt::v7::detail::buffer<char>&, fmt::v7::basic_string_view<char>, fmt::v7::basic_format_args<fmt::v7::basic_format_context<fmt::v7::detail::buffer_appender<fmt::v7::type_identity<char>::type>, fmt::v7::type_identity<char>::type> >, fmt::v7::detail::locale_ref) in /mnt/ssd01/pipline/OpenSourceDoris/clusterEnv/P0/Cluster0/be/lib/doris_be 11# doris::vectorized::NewJsonReader::_append_error_msg(rapidjson::GenericValue<rapidjson::UTF8<char>, rapidjson::MemoryPoolAllocator<rapidjson::CrtAllocator> > const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool*) at /root/doris/be/src/vec/exec/format/json/new_json_reader.cpp:924 12# doris::vectorized::NewJsonReader::_set_column_value ``` * [bug](coredump) Fix coredump in aggregation node's destruction(apache#28684) fix coredump in aggregation node's destruction * [pipelineX](improvement) Support global runtime filter (apache#28692) * [FIX](explode)fix explode array decimal (apache#28744) * fix explode with array<decimal> has specific precision at old planner * [refactor](broadcastbuffer) using a queue to remove ref and unref codes (apache#28698) Co-authored-by: yiguolei <[email protected]>Add a new class broadcastbufferholderqueue to manage holders Using shared ptr to manage holders, not use ref and unref, it is too difficult to maintain. * [enhance](blacklist) seperate blacklist conf from heartbeat (apache#28638) There is a circuit breaker lasting for 2 minutes in grpc, then if a be is down and up again, send fragments to the be fails lasting for 2 minutes. * [fix](stmt):fix CreateTableStmt toSql placed comment in wrong place (apache#27504) Issue Number: close apache#27474 Co-authored-by: tongyang.han <[email protected]> * [Improve](compile) add `__AVX2__` macro for JsonbParser (apache#28754) * [Improve](compile) add `__AVX2__` macro for JsonbParser * throw exception instead of CHECK * [Feature](materialized-view) support match logicalAggregate(logicalProject(logicalFilter(logicalOlapScan())) without agg (apache#28747) support match logicalAggregate(logicalProject(logicalFilter(logicalOlapScan())) without agg * [Fix] (schema change) forbid adding time type column (apache#28751) * [fix](group_commit) fix group commit cancel stuck (apache#28749) * [fix](scanner) fix concurrency bugs when scanner is stopped or finished (apache#28650) `ScannerContext` will schedule scanners even after stopped, and confused with `_is_finished` and `_should_stop`. Only Fix the concurrency bugs when scanner is stopped or finished reported in apache#28384 * [fix](regression) fix stream load properties case fail (apache#28680) * [fix](fe ut) fix PropertyConverterTest (apache#28722) * [improve](rpc) Log channel state before shutdown backend service client (apache#28667) * [fix](publish version) fix publish fail but return ok (apache#28425) * [test](Nereids): remove shape plan project and distribute in eager test (apache#28724) * [refactor](nereids) make NormalizeAggregate rule more clear and readable (apache#28607) * [FIX](type) fix matchExactType for complex type (apache#28233) fe matchExactType function should call type.matchTypes for its own logic, do not switch case to do special logic otherwise we may meet core in be like this. ``` F20231208 18:54:39.359673 680131 block.h:535] Check failed: _data_types[i]->is_nullable() target type: Struct(l_info:Nullable(Array(Nullable(String)))) src type: Struct(col:Nullable(Array(Nullable(UInt8)))) *** Check failure stack trace: *** @ 0x5584e952b926 google::LogMessage::SendToLog() @ 0x5584e9527ef0 google::LogMessage::Flush() @ 0x5584e952c169 google::LogMessageFatal::~LogMessageFatal() @ 0x5584cf17201e doris::vectorized::MutableBlock::merge_impl<>() @ 0x5584ceac4b1d doris::vectorized::MutableBlock::merge<>() @ 0x5584d4dd7de3 doris::vectorized::VUnionNode::get_next_const() @ 0x5584d4dd9a45 doris::vectorized::VUnionNode::get_next() @ 0x5584bce469bd std::__invoke_impl<>() @ 0x5584bce466d0 std::__invoke<>() @ 0x5584bce465c7 _ZNSt5_BindIFMN5doris8ExecNodeEFNS0_6StatusEPNS0_12RuntimeStateEPNS0_10vectorized5BlockEPbEPS1_St12_PlaceholderILi1EESC_ILi2EESC_ILi3EEEE6__callIS2_JOS4_OS7_OS8_EJLm0ELm1ELm2ELm3EEEET_OSt5tupleIJDpT0_EESt12_Index_tupleIJXspT1_EEE @ 0x5584bce46358 std::_Bind<>::operator()<>() @ 0x5584bce46208 std::__invoke_impl<>() @ 0x5584bce46178 _ZSt10__invoke_rIN5doris6StatusERSt5_BindIFMNS0_8ExecNodeEFS1_PNS0_12RuntimeStateEPNS0_10vectorized5BlockEPbEPS3_St12_PlaceholderILi1EESD_ILi2EESD_ILi3EEEEJS5_S8_S9_EENSt9enable_ifIX16is_invocable_r_vIT_T0_DpT1_EESL_E4typeEOSM_DpOSN_ @ 0x5584bce45c18 std::_Function_handler<>::_M_invoke() @ 0x5584bce6412f std::function<>::operator()() @ 0x5584bce56382 doris::ExecNode::get_next_after_projects() @ 0x5584bce26218 doris::PlanFragmentExecutor::get_vectorized_internal() @ 0x5584bce2431b doris::PlanFragmentExecutor::open_vectorized_internal() @ 0x5584bce22a96 doris::PlanFragmentExecutor::open() @ 0x5584bce27c9d doris::PlanFragmentExecutor::execute() @ 0x5584bcbdb3f8 doris::FragmentMgr::_exec_actual() @ 0x5584bcbf982f doris::FragmentMgr::exec_plan_fragment()::$_0::operator()() @ 0x5584bcbf9715 std::__invoke_impl<>() @ 0x5584bcbf96b5 _ZSt10__invoke_rIvRZN5doris11FragmentMgr18exec_plan_fragmentERKNS0_23TExecPlanFragmentParamsERKSt8functionIFvPNS0_12RuntimeStateEPNS0_6StatusEEEE3$_0JEENSt9enable_ifIX16is_invocable_r_vIT_T0_DpT1_EESH_E4typeEOSI_DpOSJ_ @ 0x5584bcbf942d std::_Function_handler<>::_M_invoke() @ 0x5584b9dfd883 std::function<>::operator()() @ 0x5584bd6e3929 doris::FunctionRunnable::run() @ 0x5584bd6cf8ce doris::ThreadPool::dispatch_thread() ``` * [test](Nereids): add assert debug log for TopnToMaxTest (apache#28755) * [improve](move-memtable) limit task num in load stream flush token (apache#28748) * [feature](Nereids) elimite inner join by foreign key (apache#28486) * [enhancement](backup/restore) support alter s3 repo info about ak/sk/token (apache#27027) in some cases: s3.session_token/AWS_TOKEN will be expired after hours, and may be upload snapshot job will failed if data is big large; as same reason, repo will be expired too when RepositoryMgr exec repo ping so it need support alter s3 repo properties about ak/sk/token and update upload snapshot job properties to continue backup. Signed-off-by: nextdreamblue <[email protected]> * [improve](transaction) extend abort transaction time (apache#28662) * [regression](p2) fix test cases result (apache#28768) regression-test/data/external_table_p2/hive/test_hive_hudi.out regression-test/data/external_table_p2/hive/test_hive_to_array.out regression-test/suites/external_table_p2/tvf/test_local_tvf_compression.groovy regression-test/suites/external_table_p2/tvf/test_path_partition_keys.groovy regression-test/data/external_table_p2/hive/test_hive_text_complex_type.out * (fix)[meta][export] fix replay export NPE issue (apache#28752) The ConnectionContext does not exist in replay thread * [Bug fix][metrics] correct fe collector type for jvm_gc (apache#28784) Co-authored-by: 胥剑旭 <[email protected]> * [Chore](decimal) set check_overflow_for_decimal to true when alter table(apache#28777) set check_overflow_for_decimal to true when alter table * [pipelineX](fix) Fix TPCH Q2 (apache#28783) * (topN)runtime_predicate is only triggered when the column name is obtained (apache#28419) Issue Number: close apache#27485 * [fix](planner)fix bug of bound conjunct to wrong tuple (apache#28811) this fix bug introduced by apache#28656 * [bug](pipelineX) Fix pipelineX bug on multiple BE (apache#28792) * [opt](inverted index) ignore_above only affects untokenized strings (apache#28819) * [improve](load) reduce lock scope in MemTableWriter active consumption (apache#28790) * [opt](Nereids) improve Shape check cases (apache#28124) * tmplate update * log tpcds stats when check shape * [fix](group commit)Fix the issue of duplicate addition of wal path when encouter exception (apache#28691) * [fix](compaction) compaction should catch exception when vertical block reader read next block (apache#28625) * [fix](Nereids) join order is not right after sql parsing (apache#28721) for sql ``` t1, t2 join t3 ``` we should generate plan like: ``` t1 join (t2 join t3) ``` but we generate: ``` (t1 join t2) join t3 ``` to follow legancy planner. * [fix](metrics) fix bvar memtable_input_block_allocated_size (apache#28725) * [enhancement](audit-log) add status code and error msg to audit log for proxy stmt (apache#27607) * [improvement](external catalog)Change log level to debug for getRowCountFromExternalSource. (apache#28801) * [fix](partial update) only report error when in strict mode partial update when finding missing rowsets during flushing memtable (apache#28764) related pr: apache#28062, apache#28674, apache#28677 fix apache#28677 * [fix](function) Fix from_second functions overflow and wrong result (apache#28685) * [Enhancement](load) Limit the number of incorrect data drops and add documents (apache#27727) In the load process, if there are problems with the original data, we will store the error data in an error_log file on the disk for subsequent debugging. However, if there are many error data, it will occupy a lot of disk space. Now we want to limit the number of error data that is saved to the disk. Be familiar with the usage of doris' import function and internal implementation process Add a new be configuration item load_error_log_limit_bytes = default value 200MB Use the newly added threshold to limit the amount of data that RuntimeState::append_error_msg_to_file writes to disk Write regression cases for testing and verification Co-authored-by: xy720 <[email protected]> * [test](partial update) add complex type regression cases for partial update (apache#28758) NOTE: There's some issue for MAP type to work with row store, so in this PR we don't have cases for MAP type Will add the support for MAP type in future. * [fix](test)fix test_create_table test case for nereids (apache#28693) * [improve](config) set mutable and masterOnly in FE config stream_load_default_memtable_on_sink_node (apache#28835) * [fix](group-commit) check if wal need recovery is abnormal (apache#28769) * [Fix](Variant) fix variant predicate rewrite OrToIn with wrong plan (apache#28695) using the name without paths info will lead to wrong In plan, e.g. ``` where cast(v:a as text) = 'hello' or cast(v:b as text) = 'world' ``` will be rewrite to: ``` where cast(v as text) in ('hello', 'world') `` This is wrong, because they are different slots * [refactor](buffer) remove download buffer since it is no longer useful (apache#28832) remove download buffer since it is no longer useful * [Feature](Variant) Implement variant new sub column access method (apache#28484) * [Feature](Variant) Implement variant new sub column access method The query SELECT v["a"]["b"] from simple_var WHERE cast(v["a"]["b"] as int) = 1 encompasses three primary testing scenarios: ``` 1. A basic test involving the variant data type. 2. A scenario dealing with GitHub event data in the context of a variant. 3. A case related to the TPC-H benchmark using a variant. ``` * [fix](memory) Add thread asynchronous purge jemalloc dirty pages (apache#28655) jemallctl purge all arena dirty pages may take several seconds, which will block memory GC and cause OOM. So purge asynchronously in a thread. * [fix](regression) fix regression error of test_compress_type (apache#28826) * [improvement](executor) Add tvf and regression test for Workload Scheduler (apache#28733) 1 Add select workload schedule policy tvf 2 Add reg test * [chore](error msg) print type info when colocate with ddl failed due to type mismatch (apache#28773) * [opt](query cancel) optimization for query cancel apache#28778 * [Fix](multi-catalog) skip hms events if hms table is not supported. (apache#28644) Co-authored-by: wangxiangyu <[email protected]> * [Enhancement](job) No need to query some backends which are not alive. (apache#28608) No need to execute some jobs if backend is not alive * [fix](paimon)fix type convert for paimon (apache#28774) fix type convert for paimon * [fix](stream load)add test case and doc for arrow type of stream load (apache#28098) add test case and doc for arrow type of stream load * [feature](mtmv)mtmv partition refresh case (apache#28787) * (enhance)(InternalQuery) Support to collect profile for intenal query (apache#28762) * [optimize](count) optimize pk exact query without reading data (apache#28494) * [opt](sessionVar)show changed sessoin var first apache#28840 “show variables” command list changed vars before not changed vars, * check stats and log memo for ds46 (apache#28396) * [bug](sharedscan) Fix shared scan bug (apache#28841) * [fix](hash join) fix stack overflow caused by evaluate case expr on huge build block (apache#28851) * [refactor](executor)remove scan group apache#28847 * [pipelineX](refactor) rename functions (apache#28846) * [Feature](inverted index) add lowercase option for inverted index analyzer (apache#28704) * [feature](Nereids) Pull up join from union all (apache#28682) * [fix](ci) tpch pipeline should not re-load data (apache#28874) * [fix](ci) tpch pipeline should not re-load data * 2 --------- Co-authored-by: stephen <[email protected]> * [exec](compress) use FragmentTransmissionCompressionCodec control the exchange compress behavior (apache#28818) * [improve](move-memtable) tweak load stream flush token num and max tasks (apache#28884) * [improve](load) remove extra layer of heavy work pool in tablet_writer_add_block (apache#28550) * [improve](load) limit delta writer flush task parallelism (apache#28883) * [fix](multi-catalog)filter impala generated path (apache#28786) file impala generated dir _imapala_insert_staging * [enhancement](udf) add prepare function for java-udf (apache#28750) * [refactor](test)Refactor workload group/schedule policy test apache#28888 [refactor](test)Refactor workload group/schedule policy test apache#28888 * [improve](move-memtable) avoid using heavy work pool during append data (apache#28745) * [enhancement](broker-load) fix-move-memtable-session-var-for-s3 (apache#28894) * [exec](load) change default parallel num from 1 to 8 in no pipeline exec engine (apache#28864) * [fix](segcompaction) disable segcompaction by default (apache#28906) Signed-off-by: freemandealer <[email protected]> * [fix](pipelineX) fix cannot runtime obtain profile on pipelineX apache#28795 * [fix](mtmv) fix failed to specify the number of buckets when bucket auto (apache#28854) Issue Number: close #xxx - fix failed to specify the number of buckets when bucket auto - delete unused SessionVariable - if mtmv used external table ,check `isMaterializedViewRewriteEnableContainForeignTable` * [bugfix](scannercore) scanner will core in deconstructor during collect profile (apache#28727) * [FIX](map)fix map with rowstore table (apache#28877) * [feature](mtmv)MTMV pause and resume (apache#28887) - PAUSE MATERIALIZED VIEW JOB ON mv1 - RESUME MATERIALIZED VIEW JOB ON mv1 - fix when drop db,not drop job - add lock for one materialized view can only run one task at a time * [feature](mtmv)add more test case1 (apache#28910) * [fix](block) add block columns size dcheck (apache#28539) * [fix](chore) update dcheck to avoid core during stress test (apache#28895) * [fix](doc) Add the usage example of bos to the documentation of s3 tvf (apache#28899) * [fix](mtmv)fix can not create mtmv all use default value (apache#28922) * [fix](parquet) the end offset of column chunk may be wrong in parquet metadata (apache#28891) * [fix](paimon)fix `like` predicate (apache#28803) fix like predict * [fix](merge-on-write) migration may cause duplicate keys for mow table (apache#28923) * [fix](mtmv) Related partition exclude null generate column when increment build materialized view (apache#28855) Infer partition column by materialized view partition column, exclude null generate column in join when increment build materialized view * [nereids] fix join fd computing bug (apache#28849) * [Fix](statistics) Fix partition name NPE and sample for all table during auto analyze (apache#28916) Fix partition name NPE and sample for all table during auto analyze. Sample for all tables because getData may have latency, which may cause full analyze a huge table and use too much resource. Sample for all tables to avoid this. Will improve the strategy later. * [fix](nereids) Fix query mv rewrite fail when mv cache build quickly (apache#28876) * [optimize](zonemap) skip zonemap if predicate does not support_zonemap (apache#28595) * [optimize](zonemap) skip zonemap if predicate does not support_zonemap apache#27608 (apache#28506) * Revert "[bugfix](scannercore) scanner will core in deconstructor during collect profile (apache#28727)" (apache#28931) This reverts commit 4066de3. * [fix] (nereids) Catch exception when mv fail and fix the npe (apache#28932) * [fix](regression-test) test_partial_update_native_insert_stmt_complex is flaky (apache#28927) * [improvement](group commit) make get column function more reliable when replaying wal (apache#28900) * [opt](nereids) convert or to inpredicate and optimize inpredicate partition prune (apache#28316) * [enhancement](bulk-load) cancel loading tasks directly without retrying when timeout exceeded (apache#28666) * [chore](test) Add testing util sync point (apache#28924) * [chore](prompt) Optimize tablet and replica prompt by pointing out what the numbers mean (apache#28925) * [fix](join) incorrect result of left semi/anti join with empty build side (apache#28898) * [docs] (DebugPoints) Update docs about Debug Points (apache#28347) --------- Co-authored-by: qinhao <[email protected]> * [feature](api) add profile text api (apache#28697) * [opt](invert index) Empty strings are not written to the index in the case of TOKENIZED (apache#28822) * [feature](pipelineX) control exchange sink by memory usage (apache#28814) * [fix](pipelineX) fix use global rf when there no shared_scans (apache#28869) * [refactor](pipelineX)do not override dependency() function in pipelineX (apache#28848) * [fix](log) regularise some BE error type and fix a load task check apache#28729 * [chore](test) correct create table statement (apache#28863) * [fix](nereids)subquery unnest need handle subquery in Not expr correnctly (apache#28713) * [feature](nereids)support decimalv2 (apache#28726) * [fix](nereids)group by expr may be bound twice in bind agg slot (apache#28771) * [doc](insert) Add group commit docs (apache#25949) * [performance](variant) support topn 2phase read for variant column (apache#28318) [performance](variant) support topn 2phase read for variant column * return residual expr of join (apache#28760) * [fix](load) fix nullptr when getting memtable flush running count (apache#28942) * [fix](load) fix nullptr when getting memtable flush running count * style * (enhance)(regression) Support `force_olap_table_replication_num=3` run test_insert_random_distribution_table case (apache#28903) * [Test](Job)Add test case (apache#28481) * [fix](doc) typo fix in auto-partition page (apache#28512) * [feature](load) enable memtable on sink node by default (apache#28963) * [Chore](Job)print log before task execute (apache#28962) * [fix](hdfs) Fix HdfsFileSystem::exists_impl crash (apache#28952) Calling hdfsGetLastExceptionRootCause without initializing ThreadLocalState will crash. This PR modifies the condition for determining the existence of a hdfs file, because hdfsExists will set errno to ENOENT when the file does not exist, we can use this condition to check whether a file existence rather than check the existence of the root cause. * [fix](block) fix be core while mutable block merge may cause different row size between columns in origin block (apache#27943) * [fix](doc) typo fix in dynamic-partition page (apache#28511) * [fix](hash join) fix column ref DCHECK failure of hash join node block mem reuse (apache#28991) Introduced by apache#28851, after evaluating build side expr, some columns in resulting block may be referenced more than once in the same block. e.g. coalesce(col_a, 'string') if col_a is nullable but actually contains no null values, in this case funcition coalesce will insert a new nullable column which references the original col_a. * [opt](assert_num_rows) support filter in AssertNumRows operator and fix some explain (apache#28935) * NEED * Update pipeline x * fix pipelinex compile * [fix](parquet_reader) misused bool pointer (apache#28986) Signed-off-by: pengyu <[email protected]> * [doc](multi-catalog)add krb and some ertificates FAQs (apache#28858) add some security docs * [fix](planner)should save original select list item before analyze (apache#28187) * [fix](planner)should save original select list item before analyze * fix test case * fix failed case * [bug](storage) Fix gc rowset bug (apache#28979) * [fix](mtmv)add log for resolve pending task (apache#28999) * add lock for resolve pending task * add lock for resolve pending task * [improvement](nereids) Get partition related table disable nullable field and complete agg matched pattern mv rules. (apache#28973) * [improvement] (nereids) Get partition related table disable nullable field and modify regression test, complete agg mv rules. * make filed not null to create partition mv * [chore](config) modify `tablet_schema_cache_recycle_interval` from 24h to 1h (apache#28980) To prevent from too many tablet schema cache in memory and lead to performance issue when hold lock to erase item * [opt](scanner) optimize the number of threads of scanners (apache#28640) 1. Remove `doris_max_remote_scanner_thread_pool_thread_num`, use `doris_scanner_thread_pool_thread_num` only. 2. Set the default value `doris_scanner_thread_pool_thread_num` as `std::max(48, CpuInfo::num_cores() * 4)` * [chore](config) modify `variant_ratio_of_defaults_as_sparse_column` from 0.95 to 1 (apache#28984) since sparse column is not stable at present * [feature](nereids)support partition property in nereids (apache#28982) * [feature](Nereids) support table valued function http_stream (apache#29004) * [feature](Load)(step2)support nereids load job schedule (apache#26356) We will Integrate new load job manager into new job scheduling framework so that the insert into task can be scheduled after the broker load sql is converted to insert into TVF(table value function) sql. issue: apache#24221 Now support: 1. load data by tvf insert into sql, but just for simple load（columns need to be defined in the table） 2. show load stmt - job id, label name, job state, time info - simple progress 3. cancel load from db 4. support that enable new load through Config.enable_nereids_load 5. can replay job after restarting doris TODO: - support partition insert job - support show statistics from BE - support multiple task and collect task statistic - support transactional task - need add ut case * [enhance](tablet) Reduce log in tablet meta (apache#28719) * [fix](case) Add sync to test case (apache#29034) * [opt](compound) Optimize by deleting the compound expr after obtaining the final result (apache#28934) * [hotfix](jdbc catalog) fix load table and column names npe (apache#28865) This fix is for the npe situation that occurs when FE of non-Master nodes initializes Jdbc Catalog metadata. * [opt](Nereids) support cast bewteen numeric and boolean in FE (apache#29006) * [improvement](jdbc catalog) Optimize connection pool caching logic (apache#28859) In the old caching logic, we only used jdbcurl, user, and password as cache keys. This may cause the old link to be still used when replacing the jar package, so we should concatenate all the parameters required for the connection pool as the key. * [fix](Nereids) runtime filter push down failed (apache#28997) project child not always NamedExpression failed msg ``` org.apache.doris.common.AnalysisException: errCode = 2, detailMessage = class org.apache.doris.nereids.trees.expressions.literal.VarcharLiteral cannot be cast to class org.apache.doris.nereids.trees.expressions.NamedExpression (org.apache.doris.nereids.trees.expressions.literal.VarcharLiteral and org.apache.doris.nereids.trees.expressions.NamedExpression are in unnamed module of loader 'app') at org.apache.doris.qe.StmtExecutor.executeByNereids(StmtExecutor.java:623) ~[classes/:?] at org.apache.doris.qe.StmtExecutor.execute(StmtExecutor.java:478) ~[classes/:?] at org.apache.doris.qe.StmtExecutor.execute(StmtExecutor.java:457) ~[classes/:?] at org.apache.doris.qe.ConnectProcessor.handleQuery(ConnectProcessor.java:245) ~[classes/:?] at org.apache.doris.qe.MysqlConnectProcessor.handleQuery(MysqlConnectProcessor.java:166) ~[classes/:?] at org.apache.doris.qe.MysqlConnectProcessor.dispatch(MysqlConnectProcessor.java:193) ~[classes/:?] at org.apache.doris.qe.MysqlConnectProcessor.processOnce(MysqlConnectProcessor.java:246) ~[classes/:?] at org.apache.doris.mysql.ReadListener.lambda$handleEvent$0(ReadListener.java:52) ~[classes/:?] at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1128) ~[?:?] at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:628) ~[?:?] at java.lang.Thread.run(Thread.java:829) ~[?:?] Caused by: java.lang.ClassCastException: class org.apache.doris.nereids.trees.expressions.literal.VarcharLiteral cannot be cast to class org.apache.doris.nereids.trees.expressions.NamedExpression (org.apache.doris.nereids.trees.expressions.literal.VarcharLiteral and org.apache.doris.nereids.trees.expressions.NamedExpression are in unnamed module of loader 'app') at org.apache.doris.nereids.trees.plans.physical.PhysicalSetOperation.pushDownRuntimeFilter(PhysicalSetOperation.java:178) ~[classes/:?] at org.apache.doris.nereids.trees.plans.physical.PhysicalHashJoin.pushDownRuntimeFilter(PhysicalHashJoin.java:229) ~[classes/:?] at org.apache.doris.nereids.processor.post.RuntimeFilterGenerator.pushDownRuntimeFilterCommon(RuntimeFilterGenerator.java:386) ~[classes/:?] ``` * [fix](Nereids) generating function should not folding to NullLiteral (apache#29003) should not fold table generating function to null when do constant folding. we should remove Generate node and replaced it by project later. * [improve](load) add profile for WaitFlushLimitTime (apache#29013) * [opt](Nereids) let inverted index work with top opt (apache#29000) * [feature](Nereids) support values inline table in query (apache#28972) * [enhencement](config) change default memtable size & loadStreamPerNode & default load parallelism (apache#28977) We change memtable size from 200MB to 100MB to achieve smoother flush performance. We change loadStreamPerNode from 20 to 60 to avoid stream rpc to be the bottleneck when enable memtable_on_sink_node. We change default s3&broker load parallelsim to make the most of CPUs on moderm multi-core systems. Signed-off-by: freemandealer <[email protected]> * [improve](move-memtable) increase load_stream_flush_token_max_tasks (apache#29011) * [feature](scan) Implement parallel scanning by dividing the tablets based on the row range (apache#28967) * [feature](scan) parallel scann on dup/mow mode * fix bugs * [refactor](create tablet) default create tablet round robin (apache#28911) * [Bug] optimize the collection import Lists dependency packge apache#28523 (apache#28579) * [fix](regression) Fix flaky test test_partial_update_2pc_schema_change (apache#29037) * [Bug](security) BE download_files function exists log print sensitive msg apache#28592 (apache#28594) * Revert "[feature](load) enable memtable on sink node by default (apache#28963)" (apache#29090) This reverts commit 17917a0. * [improvement](statistics)Remove retry load when load stats cache fail (apache#28904) Remove retry load when load stats cache fail. This case usually happens when BE is down or BE OOM, retry doesn't work in these cases and may increase BE work load. * [fix](pipeline) sort_merge should throw exception in has_next_block if got failed status (apache#29076) Test in regression-test/suites/datatype_p0/decimalv3/test_decimalv3_overflow.groovy::249 sometimes failed when there are multiple BEs and FE process report status slowly for some reason. explain select k1, k2, k1 * k2 from test_decimal128_overflow2 order by 1,2,3 -------------- +----------------------------------------------------------------------------------------------------------------------------+ | Explain String(Nereids Planner) | +----------------------------------------------------------------------------------------------------------------------------+ | PLAN FRAGMENT 0 | | OUTPUT EXPRS: | | k1[#5] | | k2[#6] | | (k1 * k2)[#7] | | PARTITION: UNPARTITIONED | | | | HAS_COLO_PLAN_NODE: false | | | | VRESULT SINK | | MYSQL_PROTOCAL | | | | 111:VMERGING-EXCHANGE | | offset: 0 | | | | PLAN FRAGMENT 1 | | | | PARTITION: HASH_PARTITIONED: k1[#0], k2[#1] | | | | HAS_COLO_PLAN_NODE: false | | | | STREAM DATA SINK | | EXCHANGE ID: 111 | | UNPARTITIONED | | | | 108:VSORT | | | order by: k1[#5] ASC, k2[#6] ASC, (k1 * k2)[#7] ASC | | | offset: 0 | | | | | 102:VOlapScanNode | | TABLE: regression_test_datatype_p0_decimalv3.test_decimal128_overflow2(test_decimal128_overflow2), PREAGGREGATION: ON | | partitions=1/1 (test_decimal128_overflow2), tablets=8/8, tabletList=22841,22843,22845 ... | | cardinality=6, avgRowSize=0.0, numNodes=1 | | pushAggOp=NONE | | projections: k1[#0], k2[#1], (k1[#0] * k2[#1]) | | project output tuple id: 1 | +----------------------------------------------------------------------------------------------------------------------------+ 36 rows in set (0.03 sec) Why failed: Multiple BEs Fragments 0 and 1 are MUST on different BEs Pipeline task of VOlapScanNode which executes k1*k2 failed sets query status to cancelled Pipeline task of VSort call try close, send Cancelled status to VMergeExchange sort_curso did not throw exception when it meets error * [feature](Nereids): support infer join when comapring mv (apache#28988) * [fix](nereids) "not is null" stats estimation fix (apache#28860) * fix not is null stats * [opt](nereids)expr normalize after filter pushdown (apache#28743) normalize expression after filter push down * [Enhancement](Wal)Support dynamic wal space limit (apache#27726) * [fix](doc) spell errors fixes query profile docs (apache#28437) * [opt](Nerieds) add infer props to expression (apache#28953) * [fix](read) remove logic of estimating count of rows to read in segment iterator to avoid wrong result of unique key. (apache#29109) * [fix](paimon) read batch by doris' batch size (apache#29039) * [fix](mtmv)add log for resolve pending task (apache#29078) * [Refact](inverted index) refactor inverted index writer init (apache#29072) * [fix](paimon)Remove the static attribute of the source for paimon (apache#29032) * [fix](planner) Fix delete statement on MOR unique table (apache#28968) add conditions for mor table removed in apache#26776 * [fix][compile]unused variable (apache#28992) * [Fix](WorkFlow)Auto replay not work * [Fix](WorkFlow)Auto replay not work * [Fix](WorkFlow)Auto replay not work --------- Signed-off-by: nextdreamblue <[email protected]> Signed-off-by: freemandealer <[email protected]> Signed-off-by: pengyu <[email protected]> Co-authored-by: lihangyu <[email protected]> Co-authored-by: Kang <[email protected]> Co-authored-by: jakevin <[email protected]> Co-authored-by: morrySnow <[email protected]> Co-authored-by: xueweizhang <[email protected]> Co-authored-by: zhangguoqiang <[email protected]> Co-authored-by: zhangdong <[email protected]> Co-authored-by: airborne12 <[email protected]> Co-authored-by: Ma1oneZhang <[email protected]> Co-authored-by: ziyang zhang <[email protected]> Co-authored-by: Siyang Tang <[email protected]> Co-authored-by: Kaijie Chen <[email protected]> Co-authored-by: Yongqiang YANG <[email protected]> Co-authored-by: yiguolei <[email protected]> Co-authored-by: minghong <[email protected]> Co-authored-by: Jibing-Li <[email protected]> Co-authored-by: Gavin Chou <[email protected]> Co-authored-by: plat1ko <[email protected]> Co-authored-by: wudongliang <[email protected]> Co-authored-by: Nitin-Kashyap <[email protected]> Co-authored-by: yiguolei <[email protected]> Co-authored-by: wangbo <[email protected]> Co-authored-by: nanfeng <[email protected]> Co-authored-by: Xinyi Zou <[email protected]> Co-authored-by: qiye <[email protected]> Co-authored-by: DuRipeng <[email protected]> Co-authored-by: wuwenchi <[email protected]> Co-authored-by: Qi Chen <[email protected]> Co-authored-by: Mingyu Chen <[email protected]> Co-authored-by: Xin Liao <[email protected]> Co-authored-by: yujun <[email protected]> Co-authored-by: walter <[email protected]> Co-authored-by: xy <[email protected]> Co-authored-by: xingying01 <[email protected]> Co-authored-by: seawinde <[email protected]> Co-authored-by: Mryange <[email protected]> Co-authored-by: starocean999 <[email protected]> Co-authored-by: bobhan1 <[email protected]> Co-authored-by: zzzxl <[email protected]> Co-authored-by: yangshijie <[email protected]> Co-authored-by: zclllyybb <[email protected]> Co-authored-by: amory <[email protected]> Co-authored-by: zhiqiang <[email protected]> Co-authored-by: xy720 <[email protected]> Co-authored-by: Gabriel <[email protected]> Co-authored-by: htyoung <[email protected]> Co-authored-by: Pxl <[email protected]> Co-authored-by: Luwei <[email protected]> Co-authored-by: meiyi <[email protected]> Co-authored-by: Ashin Gau <[email protected]> Co-authored-by: HHoflittlefish777 <[email protected]> Co-authored-by: 谢健 <[email protected]> Co-authored-by: XuJianxu <[email protected]> Co-authored-by: 胥剑旭 <[email protected]> Co-authored-by: Yoko <[email protected]> Co-authored-by: huanghaibin <[email protected]> Co-authored-by: lw112 <[email protected]> Co-authored-by: zhannngchen <[email protected]> Co-authored-by: Xiangyu Wang <[email protected]> Co-authored-by: wangxiangyu <[email protected]> Co-authored-by: AlexYue <[email protected]> Co-authored-by: TengJianPing <[email protected]> Co-authored-by: xzj7019 <[email protected]> Co-authored-by: Dongyang Li <[email protected]> Co-authored-by: stephen <[email protected]> Co-authored-by: HappenLee <[email protected]> Co-authored-by: slothever <[email protected]> Co-authored-by: zhangstar333 <[email protected]> Co-authored-by: zhengyu <[email protected]> Co-authored-by: Jerry Hu <[email protected]> Co-authored-by: HowardQin <[email protected]> Co-authored-by: qinhao <[email protected]> Co-authored-by: deardeng <[email protected]> Co-authored-by: caiconghui <[email protected]> Co-authored-by: py023 <[email protected]> Co-authored-by: zy-kkk <[email protected]> Co-authored-by: Guangming Lu <[email protected]> Co-authored-by: abmdocrt <[email protected]>
hello-stephen
pushed a commit
to hello-stephen/doris
that referenced
this pull request
Dec 28, 2023
HappenLee
pushed a commit
to HappenLee/incubator-doris
that referenced
this pull request
Jan 12, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Proposed changes
Heavy work pool is causing too many context switches, remove it from LoadStream.
bthread::Mutexin LoadStream is only used in bthread context.And blocking IO happens in
SegmentFileWriterThreadPool.Further comments
If this is a relatively large or complex change, kick off the discussion at [email protected] by explaining why you chose the solution you did and what alternatives you considered, etc...